{ "schema_version": "1.0", "updated_at": "2026-10-30T07:17:40.498Z", "source": { "name": "Models.dev", "url": "https://models.dev/models.json", "retrieved_at": "2026-10-26T09:05:47.987Z", "license": "MIT" }, "coverage": { "source_records": 606, "tracked_models": 1198, "upstream_models": 496, "historical_additions": 702, "retained_models": 0, "excluded_non_language_models": 66, "excluded_future_releases": 0, "merged_aliases": 14, "excluded_mutable_aliases": 10, "excluded_service_tiers": 11, "excluded_third_party_derivatives": 9, "date_basis": "First external availability, including limited access, as recorded by sources. Announcement dates are separate. Month precision is preserved.", "limitation": "A growing source-backed catalog, not an exhaustive census. Coverage depends on upstream and contributed records. Model variants count separately; explicitly identified aliases are merged." }, "research": { "reviewed_at": "2026-10-30T07:17:40.498Z" }, "models": [ { "id": "mistral/mistral-large-4-2026-10-29", "slug": "mistral--mistral-large-4-2026-10-29", "name": "Mistral Large 4 (October 29)", "developer": "Mistral", "developer_id": "mistral", "family": "mistral-large", "description": "Multimodal Mistral Large 4 revision released after further reinforcement learning.", "release_date": "2026-10-29", "date_precision": "day", "open_weights": true, "license": "Mistral Ecosystem Open-Weight License", "parameters_b": 1050, "active_parameters_b": 52, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://mistral.ai/news/mistral-large-4-oss/", "source_type": "primary", "sources": [ { "label": "October29 revised checkpoint release", "url": "https://mistral.ai/news/mistral-large-4-oss/", "type": "primary" }, { "label": "Official versioned model card", "url": "https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B/blob/f12dabdc0c1e99e7f7924461a6f308859b05af4d/README.md", "type": "primary" }, { "label": "Official versioned configuration", "url": "https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B/blob/f12dabdc0c1e99e7f7924461a6f308859b05af4d/config.json", "type": "primary" }, { "label": "Checkpoint license", "url": "https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B/blob/f12dabdc0c1e99e7f7924461a6f308859b05af4d/LICENSE", "type": "primary" }, { "label": "Publisher model reference", "url": "https://docs.mistral.ai/models/mistral-large-4-0.md", "type": "primary" }, { "label": "Official dated announcement index", "url": "https://mistral.ai/news/", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:13:05.399Z", "last_seen": "2026-10-30T07:13:05.399Z", "upstream_present": false, "availability": "public", "notes": "The October29 release is a further-trained revision: Mistral reports three additional weeks of reinforcement learning after the October6 public preview. FP8 and NVFP4 distributions are the same model. The 1.05T total is nominal from publisher docs; the card describes approximately1T. Active52B includes embeddings/output layers, versus49B under the other published convention. Total inclusion of the separately disclosed vision encoder and embedded DSpark draft head is not established by an independent census. Exact1048576 context comes from the released configuration." }, { "id": "diffuspace/clover-35b-a3b-preview", "slug": "diffuspace--clover-35b-a3b-preview", "name": "Clover-35B-A3B-Preview", "developer": "DiffuSpace", "developer_id": "diffuspace", "family": "clover", "description": "Text diffusion model derived from Qwen3.6-35B-A3B.", "release_date": "2026-10-28", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 35, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/Diffuspace/Clover-35B-A3B-Preview", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/Diffuspace/Clover-35B-A3B-Preview/blob/4b0402b65e523d0183a14c431955438714cb16ff/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/Diffuspace/Clover-35B-A3B-Preview/blob/4b0402b65e523d0183a14c431955438714cb16ff/config.json", "type": "config" }, { "label": "Public checkpoint files", "url": "https://huggingface.co/Diffuspace/Clover-35B-A3B-Preview/tree/4b0402b65e523d0183a14c431955438714cb16ff", "type": "weights" }, { "label": "Developer launch", "url": "https://diffuspace.ai/blog/clover-35b/", "type": "announcement" }, { "label": "Dated model release", "url": "https://huggingface.co/Diffuspace/Clover-35B-A3B-Preview/commit/4b0402b65e523d0183a14c431955438714cb16ff", "type": "release" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "notes": "The publisher release commit is dated October 28. Nominal 35B total and approximately 3B active; block diffusion on a hybrid Gated DeltaNet/full-attention MoE backbone." }, { "id": "nvidia/nemotron-3.5-transcribe-30b-a4b", "slug": "nvidia--nemotron-3.5-transcribe-30b-a4b", "name": "Nemotron 3.5 Transcribe 30B A4B", "developer": "NVIDIA", "developer_id": "nvidia", "family": "nemotron", "description": "Speech-augmented language model for transcription, translation and audio question answering.", "release_date": "2026-10-28", "date_precision": "day", "open_weights": true, "license": "OpenMDW-1.1", "parameters_b": 34, "active_parameters_b": 4.3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 65536, "modalities": { "input": [ "text", "audio" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B/blob/9d8cd6b79bfdd427b6558a5fdbf76fcaf8c75341/README.md", "source_type": "primary", "sources": [ { "label": "Official model card and downloadable weights", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B/blob/9d8cd6b79bfdd427b6558a5fdbf76fcaf8c75341/README.md", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "notes": "Included because this speech-augmented language model supports open-ended audio question answering and text instruction following, beyond transcription. Official totals include the audio encoder: 34B total and 4.3B active, of which about 3.5B active is the language backbone and 0.8B is the audio encoder. Use the documented 65,536-token supported context rather than a higher underlying config limit." }, { "id": "particle/particle-1", "slug": "particle--particle-1", "name": "Particle-1", "developer": "Particle AI", "developer_id": "particle", "family": "particle", "description": "Software-engineering model post-trained from DeepSeek V4.1 Flash to reduce token use.", "release_date": "2026-10-28", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://particle.ai/blog/introducing-particle-1", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://particle.ai/blog/introducing-particle-1", "type": "announcement" }, { "label": "Public checkpoint search", "url": "https://huggingface.co/api/models?search=particle-1&limit=100", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "notes": "The dated developer announcement confirms public API availability. Training uses expert-layer LoRA adapters; the deployed parameter total is not established and the base model’s figures are not transferred. No downloadable exact checkpoint was found in the launch links or public Hugging Face search. API documentation could not be retrieved, so context remains unspecified." }, { "id": "alibaba/qwen-audio-3.1-realtime-flash", "slug": "alibaba--qwen-audio-3.1-realtime-flash", "name": "Qwen Audio 3.1 Realtime Flash", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen-audio", "description": "Realtime conversational model with text and audio responses, tool calling, web search and voice cloning.", "release_date": "2026-10-27", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 262144, "modalities": { "input": [ "text", "audio" ], "output": [ "text", "audio" ] }, "reasoning": false, "tool_call": true, "source_url": "https://help.aliyun.com/zh/model-studio/newly-released-models", "source_type": "primary", "sources": [ { "label": "Dated first-party model availability table", "url": "https://help.aliyun.com/zh/model-studio/newly-released-models", "type": "primary" }, { "label": "Realtime input/output, reasoning and tools", "url": "https://docs.qwencloud.com/developer-guides/speech/omni-models", "type": "primary" }, { "label": "Conversation and text-only response guide", "url": "https://docs.qwencloud.com/developer-guides/speech/qwen-audio-realtime", "type": "primary" }, { "label": "Qwen public weights inventory", "url": "https://huggingface.co/api/models?author=Qwen&sort=createdAt&direction=-1&limit=50", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "notes": "Alibaba’s own model availability table dates the Flash release October 27 and specifies 262144 tokens. Native speech docs additionally limit retained audio to 50 turns or 300 seconds. Unlike a standalone TTS or transcription model, it generates conversational text, including a text-only response mode. Parameter counts and architecture are undisclosed; the current Qwen model inventory has no exact public checkpoint." }, { "id": "baidu/unlimited-ocr-2", "slug": "baidu--unlimited-ocr-2", "name": "Unlimited OCR 2", "developer": "Baidu", "developer_id": "baidu", "family": "unlimited-ocr", "description": "Vision-language model for document parsing with sparse reference sliding-window attention.", "release_date": "2026-10-27", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 3, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/baidu/Unlimited-OCR", "source_type": "primary", "sources": [ { "label": "Dated model release", "url": "https://github.com/baidu/Unlimited-OCR", "type": "primary" }, { "label": "Official model card and MIT declaration", "url": "https://huggingface.co/baidu/Unlimited-OCR-2/blob/5632329a32e103d5f1add270681e0f5c1233c6e4/README.md", "type": "primary" }, { "label": "Released model configuration", "url": "https://huggingface.co/baidu/Unlimited-OCR-2/blob/5632329a32e103d5f1add270681e0f5c1233c6e4/config.json", "type": "primary" }, { "label": "Original technical report", "url": "https://huggingface.co/baidu/Unlimited-OCR-2/resolve/5632329a32e103d5f1add270681e0f5c1233c6e4/Unlimited-OCR-2.pdf", "type": "primary" }, { "label": "Public checkpoint inventory", "url": "https://huggingface.co/api/models/baidu/Unlimited-OCR-2/revision/5632329a32e103d5f1add270681e0f5c1233c6e4", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "notes": "Baidu’s dated release notes establish October 27, independently of repository creation. The nominal 3B figure is scoped to the language decoder in technical-report Figure 3 and is not the complete multimodal model size. The report prose gives 500M activated parameters, while Figure 3 labels the decoder A570M; the active count remains unset because the discrepancy is unexplained. Released config sets 131072 positions; unlimited-horizon parsing uses bounded sliding/reference windows, not an infinite token context.", "parameters_scope": "language_backbone", "parameter_components_b": { "language_backbone": 3 } }, { "id": "ibm/granite-switch-4.1.1-30b", "slug": "ibm--granite-switch-4.1.1-30b", "name": "Granite Switch 4.1.1 30B", "developer": "IBM", "developer_id": "ibm", "family": "granite-switch", "description": "Language model composed with decision and safety adapters and a speech-to-text encoder.", "release_date": "2026-10-27", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 30, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "audio" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/ibm-granite/granite-switch-4.1-30b/blob/eb206725d03c0f1c409ce3aa512afe325fdf8731/README.md", "source_type": "primary", "sources": [ { "label": "Official version-specific model card and weights", "url": "https://huggingface.co/ibm-granite/granite-switch-4.1-30b/blob/eb206725d03c0f1c409ce3aa512afe325fdf8731/README.md", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "parameters_scope": "backbone", "notes": "The parameter value covers the 30B language backbone, not the complete composed system with adapters and a 470M speech encoder. Audio is transcribed and inserted into the chat context; the speech encoder and LLM are not jointly trained. Revision 4.1.1 changes decision-adapter weights and adds multilingual speech; the original 4.1 checkpoint remains separately tagged.", "parameter_components_b": { "backbone": 30 } }, { "id": "ibm/granite-switch-4.1.1-8b", "slug": "ibm--granite-switch-4.1.1-8b", "name": "Granite Switch 4.1.1 8B", "developer": "IBM", "developer_id": "ibm", "family": "granite-switch", "description": "Language model composed with decision and safety adapters and a speech-to-text encoder.", "release_date": "2026-10-27", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "audio" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/ibm-granite/granite-switch-4.1-8b/blob/fb4aa7877e13e5fe09ca328713dc1db72a3f93ce/README.md", "source_type": "primary", "sources": [ { "label": "Official version-specific model card and weights", "url": "https://huggingface.co/ibm-granite/granite-switch-4.1-8b/blob/fb4aa7877e13e5fe09ca328713dc1db72a3f93ce/README.md", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "parameters_scope": "backbone", "notes": "The parameter value covers the 8B language backbone, not the complete composed system with adapters and a 470M speech encoder. Audio is transcribed and inserted into the chat context; the speech encoder and LLM are not jointly trained. Revision 4.1.1 changes decision-adapter weights and adds multilingual speech; the original 4.1 checkpoint remains separately tagged.", "parameter_components_b": { "backbone": 8 } }, { "id": "inclusionai/ling-3.1-flash-1019", "slug": "inclusionai--ling-3.1-flash-1019", "name": "Ling 3.1 Flash 1019", "developer": "InclusionAI", "developer_id": "inclusionai", "family": "ling 3.1 flash 1019", "description": "Updated reasoning and tool-use checkpoint of Ling 3.1 Flash.", "release_date": "2026-10-27", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 560, "active_parameters_b": 25, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/inclusionAI/Ling-3.1-flash-1019", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/inclusionAI/Ling-3.1-flash-1019/blob/dddfee4bb43519a92761fbc5cefe3f0d3ff4aded/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/inclusionAI/Ling-3.1-flash-1019/blob/dddfee4bb43519a92761fbc5cefe3f0d3ff4aded/config.json", "type": "config" }, { "label": "Public checkpoint files", "url": "https://huggingface.co/inclusionAI/Ling-3.1-flash-1019/tree/dddfee4bb43519a92761fbc5cefe3f0d3ff4aded", "type": "weights" }, { "label": "Original model history", "url": "https://huggingface.co/inclusionAI/Ling-3.1-flash-1019/commits/main", "type": "release" }, { "label": "Developer release announcement", "url": "https://x.com/AntLingAGI/status/2115062200687878157", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "notes": "The developer’s October 27 announcement explicitly releases both 1019 and Base. The developer reports nominal 560B total and 25B active; serialized checkpoint metadata includes additional tensors. Context uses the model card’s nominal 1M limit; the underlying config lists 262144 positions. The 1019 suffix is not treated as a release date." }, { "id": "inclusionai/ling-3.1-flash-base", "slug": "inclusionai--ling-3.1-flash-base", "name": "Ling 3.1 Flash Base", "developer": "InclusionAI", "developer_id": "inclusionai", "family": "ling 3.1 flash base", "description": "Pretrained hybrid-attention MoE checkpoint for further training.", "release_date": "2026-10-27", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 560, "active_parameters_b": 25, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/inclusionAI/Ling-3.1-flash-base", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/inclusionAI/Ling-3.1-flash-base/blob/c2cad79e46fcb7b46d96f9cf876ce1f2418b6e64/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/inclusionAI/Ling-3.1-flash-base/blob/c2cad79e46fcb7b46d96f9cf876ce1f2418b6e64/config.json", "type": "config" }, { "label": "Public checkpoint files", "url": "https://huggingface.co/inclusionAI/Ling-3.1-flash-base/tree/c2cad79e46fcb7b46d96f9cf876ce1f2418b6e64", "type": "weights" }, { "label": "Original model history", "url": "https://huggingface.co/inclusionAI/Ling-3.1-flash-base/commits/main", "type": "release" }, { "label": "Developer release announcement", "url": "https://x.com/AntLingAGI/status/2115062200687878157", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "notes": "The developer’s October 27 announcement explicitly releases both 1019 and Base. The developer reports nominal 560B total and 25B active; serialized checkpoint metadata includes additional tensors." }, { "id": "liquid/lfm3-1.7b", "slug": "liquid--lfm3-1.7b", "name": "LFM3-1.7B", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm3", "description": "Compact hybrid language model for local agents and tool use.", "release_date": "2026-10-27", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 1.73, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "hybrid", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.liquid.ai/blog/lfm3", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/lfm3", "type": "primary" }, { "label": "Official model card and downloadable weights", "url": "https://huggingface.co/LiquidAI/LFM3-1.7B/blob/ac2586a34f1a1c03329dbdcdefabd8b6478ddaa7/README.md", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "notes": "Tri-hybrid convolution, sliding-window and global-attention backbone. Nominal count from the model card; the BF16 checkpoint metadata reports 1,728,801,536 parameters." }, { "id": "liquid/lfm3-1.7b-base", "slug": "liquid--lfm3-1.7b-base", "name": "LFM3-1.7B-Base", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm3", "description": "Pretrained and mid-trained hybrid language model.", "release_date": "2026-10-27", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 1.73, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "hybrid", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://www.liquid.ai/blog/lfm3", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/lfm3", "type": "primary" }, { "label": "Official model card and downloadable weights", "url": "https://huggingface.co/LiquidAI/LFM3-1.7B-Base/blob/545fc8f4c7b5fa9f405105274cd29c4f8f44b526/README.md", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "notes": "Pretrained and mid-trained base checkpoint, separate from the post-trained LFM3-1.7B. The card documents thinking traces in the chat template. Nominal count from the model card; BF16 checkpoint metadata reports 1,728,801,536 parameters." }, { "id": "liquid/lfm3-audio-2b", "slug": "liquid--lfm3-audio-2b", "name": "LFM3-Audio-2B", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm3", "description": "Audio-language model with text and speech generation.", "release_date": "2026-10-27", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 2, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "hybrid", "context_window": 32000, "modalities": { "input": [ "text", "audio" ], "output": [ "text", "audio" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.liquid.ai/blog/lfm3", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/lfm3", "type": "primary" }, { "label": "Official model card and downloadable weights", "url": "https://huggingface.co/LiquidAI/LFM3-Audio-2B/blob/9d49701c372d6011d900f6abb40c6b296fb706dd/README.md", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "notes": "Official model card specifies 2B total and a supported 32K input-plus-output context. Reasoning is supported for text-only responses. LiquidCodec is distributed alongside the generative checkpoint; no independent full-system tensor census is asserted. Context is stored as nominal 32000 from the card’s 32K notation; the exact limit is not established." }, { "id": "liquid/lfm3-omni-2b", "slug": "liquid--lfm3-omni-2b", "name": "LFM3-Omni-2B", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm3", "description": "Multimodal model with text, image, video and audio input and text or speech output.", "release_date": "2026-10-27", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 2.17, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "hybrid", "context_window": 32000, "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text", "audio" ] }, "reasoning": false, "tool_call": true, "source_url": "https://www.liquid.ai/blog/lfm3", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/lfm3", "type": "primary" }, { "label": "Official model card and downloadable weights", "url": "https://huggingface.co/LiquidAI/LFM3-Omni-2B/blob/5797d07c125503bd4f4c0e7f915ae2d3d703e8de/README.md", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "notes": "Official model card specifies 2.17B total and a supported 32K input-plus-output context, despite a larger inherited text config. The card explicitly states this is a non-reasoning model. LiquidCodec is distributed alongside the generative checkpoint. Context is stored as nominal 32000 from the card’s 32K notation; the exact limit is not established." }, { "id": "liquid/lfm3-vl-2b", "slug": "liquid--lfm3-vl-2b", "name": "LFM3-VL-2B", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm3", "description": "Compact vision-language model supporting images and video.", "release_date": "2026-10-27", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 1.92, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "hybrid", "context_window": 128000, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.liquid.ai/blog/lfm3", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/lfm3", "type": "primary" }, { "label": "Official model card and downloadable weights", "url": "https://huggingface.co/LiquidAI/LFM3-VL-2B/blob/ba44bd51db20a2753f28c273a1571ccaa210d954/README.md", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "notes": "Official card gives 1.92B total: 1.73B language model, 164M vision encoder and 25M projector. Its 128K context corresponds to max_position_embeddings=128000 in the published config." }, { "id": "microsoft/mai-cyber-1.1-flash", "slug": "microsoft--mai-cyber-1.1-flash", "name": "MAI Cyber 1.1 Flash", "developer": "Microsoft", "developer_id": "microsoft", "family": "mai-cyber", "description": "Cybersecurity language model available to verified customers through Microsoft’s MDASH scanning system.", "release_date": "2026-10-27", "date_precision": "day", "open_weights": false, "license": "Microsoft MDASH terms", "parameters_b": 137, "active_parameters_b": 5, "parameters_source": "primary_source", "architecture": "moe", "context_window": 256000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://microsoft.ai/news/mai-cyber-1-1-flash/", "source_type": "primary", "sources": [ { "label": "Microsoft release announcement", "url": "https://microsoft.ai/news/mai-cyber-1-1-flash/", "type": "announcement" }, { "label": "MAI Cyber 1.1 Flash model card", "url": "https://microsoft.ai/pdf/MAI-Cyber-1.1-Flash-ORA-Model-Card.pdf", "type": "model_card" }, { "label": "Microsoft model page", "url": "https://microsoft.ai/models/mai-cyber-1-1-flash/", "type": "docs" }, { "label": "MDASH deployment documentation", "url": "https://learn.microsoft.com/en-us/security-exposure-management/mdash-foundry-integration", "type": "docs" }, { "label": "Microsoft public checkpoint search", "url": "https://huggingface.co/api/models?author=microsoft&search=MAI-Cyber&limit=100", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "notes": "Released October 27, 2026 to verified MDASH customers, with no standalone general-access model endpoint or public weights. The model card reports 137B total and 5B active parameters and a nominal 256k context window; an exact power-of-two context limit is not established. This is a fine-tune of MAI Cyber 1 Flash, not the MAI Code 1.1 Flash model." }, { "id": "sparkllm/spark-x2.6-1.7b", "slug": "sparkllm--spark-x2.6-1.7b", "name": "Spark-X2.6-1.7B", "developer": "SparkLLM", "developer_id": "sparkllm", "family": "spark", "description": "Compact text model with configurable thinking and tool use.", "release_date": "2026-10-27", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 1.7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 1048576, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/XHToken/Spark-X2.6-1.7B", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/XHToken/Spark-X2.6-1.7B/blob/f18c032c7675f807c0fdffe281dbe124edba2605/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/XHToken/Spark-X2.6-1.7B/blob/f18c032c7675f807c0fdffe281dbe124edba2605/config.json", "type": "config" }, { "label": "Public checkpoint files", "url": "https://huggingface.co/XHToken/Spark-X2.6-1.7B/tree/f18c032c7675f807c0fdffe281dbe124edba2605", "type": "weights" }, { "label": "Developer launch post", "url": "https://bsky.app/profile/sparkllm.bsky.social/post/3myuj7ud7m62n", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "notes": "Release date comes from the developer’s dated launch post. Parameter size is the nominal model-card label; the published configuration allows 1048576 positions. Quantized builds are distributions of the same model." }, { "id": "openai/gpt-6.1-terra", "slug": "openai--gpt-6.1-terra", "name": "GPT-6.1 Terra", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-6", "description": "Reasoning model for visual analysis, computer use and writing, available through the OpenAI API, ChatGPT Work and Codex.", "release_date": "2026-10-26", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1050000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://developers.openai.com/api/docs/changelog", "source_type": "primary", "sources": [ { "label": "OpenAI API release notes", "url": "https://developers.openai.com/api/docs/changelog", "type": "announcement" }, { "label": "GPT-6.1 Terra model reference", "url": "https://developers.openai.com/api/docs/models/gpt-6.1-terra", "type": "docs" }, { "label": "Release announcement", "url": "https://openai.com/index/introducing-gpt-6-1-terra/", "type": "announcement" }, { "label": "OpenAI public checkpoint search", "url": "https://huggingface.co/api/models?author=openai&search=gpt-6.1&limit=100", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "notes": "Public API release on October 26, 2026. Context is 1,050,000 tokens, with 922,000 maximum input and 128,000 maximum output. Function calling requires Responses. Total parameters, active parameters and architecture are not disclosed in the reviewed model documentation. No downloadable exact checkpoint was found in OpenAI’s public weight inventory." }, { "id": "tencent/hy-simulmt-30b-a3b", "slug": "tencent--hy-simulmt-30b-a3b", "name": "Hy-SimulMT-30B-A3B", "developer": "Tencent", "developer_id": "tencent", "family": "hy-simulmt", "description": "Simultaneous speech translation model producing translated text.", "release_date": "2026-10-26", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 30.394794496, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "audio" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/tencent/Hy-SimulMT-30B-A3B", "source_type": "primary", "sources": [ { "label": "Pinned release notes and model card", "url": "https://huggingface.co/tencent/Hy-SimulMT-30B-A3B/blob/56c233df75bf09e386aa6282b01730b461afc83f/README.md", "type": "primary" }, { "label": "Pinned released model configuration", "url": "https://huggingface.co/tencent/Hy-SimulMT-30B-A3B/blob/56c233df75bf09e386aa6282b01730b461afc83f/config.json", "type": "primary" }, { "label": "Official unquantized checkpoint parameter metadata", "url": "https://huggingface.co/api/models/tencent/Hy-SimulMT-30B-A3B/revision/56c233df75bf09e386aa6282b01730b461afc83f", "type": "primary" }, { "label": "Developer inference and training repository", "url": "https://github.com/Tencent-Hunyuan/Hy-SimulMT", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:13:45.595Z", "last_seen": "2026-10-30T07:13:45.595Z", "upstream_present": false, "availability": "public", "notes": "The official model-card News section explicitly dates release October 26. Repository creation October 22 and checkpoint upload October 24 are not substituted for the public launch. Total is the published unquantized safetensors element count, including the audio encoder; a nominal 3B active count appears only in the checkpoint name and is left unset without a separately defined count. The two released trained checkpoints have different output objectives; FP8/GGUF copies are excluded. Speech output and voice cloning are supplied by a separately configured TTS service, so this checkpoint’s output is text only. Config allows 262144 positions; reference speech runtimes use shorter audio chunks." }, { "id": "tencent/hy-simulmt-30b-a3b-interleave", "slug": "tencent--hy-simulmt-30b-a3b-interleave", "name": "Hy-SimulMT-30B-A3B-Interleave", "developer": "Tencent", "developer_id": "tencent", "family": "hy-simulmt", "description": "Simultaneous speech translation model producing source text and translated text.", "release_date": "2026-10-26", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 30.394794496, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "audio" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/tencent/Hy-SimulMT-30B-A3B-Interleave", "source_type": "primary", "sources": [ { "label": "Pinned release notes and model card", "url": "https://huggingface.co/tencent/Hy-SimulMT-30B-A3B-Interleave/blob/e988e52e19ec4002e2763ec46fdff997c68fe859/README.md", "type": "primary" }, { "label": "Pinned released model configuration", "url": "https://huggingface.co/tencent/Hy-SimulMT-30B-A3B-Interleave/blob/e988e52e19ec4002e2763ec46fdff997c68fe859/config.json", "type": "primary" }, { "label": "Official unquantized checkpoint parameter metadata", "url": "https://huggingface.co/api/models/tencent/Hy-SimulMT-30B-A3B-Interleave/revision/e988e52e19ec4002e2763ec46fdff997c68fe859", "type": "primary" }, { "label": "Developer inference and training repository", "url": "https://github.com/Tencent-Hunyuan/Hy-SimulMT", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:13:45.595Z", "last_seen": "2026-10-30T07:13:45.595Z", "upstream_present": false, "availability": "public", "notes": "The official model-card News section explicitly dates release October 26. Repository creation October 22 and checkpoint upload October 24 are not substituted for the public launch. Total is the published unquantized safetensors element count, including the audio encoder; a nominal 3B active count appears only in the checkpoint name and is left unset without a separately defined count. The two released trained checkpoints have different output objectives; FP8/GGUF copies are excluded. Speech output and voice cloning are supplied by a separately configured TTS service, so this checkpoint’s output is text only. Config allows 262144 positions; reference speech runtimes use shorter audio chunks." }, { "id": "nanbeige/nanbeige4.3-33b", "slug": "nanbeige--nanbeige4.3-33b", "name": "Nanbeige4.3-33B", "developer": "Nanbeige", "developer_id": "nanbeige", "family": "nanbeige4.3", "description": "Text model for coding, reasoning and tool use.", "release_date": "2026-10-25", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 33, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 524288, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/Nanbeige/Nanbeige4.3-33B", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/Nanbeige/Nanbeige4.3-33B/blob/df919e1e2cf077a22204c78b6c826bf3db18bc08/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/Nanbeige/Nanbeige4.3-33B/blob/df919e1e2cf077a22204c78b6c826bf3db18bc08/config.json", "type": "config" }, { "label": "Public checkpoint files", "url": "https://huggingface.co/Nanbeige/Nanbeige4.3-33B/tree/df919e1e2cf077a22204c78b6c826bf3db18bc08", "type": "weights" }, { "label": "Original model history", "url": "https://huggingface.co/Nanbeige/Nanbeige4.3-33B/commits/main", "type": "release" }, { "label": "Developer release announcement", "url": "https://x.com/nanbeige/status/2114328933181333798", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "notes": "The developer’s October 25 announcement links both the post-trained and Base checkpoints. Parameter count is the nominal 33B model-card label; the Base and post-trained checkpoints have different context limits." }, { "id": "nanbeige/nanbeige4.3-33b-base", "slug": "nanbeige--nanbeige4.3-33b-base", "name": "Nanbeige4.3-33B-Base", "developer": "Nanbeige", "developer_id": "nanbeige", "family": "nanbeige4.3", "description": "Pretrained text model for further training.", "release_date": "2026-10-25", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 33, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/Nanbeige/Nanbeige4.3-33B-Base", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/Nanbeige/Nanbeige4.3-33B-Base/blob/e061adde974ffea8cd3586ae43d08ac640afe56e/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/Nanbeige/Nanbeige4.3-33B-Base/blob/e061adde974ffea8cd3586ae43d08ac640afe56e/config.json", "type": "config" }, { "label": "Public checkpoint files", "url": "https://huggingface.co/Nanbeige/Nanbeige4.3-33B-Base/tree/e061adde974ffea8cd3586ae43d08ac640afe56e", "type": "weights" }, { "label": "Original model history", "url": "https://huggingface.co/Nanbeige/Nanbeige4.3-33B-Base/commits/main", "type": "release" }, { "label": "Developer release announcement", "url": "https://x.com/nanbeige/status/2114328933181333798", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "notes": "The developer’s October 25 announcement links both the post-trained and Base checkpoints. Parameter count is the nominal 33B model-card label; the Base and post-trained checkpoints have different context limits." }, { "id": "moonshotai/kimi-k3.1-pro", "slug": "moonshotai--kimi-k3.1-pro", "name": "Kimi K3.1 Pro", "developer": "Moonshot AI", "developer_id": "moonshotai", "family": "kimi-k3", "description": "Multimodal Kimi model with 1M context and always-on reasoning for coding, 3D creation, financial research, and agentic work", "release_date": "2026-10-23", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 2800, "active_parameters_b": 104, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.kimi.com/en/blog/kimi-k3-1-pro", "source_type": "primary", "sources": [ { "label": "Kimi K3.1 Pro launch", "url": "https://www.kimi.com/en/blog/kimi-k3-1-pro", "type": "primary" }, { "label": "Dated Kimi Code launch", "url": "https://www.kimi.com/code/docs/en/kimi-code/whats-new.html", "type": "primary" }, { "label": "Kimi Code model comparison", "url": "https://www.kimi.com/code/docs/en/kimi-code/models.html", "type": "primary" }, { "label": "Kimi K3.1 Pro API guide", "url": "https://platform.kimi.ai/docs/guide/kimi-k3-1-pro-quickstart", "type": "primary" }, { "label": "Moonshot API service terms", "url": "https://platform.kimi.ai/docs/agreement/modeluse", "type": "primary" }, { "label": "Kimi K3 technical report, architecture and Table 1", "url": "https://arxiv.org/html/2607.24653v1", "type": "primary" }, { "label": "Kimi K3 original model card", "url": "https://huggingface.co/moonshotai/Kimi-K3", "type": "primary" }, { "label": "Secondary corroboration: Artificial Analysis Pro launch", "url": "https://artificialanalysis.ai/articles/kimi-k3-1-pro-launch-benchmarks", "type": "catalog" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/moonshotai/kimi-k3.1-pro.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The 2.8T total and 104B active figures are nominal lineage-based inferences from Moonshot's explicit description of Pro as post-trained K3 and the official K3 model card; Artificial Analysis corroborates these Pro specifications. They are not a separate Pro tensor census. K3's finer report values of 2.78T/104.2B are not asserted as exact Pro counts. Moonshot calls Pro 3T-class, not exactly 3000B. MoE architecture follows the same documented K3 lineage. As of October 27, 2026, hosted access is proprietary and full weights are scheduled by November 4; the future weight license is not established.", "field_status": {}, "audited_at": "2026-10-27T06:53:21Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/365.json" }, { "id": "nvidia/nemotron-3.5-bolt-e4b", "slug": "nvidia--nemotron-3.5-bolt-e4b", "name": "Nemotron 3.5 Bolt E4B", "developer": "NVIDIA", "developer_id": "nvidia", "family": "nemotron", "description": "Gemma-derived model for local conversation and tool use.", "release_date": "2026-10-23", "date_precision": "day", "open_weights": true, "license": "OpenMDW-1.1", "parameters_b": 8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 65536, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://research.nvidia.com/labs/nemotron/NVIDIA-Nemotron-3.5-Bolt-E4B-4GB/", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://research.nvidia.com/labs/nemotron/NVIDIA-Nemotron-3.5-Bolt-E4B-4GB/", "type": "primary" }, { "label": "Official model card and downloadable weights", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-BF16/blob/49843eed8db1f1c616c47af75cfd36e07aec0667/README.md", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "notes": "The official card lists 8B total and 4B effective parameters. Effective size is not asserted as a MoE active count. The card specifies text input/output despite inherited multimodal library examples; the serving recipe disables image, video and audio inputs." }, { "id": "recursive/flint", "slug": "recursive--flint", "name": "Flint", "developer": "Recursive", "developer_id": "recursive", "family": "flint", "description": "Recursive text model for coding agents, with open weights, reasoning and tool calling.", "release_date": "2026-10-21", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 39.8, "active_parameters_b": 2.5, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://recursive.com/updates/introducing-recursive-flint", "source_type": "primary", "sources": [ { "label": "Introducing Recursive Flint", "url": "https://recursive.com/updates/introducing-recursive-flint", "type": "primary" }, { "label": "Recursive update archive", "url": "https://recursive.com/updates", "type": "primary" }, { "label": "Flint pinned BF16 model card", "url": "https://huggingface.co/RecursiveSI/flint/blob/2a247be0c0aec895ccb0ac7ca92229ec6df37288/README.md", "type": "primary" }, { "label": "Flint pinned configuration", "url": "https://huggingface.co/RecursiveSI/flint/blob/2a247be0c0aec895ccb0ac7ca92229ec6df37288/config.json", "type": "primary" }, { "label": "Flint complete index and all 15 bounded shard headers", "url": "https://huggingface.co/RecursiveSI/flint/blob/2a247be0c0aec895ccb0ac7ca92229ec6df37288/model.safetensors.index.json", "type": "primary" }, { "label": "Published Flint SGLang implementation package", "url": "https://huggingface.co/RecursiveSI/flint/blob/2a247be0c0aec895ccb0ac7ca92229ec6df37288/sglang/flint_sglang-0.1.4-py3-none-any.whl", "type": "primary" }, { "label": "Flint native chat template", "url": "https://huggingface.co/RecursiveSI/flint/blob/2a247be0c0aec895ccb0ac7ca92229ec6df37288/chat_template.jinja", "type": "primary" }, { "label": "Flint pinned SGLang serving guide", "url": "https://huggingface.co/RecursiveSI/flint/blob/2a247be0c0aec895ccb0ac7ca92229ec6df37288/sglang/README.md", "type": "primary" }, { "label": "Recursive SGLang documentation", "url": "https://docs.recursive.com/gpu/sglang", "type": "primary" }, { "label": "Flint Apache License2.0", "url": "https://huggingface.co/RecursiveSI/flint/blob/2a247be0c0aec895ccb0ac7ca92229ec6df37288/LICENSE", "type": "primary" }, { "label": "Rec pinned Flint INT4 deployment profile", "url": "https://github.com/recursive-oss/rec/blob/0e3150ca4262f572dee25c231d9ac7148414d2af/packaging/flint-int4-mlx.json", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/recursive/flint.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/RecursiveSI/flint", "type": "weights" }, { "label": "Introducing Recursive Flint", "url": "https://www.recursive.com/updates/introducing-recursive-flint", "type": "announcement" }, { "label": "Hugging Face (INT4)", "url": "https://huggingface.co/RecursiveSI/flint-int4", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "39.8B total and approximately 2.5B active are developer nominal counts. The active convention includes the output head and excludes the full input embedding matrix; the active body alone is approximately 2.2B. All 15 BF16 shard headers reconcile to 39,785,287,456 learned elements, including learned attention sinks; runtime-only zero router biases and separate quantized artifacts are excluded. Context 262144 is the documented BF16 configuration. The launch claims up to 512K and recommends 256K; the inspected checkpoint and serving guide do not provide a matching 512K configuration, and no native training-length claim is inferred. Reasoning is model-selected each turn; no universal thinking toggle is documented.", "field_status": {}, "audited_at": "2026-10-27T03:54:43.486298Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/476.json" }, { "id": "sparkllm/spark-x2.6-4b", "slug": "sparkllm--spark-x2.6-4b", "name": "Spark-X2.6-4B", "developer": "SparkLLM", "developer_id": "sparkllm", "family": "spark", "description": "Compact text model with configurable thinking and tool use.", "release_date": "2026-10-20", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 4, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 1048576, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/XHToken/Spark-X2.6-4B", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/XHToken/Spark-X2.6-4B/blob/d1b5f26071101c27fcd07728e7a7770e428281f4/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/XHToken/Spark-X2.6-4B/blob/d1b5f26071101c27fcd07728e7a7770e428281f4/config.json", "type": "config" }, { "label": "Public checkpoint files", "url": "https://huggingface.co/XHToken/Spark-X2.6-4B/tree/d1b5f26071101c27fcd07728e7a7770e428281f4", "type": "weights" }, { "label": "Developer launch post", "url": "https://bsky.app/profile/sparkllm.bsky.social/post/3mydbovzh752y", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "notes": "Release date comes from the developer’s dated launch post. Parameter size is the nominal model-card label; the published configuration allows 1048576 positions. Quantized builds are distributions of the same model." }, { "id": "inception/mercury-3", "slug": "inception--mercury-3", "name": "Mercury 3", "developer": "Inception", "developer_id": "inception", "family": "mercury", "description": "Inception's diffusion reasoning model for chat, multi-step problem solving, and tool use", "release_date": "2026-10-19", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "diffusion", "context_window": 1000000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.inceptionlabs.ai/blog/introducing-mercury-3", "source_type": "primary", "sources": [ { "label": "Inception Mercury 3 launch", "url": "https://www.inceptionlabs.ai/blog/introducing-mercury-3", "type": "primary" }, { "label": "Inception model documentation", "url": "https://docs.inceptionlabs.ai/get-started/models", "type": "primary" }, { "label": "Inception streaming and diffusion", "url": "https://docs.inceptionlabs.ai/capabilities/streaming", "type": "primary" }, { "label": "Inception terms of use", "url": "https://www.inceptionlabs.ai/docs/terms-of-use", "type": "primary" }, { "label": "Inception research archive", "url": "https://www.inceptionlabs.ai/research", "type": "primary" }, { "label": "Inception reasoning effort", "url": "https://docs.inceptionlabs.ai/capabilities/reasoning-efforts", "type": "primary" }, { "label": "Inception tool use", "url": "https://docs.inceptionlabs.ai/capabilities/tool-use", "type": "primary" }, { "label": "Inception modality FAQ", "url": "https://docs.inceptionlabs.ai/resources/faq", "type": "primary" }, { "label": "Official dated release index", "url": "https://www.inceptionlabs.ai/blog", "type": "primary" }, { "label": "Inception co-founder on model-size nondisclosure", "url": "https://news.ycombinator.com/item?id=50156847", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/inception/mercury-3.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Public launch was October 19, 2026; October 16 is OpenRouter catalog creation. Inception reports more than 40 times Mercury 1 Mini’s total parameters but has not disclosed an absolute total or active count. Its co-founder explicitly confirmed model-size nondisclosure. Diffusion is confirmed; the backbone and dense/MoE structure remain undisclosed. The 1M context is the developer’s nominal specification.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T13:43:58.089113Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/240.json" }, { "id": "inclusionai/ming-3.0-audio-flash", "slug": "inclusionai--ming-3.0-audio-flash", "name": "Ming-3.0-audio-flash", "developer": "InclusionAI", "developer_id": "inclusionai", "family": "ming-3.0", "description": "Audio-language model for spoken questions, audio understanding and optional speech responses.", "release_date": "2026-10-19", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 126, "active_parameters_b": 6.5, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text", "audio" ], "output": [ "text", "audio" ] }, "reasoning": null, "tool_call": true, "source_url": "https://huggingface.co/inclusionAI/Ming-3.0-audio-flash", "source_type": "primary", "sources": [ { "label": "Pinned official model card", "url": "https://huggingface.co/inclusionAI/Ming-3.0-audio-flash/blob/db94e0de19f6d05adc0413c255a28eb480cd1fae/README.md", "type": "primary" }, { "label": "Pinned checkpoint configuration", "url": "https://huggingface.co/inclusionAI/Ming-3.0-audio-flash/blob/db94e0de19f6d05adc0413c255a28eb480cd1fae/config.json", "type": "primary" }, { "label": "Official checkpoint inventory", "url": "https://huggingface.co/api/models/inclusionAI/Ming-3.0-audio-flash/revision/db94e0de19f6d05adc0413c255a28eb480cd1fae", "type": "primary" }, { "label": "Ant API release chronology", "url": "https://developer.ant-ling.com/zh-CN/docs/getting-started/changelog/", "type": "primary" }, { "label": "AntLing public launch announcement", "url": "https://x.com/AntLingAGI/status/2112210811997540475", "type": "primary" }, { "label": "Read-only mirror of original AntLing launch thread", "url": "https://teapaw.org/AntLingAGI/status/2112210811997540475", "type": "archive" }, { "label": "Official model and inference repository", "url": "https://github.com/inclusionAI/Ming-3.0-audio-flash", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:13:45.595Z", "last_seen": "2026-10-30T07:13:45.595Z", "upstream_present": false, "availability": "public", "notes": "Nominal total and active counts are explicitly reported by inclusionAI; they are not a full tensor census or an independently defined Talker-inclusive count. HF safetensors metadata has a different stored-element total. Context comes from released text_config.max_position_embeddings. October 19 is the earliest verified public launch: the Ant API changelog dates Flash access and the AntLing open-source announcement names both models. September 3 repository creation and earlier commits do not prove public access; October weight revisions replace the earlier ASR-oriented artifacts. X was blocked; the original author’s announcement was read through its public mirror. Speech output uses the included optional Talker." }, { "id": "inclusionai/ming-3.0-audio-tiny", "slug": "inclusionai--ming-3.0-audio-tiny", "name": "Ming-3.0-audio-tiny", "developer": "InclusionAI", "developer_id": "inclusionai", "family": "ming-3.0", "description": "Audio-language model for spoken questions, audio understanding and optional speech responses.", "release_date": "2026-10-19", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 9.3, "active_parameters_b": 2.7, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text", "audio" ], "output": [ "text", "audio" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/inclusionAI/Ming-3.0-audio-tiny", "source_type": "primary", "sources": [ { "label": "Pinned official model card", "url": "https://huggingface.co/inclusionAI/Ming-3.0-audio-tiny/blob/401b1abc63de0862dfbc9cd2a40895019d742d09/README.md", "type": "primary" }, { "label": "Pinned checkpoint configuration", "url": "https://huggingface.co/inclusionAI/Ming-3.0-audio-tiny/blob/401b1abc63de0862dfbc9cd2a40895019d742d09/config.json", "type": "primary" }, { "label": "Official checkpoint inventory", "url": "https://huggingface.co/api/models/inclusionAI/Ming-3.0-audio-tiny/revision/401b1abc63de0862dfbc9cd2a40895019d742d09", "type": "primary" }, { "label": "Ant API release chronology", "url": "https://developer.ant-ling.com/zh-CN/docs/getting-started/changelog/", "type": "primary" }, { "label": "AntLing public launch announcement", "url": "https://x.com/AntLingAGI/status/2112210811997540475", "type": "primary" }, { "label": "Read-only mirror of original AntLing launch thread", "url": "https://teapaw.org/AntLingAGI/status/2112210811997540475", "type": "archive" }, { "label": "Official model and inference repository", "url": "https://github.com/inclusionAI/Ming-3.0-audio-flash", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:13:45.595Z", "last_seen": "2026-10-30T07:13:45.595Z", "upstream_present": false, "availability": "public", "notes": "Nominal total and active counts are explicitly reported by inclusionAI; they are not a full tensor census or an independently defined Talker-inclusive count. HF safetensors metadata has a different stored-element total. Context comes from released text_config.max_position_embeddings. October 19 is the earliest verified public launch: the Ant API changelog dates Flash access and the AntLing open-source announcement names both models. September 3 repository creation and earlier commits do not prove public access; October weight revisions replace the earlier ASR-oriented artifacts. X was blocked; the original author’s announcement was read through its public mirror. Speech output uses the included optional Talker." }, { "id": "minimax/MiniMax-M3.1-Flash", "slug": "minimax--minimax-m3.1-flash", "name": "MiniMax-M3.1-Flash", "developer": "MiniMax", "developer_id": "minimax", "family": "minimax", "description": "Native multimodal MoE model for coding and agentic tasks, with text, image and video inputs, text output, a 1M-token context window and adjustable reasoning effort.", "release_date": "2026-10-19", "date_precision": "day", "open_weights": true, "license": "MiniMax Community License", "parameters_b": 428, "active_parameters_b": 23, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.minimax.io/blog/minimax-m31f", "source_type": "primary", "sources": [ { "label": "MiniMax M3.1 Flash: Redefining Agent Performance and Cost", "url": "https://www.minimax.io/blog/minimax-m31f", "type": "primary" }, { "label": "MiniMax-M3.1-Flash official model card", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M3.1-Flash/blob/262ae6830934f3512211081b8a3f16425b93a239/README.md", "type": "primary" }, { "label": "MiniMax-M3.1-Flash checkpoint configuration", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M3.1-Flash/blob/262ae6830934f3512211081b8a3f16425b93a239/config.json", "type": "weights" }, { "label": "MiniMax Community License for M3.1-Flash", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M3.1-Flash/blob/262ae6830934f3512211081b8a3f16425b93a239/LICENSE", "type": "weights" }, { "label": "Official M3.1-Flash checkpoint shard index", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M3.1-Flash/blob/262ae6830934f3512211081b8a3f16425b93a239/model.safetensors.index.json", "type": "weights" }, { "label": "MiniMax Model Invocation documentation", "url": "https://platform.minimax.io/docs/guides/text-generation", "type": "primary" }, { "label": "Tool Use and Interleaved Thinking", "url": "https://platform.minimax.io/docs/guides/text-m3-function-call", "type": "primary" }, { "label": "Official checkpoint repository metadata", "url": "https://huggingface.co/api/models/MiniMaxAI/MiniMax-M3.1-Flash", "type": "weights" }, { "label": "MiniMax model release notes", "url": "https://platform.minimax.io/docs/release-notes/models", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/minimax/MiniMax-M3.1-Flash.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Counts are MiniMax's nominal approximately 428B total and 23B active per token; the model card does not define the DSpark count boundary or active-component convention. Quantized stored-element totals are not parameter totals. API/card context is 1,000,000 tokens; checkpoint max_position_embeddings is 1,048,576. Public weights use the conditional MiniMax Community License.", "field_status": {}, "audited_at": "2026-10-26T22:38:26.260376Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/308.json" }, { "id": "deepseek/deepseek-v4.1-flash-1017", "slug": "deepseek--deepseek-v4.1-flash-1017", "name": "DeepSeek V4.1 Flash 1017", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-flash", "description": "October 17 update of DeepSeek V4.1 Flash with improved reasoning, coding, agentic tasks, and visual understanding", "release_date": "2026-10-17", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 552, "active_parameters_b": 16, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepseek/deepseek-v4.1-flash-1017.toml", "source_type": "catalog", "sources": [ { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepseek/deepseek-v4.1-flash-1017.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017", "type": "weights" }, { "label": "DeepSeek October 17 changelog", "url": "https://api-docs.deepseek.com/updates/", "type": "primary" }, { "label": "DeepSeek V4.1 technical report", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash/resolve/main/DeepSeek_V41_Tech_Report.pdf", "type": "primary" }, { "label": "Exact 1017 configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017/blob/main/config.json", "type": "primary" }, { "label": "DeepSeek model-version API documentation", "url": "https://api-docs.deepseek.com/quick_start/pricing/", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "552B is the nominal backbone count; the technical report separately lists 196B Engram parameters (748B nominal combined), while Hugging Face tensor metadata displays 763B for the checkpoint. Active parameters are phase-dependent: 8B per input token during prefill and 16B per output token during decode. The 1017 card links the launch report; its config is byte-identical to the launch config. Context is advertised as 1M; the weights configure 1,048,576 positions.", "parameters_scope": "backbone", "active_parameters_scope": "decode", "parameter_components_b": { "backbone": 552, "engram": 196 }, "active_parameters_by_phase_b": { "prefill": 8, "decode": 16 }, "field_status": {}, "audited_at": "2026-10-26T13:46:34.227057Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/002.json" }, { "id": "elorian/glimpse", "slug": "elorian--glimpse", "name": "Glimpse", "developer": "Elorian", "developer_id": "elorian", "family": "glimpse", "description": null, "release_date": "2026-10-16", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://elorian.ai/blog/introducing-glimpse", "source_type": "primary", "sources": [ { "label": "Elorian Glimpse launch", "url": "https://elorian.ai/blog/introducing-glimpse", "type": "primary" }, { "label": "Elorian company and model description", "url": "https://elorian.ai/", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:07:44.790Z", "last_seen": "2026-10-30T07:07:44.790Z", "upstream_present": false, "availability": "public", "notes": "Company post-training of an unnamed open-weight base model. Launch confirms public API and chat access on October16. Glimpse checkpoint weights, license, parameter counts, backbone and context length are not disclosed in the reviewed announcement; the base model's weight status is not inherited." }, { "id": "nvidia/nemotron-labs-dory1.0-4b", "slug": "nvidia--nemotron-labs-dory1.0-4b", "name": "Nemotron Labs Dory 1.0 4B", "developer": "NVIDIA", "developer_id": "nvidia", "family": "nemotron", "description": "Latent-recurrent language model with configurable layer looping.", "release_date": "2026-10-16", "date_precision": "day", "open_weights": true, "license": "OpenMDW-1.1", "parameters_b": 4, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-Labs-Dory1.0-4B-BF16/blob/48e0c3faac0aaa8b93fc4e68256d7e58b183645e/README.md", "source_type": "primary", "sources": [ { "label": "Official model card and downloadable weights", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-Labs-Dory1.0-4B-BF16/blob/48e0c3faac0aaa8b93fc4e68256d7e58b183645e/README.md", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "notes": "Looped latent-recurrent normalized-GPT with shared recurrent layers. Repeating layers changes inference compute, not the number of stored parameters. Official card gives the October 16 release date and 4B nominal parameters." }, { "id": "ifm/jais-3-0.9b", "slug": "ifm--jais-3-0.9b", "name": "Jais 3 0.9B", "developer": "Institute of Foundation Models", "developer_id": "ifm", "family": "jais-3", "description": "Arabic-focused language model adapted by its publisher from K2 Horizon.", "release_date": "2026-10-15", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 1.078285824, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://ifm.ai/blog/jais3/", "source_type": "primary", "sources": [ { "label": "Dated publisher family release", "url": "https://ifm.ai/blog/jais3/", "type": "primary" }, { "label": "Official versioned model card", "url": "https://huggingface.co/inception42/Jais-3-0.9B/blob/ef8b0e93cae13e7fdb6c711770ab00e03616baac/README.md", "type": "primary" }, { "label": "Versioned model configuration", "url": "https://huggingface.co/inception42/Jais-3-0.9B/blob/ef8b0e93cae13e7fdb6c711770ab00e03616baac/config.json", "type": "primary" }, { "label": "Official checkpoint parameter metadata", "url": "https://huggingface.co/api/models/IFM/Jais-3-0.9B", "type": "primary" }, { "label": "Publisher collection links checkpoints", "url": "https://huggingface.co/collections/IFM/jais-3", "type": "primary" } ], "aliases": [ "inception42/jais-3-0.9b" ], "first_seen": "2026-10-30T07:13:05.399Z", "last_seen": "2026-10-30T07:13:05.399Z", "upstream_present": false, "availability": "public", "notes": "Final trained Jais3checkpoint; intermediate training stages and quantized copies are not separate models. Total follows the official HuggingFace safetensors metadata, which exceeds the nominal size in the name; the difference is not independently reconciled. The versioned configuration establishes exact context. The collection links the current inception42repository, and the earlier IFMURL resolves to it. Text-only; JaisASR is a separate speech model, not a native audio modality. Tool calling was not documented for the0.9Bcard." }, { "id": "ifm/jais-3-3.7b", "slug": "ifm--jais-3-3.7b", "name": "Jais 3 3.7B", "developer": "Institute of Foundation Models", "developer_id": "ifm", "family": "jais-3", "description": "Arabic-focused language model adapted by its publisher from K2 Horizon.", "release_date": "2026-10-15", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 5.05825536, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 524288, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ifm.ai/blog/jais3/", "source_type": "primary", "sources": [ { "label": "Dated publisher family release", "url": "https://ifm.ai/blog/jais3/", "type": "primary" }, { "label": "Official versioned model card", "url": "https://huggingface.co/inception42/Jais-3-3.7B/blob/78d52c03900ef18ee3ebae631426556df6bd925a/README.md", "type": "primary" }, { "label": "Versioned model configuration", "url": "https://huggingface.co/inception42/Jais-3-3.7B/blob/78d52c03900ef18ee3ebae631426556df6bd925a/config.json", "type": "primary" }, { "label": "Official checkpoint parameter metadata", "url": "https://huggingface.co/api/models/IFM/Jais-3-3.7B", "type": "primary" }, { "label": "Publisher collection links checkpoints", "url": "https://huggingface.co/collections/IFM/jais-3", "type": "primary" } ], "aliases": [ "inception42/jais-3-3.7b" ], "first_seen": "2026-10-30T07:13:05.399Z", "last_seen": "2026-10-30T07:13:05.399Z", "upstream_present": false, "availability": "public", "notes": "Final trained Jais3checkpoint; intermediate training stages and quantized copies are not separate models. Total follows the official HuggingFace safetensors metadata, which exceeds the nominal size in the name; the difference is not independently reconciled. The versioned configuration establishes exact context. The collection links the current inception42repository, and the earlier IFMURL resolves to it. Text-only; JaisASR is a separate speech model, not a native audio modality." }, { "id": "ifm/jais-3-32b", "slug": "ifm--jais-3-32b", "name": "Jais 3 32B", "developer": "Institute of Foundation Models", "developer_id": "ifm", "family": "jais-3", "description": "Arabic-focused language model adapted by its publisher from K2 Horizon.", "release_date": "2026-10-15", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 34.77930496, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 524288, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ifm.ai/blog/jais3/", "source_type": "primary", "sources": [ { "label": "Dated publisher family release", "url": "https://ifm.ai/blog/jais3/", "type": "primary" }, { "label": "Official versioned model card", "url": "https://huggingface.co/inception42/Jais-3-32B/blob/2d4da325cfa0d28b9f2eb4ddac2baa1ae0ed6063/README.md", "type": "primary" }, { "label": "Versioned model configuration", "url": "https://huggingface.co/inception42/Jais-3-32B/blob/2d4da325cfa0d28b9f2eb4ddac2baa1ae0ed6063/config.json", "type": "primary" }, { "label": "Official checkpoint parameter metadata", "url": "https://huggingface.co/api/models/IFM/Jais-3-32B", "type": "primary" }, { "label": "Publisher collection links checkpoints", "url": "https://huggingface.co/collections/IFM/jais-3", "type": "primary" } ], "aliases": [ "inception42/jais-3-32b" ], "first_seen": "2026-10-30T07:13:05.399Z", "last_seen": "2026-10-30T07:13:05.399Z", "upstream_present": false, "availability": "public", "notes": "Final trained Jais3checkpoint; intermediate training stages and quantized copies are not separate models. Total follows the official HuggingFace safetensors metadata, which exceeds the nominal size in the name; the difference is not independently reconciled. The versioned configuration establishes exact context. The collection links the current inception42repository, and the earlier IFMURL resolves to it. Text-only; JaisASR is a separate speech model, not a native audio modality." }, { "id": "ifm/jais-3-375b-a23b", "slug": "ifm--jais-3-375b-a23b", "name": "Jais 3 375B A23B", "developer": "Institute of Foundation Models", "developer_id": "ifm", "family": "jais-3", "description": "Arabic-focused language model adapted by its publisher from K2 Horizon.", "release_date": "2026-10-15", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 379.167159168, "active_parameters_b": 23, "parameters_source": "primary_source", "architecture": "moe", "context_window": 524288, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ifm.ai/blog/jais3/", "source_type": "primary", "sources": [ { "label": "Dated publisher family release", "url": "https://ifm.ai/blog/jais3/", "type": "primary" }, { "label": "Official versioned model card", "url": "https://huggingface.co/inception42/Jais-3-375B-A23B/blob/0478c3bc69be313f1c4f898b55c8905413195cf9/README.md", "type": "primary" }, { "label": "Versioned model configuration", "url": "https://huggingface.co/inception42/Jais-3-375B-A23B/blob/0478c3bc69be313f1c4f898b55c8905413195cf9/config.json", "type": "primary" }, { "label": "Official checkpoint parameter metadata", "url": "https://huggingface.co/api/models/IFM/Jais-3-375B-A23B", "type": "primary" }, { "label": "Publisher collection links checkpoints", "url": "https://huggingface.co/collections/IFM/jais-3", "type": "primary" } ], "aliases": [ "inception42/jais-3-375b-a23b" ], "first_seen": "2026-10-30T07:13:05.399Z", "last_seen": "2026-10-30T07:13:05.399Z", "upstream_present": false, "availability": "public", "notes": "Final trained Jais3checkpoint; intermediate training stages and quantized copies are not separate models. Total follows the official HuggingFace safetensors metadata, which exceeds the nominal size in the name; the difference is not independently reconciled. Activecount is the publisher-reported nominal convention, not an independent full-network active census. The versioned configuration establishes exact context. The collection links the current inception42repository, and the earlier IFMURL resolves to it. Text-only; JaisASR is a separate speech model, not a native audio modality." }, { "id": "ifm/jais-3-7b", "slug": "ifm--jais-3-7b", "name": "Jais 3 7B", "developer": "Institute of Foundation Models", "developer_id": "ifm", "family": "jais-3", "description": "Arabic-focused language model adapted by its publisher from K2 Horizon.", "release_date": "2026-10-15", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 8.99917824, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 524288, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ifm.ai/blog/jais3/", "source_type": "primary", "sources": [ { "label": "Dated publisher family release", "url": "https://ifm.ai/blog/jais3/", "type": "primary" }, { "label": "Official versioned model card", "url": "https://huggingface.co/inception42/Jais-3-7B/blob/c7daf3c43355921602398a0d97bbeba05bfbe81b/README.md", "type": "primary" }, { "label": "Versioned model configuration", "url": "https://huggingface.co/inception42/Jais-3-7B/blob/c7daf3c43355921602398a0d97bbeba05bfbe81b/config.json", "type": "primary" }, { "label": "Official checkpoint parameter metadata", "url": "https://huggingface.co/api/models/IFM/Jais-3-7B", "type": "primary" }, { "label": "Publisher collection links checkpoints", "url": "https://huggingface.co/collections/IFM/jais-3", "type": "primary" } ], "aliases": [ "inception42/jais-3-7b" ], "first_seen": "2026-10-30T07:13:05.399Z", "last_seen": "2026-10-30T07:13:05.399Z", "upstream_present": false, "availability": "public", "notes": "Final trained Jais3checkpoint; intermediate training stages and quantized copies are not separate models. Total follows the official HuggingFace safetensors metadata, which exceeds the nominal size in the name; the difference is not independently reconciled. The versioned configuration establishes exact context. The collection links the current inception42repository, and the earlier IFMURL resolves to it. Text-only; JaisASR is a separate speech model, not a native audio modality." }, { "id": "ifm/jais-3-mova-36b-a4b", "slug": "ifm--jais-3-mova-36b-a4b", "name": "Jais 3 MoVA 36B A4B", "developer": "Institute of Foundation Models", "developer_id": "ifm", "family": "jais-3", "description": "Arabic-focused language model adapted by its publisher from K2 Horizon.", "release_date": "2026-10-15", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 37.44479202, "active_parameters_b": 4, "parameters_source": "primary_source", "architecture": "moe", "context_window": 524288, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ifm.ai/blog/jais3/", "source_type": "primary", "sources": [ { "label": "Dated publisher family release", "url": "https://ifm.ai/blog/jais3/", "type": "primary" }, { "label": "Official versioned model card", "url": "https://huggingface.co/inception42/Jais-3-MoVA-36B-A4B/blob/3170dd9cdcebbdc214514b255e334b9ca6c2c265/README.md", "type": "primary" }, { "label": "Versioned model configuration", "url": "https://huggingface.co/inception42/Jais-3-MoVA-36B-A4B/blob/3170dd9cdcebbdc214514b255e334b9ca6c2c265/config.json", "type": "primary" }, { "label": "Official checkpoint parameter metadata", "url": "https://huggingface.co/api/models/IFM/Jais-3-MoVA-36B-A4B", "type": "primary" }, { "label": "Publisher collection links checkpoints", "url": "https://huggingface.co/collections/IFM/jais-3", "type": "primary" } ], "aliases": [ "inception42/jais-3-mova-36b-a4b" ], "first_seen": "2026-10-30T07:13:05.399Z", "last_seen": "2026-10-30T07:13:05.399Z", "upstream_present": false, "availability": "public", "notes": "Final trained Jais3checkpoint; intermediate training stages and quantized copies are not separate models. Total follows the official HuggingFace safetensors metadata, which exceeds the nominal size in the name; the difference is not independently reconciled. Activecount is the publisher-reported nominal convention, not an independent full-network active census. The versioned configuration establishes exact context. The collection links the current inception42repository, and the earlier IFMURL resolves to it. Text-only; JaisASR is a separate speech model, not a native audio modality." }, { "id": "meta/muse-spark-open", "slug": "meta--muse-spark-open", "name": "Muse Spark Open", "developer": "Meta", "developer_id": "meta", "family": "muse", "description": "Muse Spark Open is Meta’s open-weight multimodal MoE Transformer, published as 816B total and 42B active parameters, for coding and agentic work with a 1,048,576-token context window.", "release_date": "2026-10-15", "date_precision": "day", "open_weights": true, "license": "Muse Model License", "parameters_b": 816, "active_parameters_b": 42, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://research.meta.ai/blog/muse-spark-open", "source_type": "primary", "sources": [ { "label": "Announcement", "url": "https://research.meta.ai/blog/muse-spark-open", "type": "announcement" }, { "label": "Hugging Face (BF16)", "url": "https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B", "type": "weights" }, { "label": "Developer docs", "url": "https://dev.meta.ai/docs/muse-spark-open", "type": "docs" }, { "label": "Meta prompting guide", "url": "https://dev.meta.ai/docs/muse-spark-open/prompting", "type": "primary" }, { "label": "Official BF16 metadata", "url": "https://huggingface.co/api/models/meta-models/Muse-Spark-Open-816B-A42B", "type": "weights" }, { "label": "Official NVFP4 metadata", "url": "https://huggingface.co/api/models/meta-models/Muse-Spark-Open-816B-A42B-NVFP4", "type": "weights" }, { "label": "Transformers model implementation", "url": "https://github.com/huggingface/transformers/blob/main/src/transformers/models/muse_spark/modeling_muse_spark.py", "type": "primary" }, { "label": "Meta model lineup", "url": "https://dev.meta.ai/docs/models", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/meta/muse-spark-open.toml", "type": "catalog" }, { "label": "License", "url": "https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B/blob/main/LICENSE.md", "type": "license" }, { "label": "Hugging Face (NVFP4)", "url": "https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B-NVFP4", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "816B total and 42B active are Meta’s nominal published figures, not independently counted exact totals. BF16 repository metadata aggregates 822.036529664B stored tensor elements; its difference from the advertised 816B remains unreconciled because checkpoint config, tensor index and header access returned 401. The full released architecture includes language, vision and audio components; no verified component accounting establishes which are included in the nominal counts. Quantized storage elements, scale tensors and buffers are not automatically learned parameters. The 256 experts use top-8 routing; shared pre/post projections are not additional shared experts. This Open release is distinct from hosted Muse Spark 1.1, 1.2 and 1.3.", "field_status": {}, "audited_at": "2026-10-26T21:55:16.284223Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/285.json" }, { "id": "nvidia/nemotron-3.5-bolt-74b-a9b", "slug": "nvidia--nemotron-3.5-bolt-74b-a9b", "name": "Nemotron 3.5 Bolt 74B A9B", "developer": "NVIDIA", "developer_id": "nvidia", "family": "nemotron", "description": "Multimodal mixture-of-experts model for local agent workflows.", "release_date": "2026-10-15", "date_precision": "day", "open_weights": true, "license": "OpenMDW-1.1", "parameters_b": 74, "active_parameters_b": 9, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://blogs.nvidia.com/blog/nemotron-3-5-super-vl/", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://blogs.nvidia.com/blog/nemotron-3-5-super-vl/", "type": "primary" }, { "label": "Official model card and downloadable weights", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16/blob/e09a4cedb1787db1760375ee5539f222cae0c672/README.md", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "notes": "Nominal 74B total and 9B active from the official card. The BF16 card documents a 256K shared context window. Quantized 24GB and BF16 distributions represent the same named model." }, { "id": "nvidia/nemotron-3.5-super-vl-120b-a12b", "slug": "nvidia--nemotron-3.5-super-vl-120b-a12b", "name": "Nemotron 3.5 Super VL 120B A12B", "developer": "NVIDIA", "developer_id": "nvidia", "family": "nemotron", "description": "NVIDIA vision-language model with a C-RADIO encoder and hybrid Mamba-2, attention and LatentMoE decoder, trained for reasoning and tool use over text, images and video.", "release_date": "2026-10-15", "date_precision": "day", "open_weights": true, "license": "OpenMDW-1.1", "parameters_b": 124.453374976, "active_parameters_b": 12, "parameters_source": "primary_source", "architecture": "hybrid-ssm-transformer-moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16/blob/fb0ee0bc6b3f70342afd8d84cb80dbadbaa8a1fe/README.md", "source_type": "primary", "sources": [ { "label": "Pinned Nemotron 3.5 Super VL BF16 model card", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16/blob/fb0ee0bc6b3f70342afd8d84cb80dbadbaa8a1fe/README.md", "type": "primary" }, { "label": "Pinned Super VL model configuration", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16/blob/fb0ee0bc6b3f70342afd8d84cb80dbadbaa8a1fe/config.json", "type": "primary" }, { "label": "Complete BF16 checkpoint tensor index", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16/blob/fb0ee0bc6b3f70342afd8d84cb80dbadbaa8a1fe/model.safetensors.index.json", "type": "primary" }, { "label": "Pinned vision-language composition and connector", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16/blob/fb0ee0bc6b3f70342afd8d84cb80dbadbaa8a1fe/modeling_nemotron_h_omni.py", "type": "primary" }, { "label": "Pinned C-RADIO implementation and checkpoint conversion", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16/blob/fb0ee0bc6b3f70342afd8d84cb80dbadbaa8a1fe/modeling_radio.py", "type": "primary" }, { "label": "Nemotron 3.5 Super and Lightning technical report, October 21, 2026", "url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3.5-Super-and-Lightning-Technical-Report.pdf", "type": "primary" }, { "label": "Nemotron 3.5 Super training recipe and architecture", "url": "https://docs.nvidia.com/nemotron/latest/nemotron/super35/README.html", "type": "primary" }, { "label": "October 15 Super VL launch announcement", "url": "https://blogs.nvidia.com/blog/nemotron-3-5-super-vl/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/nvidia/nemotron-3.5-super-vl-120b-a12b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete BF16 checkpoint: 124.453374976B learned parameters, including vision, connector and one shared MTP module. Publisher 12B active is nominal for the language decoder; vision encoding and speculative drafting are separate phases. Excludes persistent buffers and identity LayerScale vectors initialized by the loader but absent from released weights.", "active_parameters_scope": "language_backbone", "parameter_components_b": { "language_backbone": 120.66868736, "mtp": 2.942324736, "vision_encoder": 0.65361152, "vision_connector": 0.18875136 }, "field_status": {}, "audited_at": "2026-10-27T00:56:03.868222+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/388.json" }, { "id": "ibm/granite-switch-4.1-30b", "slug": "ibm--granite-switch-4.1-30b", "name": "Granite Switch 4.1 30B", "developer": "IBM", "developer_id": "ibm", "family": "granite-switch", "description": "Language model composed with decision and safety adapters and a speech-to-text encoder.", "release_date": "2026-10-14", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 30, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "audio" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/ibm-granite/granite-switch-4.1-30b/blob/granite-switch-4.1-30b/README.md", "source_type": "primary", "sources": [ { "label": "Official version-specific model card and weights", "url": "https://huggingface.co/ibm-granite/granite-switch-4.1-30b/blob/granite-switch-4.1-30b/README.md", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "parameters_scope": "backbone", "notes": "The parameter value covers the 30B language backbone, not the complete composed system with adapters and a 470M speech encoder. Audio is transcribed and inserted into the chat context; the speech encoder and LLM are not jointly trained. Original revision; its dated model card is preserved at the named release tag.", "parameter_components_b": { "backbone": 30 } }, { "id": "ibm/granite-switch-4.1-8b", "slug": "ibm--granite-switch-4.1-8b", "name": "Granite Switch 4.1 8B", "developer": "IBM", "developer_id": "ibm", "family": "granite-switch", "description": "Language model composed with decision and safety adapters and a speech-to-text encoder.", "release_date": "2026-10-14", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "audio" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/ibm-granite/granite-switch-4.1-8b/blob/granite-switch-4.1-8b/README.md", "source_type": "primary", "sources": [ { "label": "Official version-specific model card and weights", "url": "https://huggingface.co/ibm-granite/granite-switch-4.1-8b/blob/granite-switch-4.1-8b/README.md", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "parameters_scope": "backbone", "notes": "The parameter value covers the 8B language backbone, not the complete composed system with adapters and a 470M speech encoder. Audio is transcribed and inserted into the chat context; the speech encoder and LLM are not jointly trained. Original revision; its dated model card is preserved at the named release tag.", "parameter_components_b": { "backbone": 8 } }, { "id": "xai/grok-voice-3.0", "slug": "xai--grok-voice-3.0", "name": "Grok Voice 3.0", "developer": "xAI", "developer_id": "xai", "family": "grok-voice", "description": null, "release_date": "2026-10-14", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text", "audio" ], "output": [ "text", "audio" ] }, "reasoning": null, "tool_call": null, "source_url": "https://docs.x.ai/developers/release-notes", "source_type": "primary", "sources": [ { "label": "Dated publisher release notes", "url": "https://docs.x.ai/developers/release-notes", "type": "documentation" }, { "label": "Conversational audio capability", "url": "https://docs.x.ai/developers/models/grok-voice-3.0", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:31:17.277Z", "last_seen": "2026-10-29T10:31:17.277Z", "upstream_present": false, "availability": "public", "notes": "Conversational speech model; separate from the dedicated transcription service. " }, { "id": "google/gemini-4-argon", "slug": "google--gemini-4-argon", "name": "Gemini 4 Argon", "developer": "Google", "developer_id": "google", "family": "gemini", "description": "Google Gemini reasoning model with multimodal input, text output, function calling, and a unified 1,048,576-token input/output context window.", "release_date": "2026-10-13", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ai.google.dev/gemini-api/docs/models/gemini-4-argon", "source_type": "primary", "sources": [ { "label": "Google DeepMind model card", "url": "https://deepmind.google/models/model-cards/gemini-4-argon/", "type": "primary" }, { "label": "Gemini API specification", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-4-argon", "type": "primary" }, { "label": "Public release announcement", "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-argon/", "type": "primary" }, { "label": "Earlier restricted rollout", "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/", "type": "primary" }, { "label": "Gemini API terms", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Safety and security report", "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf", "type": "primary" }, { "label": "Google Cloud release date", "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/4-argon", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemini-4-argon.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Sparse MoE Transformer; Google identifies Argon as a new model, not a fine-tune of an earlier one. Total and active parameter counts are undisclosed. Context is a shared input/output budget of 1,048,576 tokens. September 30, 2026 was a restricted Fairwind rollout; public API availability began October 13, 2026.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T13:08:01.764454Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/213.json" }, { "id": "liquid/lfm2.5-24b-a2b", "slug": "liquid--lfm2.5-24b-a2b", "name": "LFM2.5-24B-A2B", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm2.5", "description": "Hybrid mixture-of-experts reasoning model for local agents.", "release_date": "2026-10-13", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 24, "active_parameters_b": 2.45, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.liquid.ai/blog/lfm2-5-24b-a2b", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/lfm2-5-24b-a2b", "type": "primary" }, { "label": "Official model card and downloadable weights", "url": "https://huggingface.co/LiquidAI/LFM2.5-24B-A2B/blob/3fcfdbfa1a1b6d8166846a6323e69f1e1254440d/README.md", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "notes": "Distinct from LFM2-24B-A2B. The developer release explicitly publishes both base and post-trained checkpoints on October 13; card specifies 24.0B total and 2.45B active parameters." }, { "id": "liquid/lfm2.5-24b-a2b-base", "slug": "liquid--lfm2.5-24b-a2b-base", "name": "LFM2.5-24B-A2B-Base", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm2.5", "description": "Pretrained hybrid mixture-of-experts language model.", "release_date": "2026-10-13", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 24, "active_parameters_b": 2.45, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.liquid.ai/blog/lfm2-5-24b-a2b", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/lfm2-5-24b-a2b", "type": "primary" }, { "label": "Official model card and downloadable weights", "url": "https://huggingface.co/LiquidAI/LFM2.5-24B-A2B-Base/blob/9156f24f4a227a259b09011f4e0295aa8f57db01/README.md", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "notes": "Distinct from LFM2-24B-A2B. The developer release explicitly publishes both base and post-trained checkpoints on October 13; card specifies 24.0B total and 2.45B active parameters." }, { "id": "meta/muse-wonder-1.0", "slug": "meta--muse-wonder-1.0", "name": "Muse Wonder", "developer": "Meta", "developer_id": "meta", "family": "muse", "description": "Meta's hosted multimodal reasoning model for coding, computer use, and tasks that require tools across multiple steps. Available through Meta Model API and Muse Code.", "release_date": "2026-10-12", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1048576, "modalities": { "input": [ "text", "image", "video", "pdf", "audio" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://research.meta.ai/blog/muse-wonder", "source_type": "primary", "sources": [ { "label": "Announcement", "url": "https://research.meta.ai/blog/muse-wonder", "type": "announcement" }, { "label": "Model card", "url": "https://research.meta.ai/static/muse-wonder-model-card", "type": "model_card" }, { "label": "Meta API model specifications", "url": "https://dev.meta.ai/docs/models", "type": "primary" }, { "label": "Meta API reasoning", "url": "https://dev.meta.ai/docs/reasoning", "type": "primary" }, { "label": "Meta Model API terms", "url": "https://dev.meta.ai/legal/terms-of-service", "type": "primary" }, { "label": "Muse Wonder model page", "url": "https://dev.meta.ai/models/muse-wonder", "type": "primary" }, { "label": "Spark Open lineage (PyTorch Team at Meta)", "url": "https://pytorch.org/blog/muse-spark-open/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/meta/muse-wonder-1.0.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "This entry is muse-wonder-1.0, distinct from subscription-only Muse Wonder Full. Max reasoning effort does not select Full; none is unsupported. Total parameters, active parameters and architecture remain undisclosed. Meta describes larger bases than Spark 1.3, so Spark Open’s 816B total/42B active cannot be transferred through the documented Spark lineage. October 23 is a model-card evaluation correction, not the release date.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T14:28:29.785480Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/286.json" }, { "id": "meta/muse-wonder-full", "slug": "meta--muse-wonder-full", "name": "Muse Wonder Full", "developer": "Meta", "developer_id": "meta", "family": "muse", "description": "Meta's multimodal reasoning model for coding, computer use and tasks involving tools, available through Muse Code subscriptions. Full is a separate variant from the Muse Wonder API model.", "release_date": "2026-10-12", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1000000, "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://research.meta.ai/blog/muse-wonder", "source_type": "primary", "sources": [ { "label": "Announcement", "url": "https://research.meta.ai/blog/muse-wonder", "type": "announcement" }, { "label": "Model card", "url": "https://research.meta.ai/static/muse-wonder-model-card", "type": "model_card" }, { "label": "Meta Full-specific context and access", "url": "https://dev.meta.ai/models/muse-wonder", "type": "primary" }, { "label": "Meta model availability and variant distinction", "url": "https://dev.meta.ai/docs/models", "type": "primary" }, { "label": "Meta developer launch and capabilities", "url": "https://dev.meta.ai/resources/blog/build-with-muse-wonder", "type": "primary" }, { "label": "Meta service terms", "url": "https://dev.meta.ai/legal/terms-of-service", "type": "primary" }, { "label": "Meta team Spark Open counts and lineage", "url": "https://pytorch.org/blog/muse-spark-open/", "type": "primary" }, { "label": "Meta team Spark Open training release identity", "url": "https://pytorch.org/blog/torchtitan-update-new-sota-models-performance-at-scale-titanrl/", "type": "primary" }, { "label": "Meta leadership Wonder and Spark pretrain distinction", "url": "https://cheekypint.transistor.fm/35/transcript", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/meta/muse-wonder-full.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Full is subscription-only in Muse Code, not the ordinary Wonder API model at high or max reasoning effort. Context is the published nominal 1M; an exact Full limit of 1,048,576 is not established. Model-card audio modality does not establish direct audio support on every release surface, and PDF support is not transferred from ordinary Wonder API documentation. Full total/active parameters and architecture remain undisclosed. Meta describes larger Wonder bases and a separate pretrain from Spark, so Spark Open's 816B total/42B active counts do not establish Full's size.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T14:30:29.060121Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/287.json" }, { "id": "openbmb/minicpm-v-4.7-1b", "slug": "openbmb--minicpm-v-4.7-1b", "name": "MiniCPM-V 4.7 1B", "developer": "OpenBMB", "developer_id": "openbmb", "family": "minicpm-v", "description": "Vision-language model for single-image, multi-image and video understanding, with optional thinking and tool calling.", "release_date": "2026-10-12", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 1.554476272, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "hybrid", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openbmb.cn/docs/en/changelog", "source_type": "primary", "sources": [ { "label": "MiniCPM-V 4.7 1B model card", "url": "https://huggingface.co/openbmb/MiniCPM-V-4.7-1B/blob/bb99eab4113676f30830ffa73bac1f0b2a5f1dea/README.md", "type": "primary" }, { "label": "OpenBMB release changelog", "url": "https://openbmb.cn/docs/en/changelog", "type": "primary" }, { "label": "MiniCPM-V project README and News", "url": "https://github.com/OpenBMB/MiniCPM-V/blob/1915c89a2508f1cb959991038173a993ea67ffb2/README.md", "type": "primary" }, { "label": "Pinned MiniCPM-V 4.7 1B configuration", "url": "https://huggingface.co/openbmb/MiniCPM-V-4.7-1B/blob/bb99eab4113676f30830ffa73bac1f0b2a5f1dea/config.json", "type": "primary" }, { "label": "Complete singleton BF16 checkpoint tensor header", "url": "https://huggingface.co/openbmb/MiniCPM-V-4.7-1B/resolve/bb99eab4113676f30830ffa73bac1f0b2a5f1dea/model.safetensors", "type": "primary" }, { "label": "Pinned official checkpoint inventory", "url": "https://huggingface.co/api/models/openbmb/MiniCPM-V-4.7-1B/revision/bb99eab4113676f30830ffa73bac1f0b2a5f1dea", "type": "primary" }, { "label": "Native MiniCPM-V 4.7 model implementation", "url": "https://github.com/huggingface/transformers/blob/a906d3c4b65095f2308b6a6a193e934d03b8eb5d/src/transformers/models/minicpmv4_7/modeling_minicpmv4_7.py", "type": "primary" }, { "label": "Native MiniCPM-V 4.7 configuration defaults", "url": "https://github.com/huggingface/transformers/blob/a906d3c4b65095f2308b6a6a193e934d03b8eb5d/src/transformers/models/minicpmv4_7/configuration_minicpmv4_7.py", "type": "primary" }, { "label": "Native Qwen3.5 text implementation used by the checkpoint", "url": "https://github.com/huggingface/transformers/blob/a906d3c4b65095f2308b6a6a193e934d03b8eb5d/src/transformers/models/qwen3_5/modeling_qwen3_5.py", "type": "primary" }, { "label": "Pinned conversation template", "url": "https://huggingface.co/openbmb/MiniCPM-V-4.7-1B/blob/bb99eab4113676f30830ffa73bac1f0b2a5f1dea/chat_template.jinja", "type": "primary" }, { "label": "MiniCPM-V 4.7 knowledge-base model table", "url": "https://openbmb.cn/docs/en/guides/minicpm-v/minicpm-v-4.7", "type": "primary" }, { "label": "Official checkpoint commit history", "url": "https://huggingface.co/api/models/openbmb/MiniCPM-V-4.7-1B/commits/main", "type": "primary" }, { "label": "Later DSpark checkpoint for the exact 1B target", "url": "https://huggingface.co/openbmb/MiniCPM-V-4.7-1B-DSpark", "type": "primary" }, { "label": "MiniCPM-V series overview", "url": "https://openbmb.cn/docs/en/guides/minicpm-v", "type": "primary" }, { "label": "Officially linked MiniCPM-V 4.7 cookbook model table", "url": "https://github.com/OpenSQZ/MiniCPM-V-CookBook/blob/main/docs/pages/v4.7/overview.md", "type": "primary" }, { "label": "OpenBMB on-device model comparison table", "url": "https://github.com/OpenBMB/MiniCPM-V-Apps/blob/main/README.md", "type": "primary" }, { "label": "OpenBMB MiniCPM-V 4.7 1B original benchmark image", "url": "https://raw.githubusercontent.com/OpenBMB/MiniCPM-o/main/assets/minicpmv4.7/overall_performance_1b_instruct.png", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openbmb/minicpm-v-4.7-1b.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/openbmb/MiniCPM-V-4.7-1B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete BF16 checkpoint: 1,554,476,272 learned parameters, comprising 1,006,209,856 in the language model and output head, 522,299,312 in the vision tower and early merger, and 25,967,104 in the visual-to-language merger. OpenBMB currently reports nominal 1.55B; its benchmark image labels Size as 1.3B without defining the scope. Retain the exact complete count. No separate active count was found for this checkpoint; sibling MoE and separate DSpark counts do not apply. The language backbone combines dense MLPs with 18 GatedDeltaNet and six full-attention layers. Inert MTP metadata adds no module. Context is configured at 262,144 tokens. Public release is documented October 12, 2026; the earlier October 6 upload does not establish public availability.", "parameter_components_b": { "language_model_and_head": 1.006209856, "vision_tower_with_early_merger": 0.522299312, "visual_to_language_merger": 0.025967104 }, "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T17:27:23.657767+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/459.json" }, { "id": "openbmb/minicpm-v-4.7-35b-a3b", "slug": "openbmb--minicpm-v-4.7-35b-a3b", "name": "MiniCPM-V 4.7 35B-A3B", "developer": "OpenBMB", "developer_id": "openbmb", "family": "minicpm-v", "description": "Multimodal MoE model for text, image and video input, with optional thinking and function calling.", "release_date": "2026-10-12", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 35.212875824, "active_parameters_b": 3.5, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openbmb.cn/docs/en/changelog", "source_type": "primary", "sources": [ { "label": "MiniCPM-V 4.7 35B-A3B model card", "url": "https://huggingface.co/openbmb/MiniCPM-V-4.7-35B-A3B/blob/ee2d0a272c9d55db52a8b6841b6b24c9d0dda0a3/README.md", "type": "primary" }, { "label": "OpenBMB changelog: MiniCPM-V 4.7", "url": "https://openbmb.cn/docs/en/changelog", "type": "primary" }, { "label": "MiniCPM-V repository news and model overview", "url": "https://github.com/OpenBMB/MiniCPM-V/blob/main/README.md", "type": "primary" }, { "label": "Pinned MiniCPM-V 4.7 35B-A3B configuration", "url": "https://huggingface.co/openbmb/MiniCPM-V-4.7-35B-A3B/blob/ee2d0a272c9d55db52a8b6841b6b24c9d0dda0a3/config.json", "type": "primary" }, { "label": "Pinned BF16 shards and safetensors index", "url": "https://huggingface.co/openbmb/MiniCPM-V-4.7-35B-A3B/tree/ee2d0a272c9d55db52a8b6841b6b24c9d0dda0a3", "type": "primary" }, { "label": "OpenBMB-contributed Transformers MiniCPM-V 4.7 implementation", "url": "https://github.com/huggingface/transformers/blob/v5.18.0/src/transformers/models/minicpmv4_7/modeling_minicpmv4_7.py", "type": "primary" }, { "label": "MiniCPM-V 4.7 configuration defaults", "url": "https://github.com/huggingface/transformers/blob/v5.18.0/src/transformers/models/minicpmv4_7/configuration_minicpmv4_7.py", "type": "primary" }, { "label": "Selected Qwen3.5-MoE text implementation", "url": "https://github.com/huggingface/transformers/blob/v5.18.0/src/transformers/models/qwen3_5_moe/modeling_qwen3_5_moe.py", "type": "primary" }, { "label": "MiniCPM-V 4.7 model list and capabilities", "url": "https://openbmb.cn/docs/en/guides/minicpm-v/minicpm-v-4.7", "type": "primary" }, { "label": "OpenBMB on-device model table", "url": "https://github.com/OpenBMB/MiniCPM-V-Apps/blob/main/README.md", "type": "primary" }, { "label": "MiniCPM-V 4.7 inference and tool-calling examples", "url": "https://github.com/OpenBMB/MiniCPM-V/blob/main/docs/minicpm_v4dot7_en.md", "type": "primary" }, { "label": "Pinned 35B-A3B chat template", "url": "https://huggingface.co/openbmb/MiniCPM-V-4.7-35B-A3B/blob/ee2d0a272c9d55db52a8b6841b6b24c9d0dda0a3/chat_template.jinja", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openbmb/minicpm-v-4.7-35b-a3b.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/openbmb/MiniCPM-V-4.7-35B-A3B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released October 12, 2026; October 7 repository creation and early initialization commits do not establish earlier public availability. Complete BF16 checkpoint: 35.212875824B parameters, including 34.659889792B language model/output head, 0.522299312B vision tower and 0.030686720B visual projector. OpenBMB reports approximately 3.5B activated; the A3B name and 3B app-table figure refer to language-model shorthand. The active figure does not define embedding or visual-prefill accounting and is not an exact per-token count. Language layers combine 30 GatedDeltaNet and 10 gated full-attention blocks, with 256 routed experts/top 8 plus a shared expert. MTP is not instantiated by the selected inference model; DSpark draft weights are separate. Context is documented/configured at 262144 tokens. Thinking is optional and defaults off in the pinned October 19 template update.", "field_status": {}, "audited_at": "2026-10-27T03:18:52.341072Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/460.json" }, { "id": "lightonai/lightonocr-3-0.8b", "slug": "lightonai--lightonocr-3-0.8b", "name": "LightOnOCR-3 0.8B", "developer": "LightOn", "developer_id": "lightonai", "family": "lightonocr-3", "description": "Vision-language model for page transcription, document layout extraction, image descriptions and chart-to-table conversion.", "release_date": "2026-10-08", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 0.85298592, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 262144, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": false, "tool_call": false, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/lightonai/lightonocr-3-0.8b.toml", "source_type": "catalog", "sources": [ { "label": "LightOnOCR-3-0.8B model card", "url": "https://huggingface.co/lightonai/LightOnOCR-3-0.8B/blob/36d8636925327c671bd8e479c517a004f7566bd9/README.md", "type": "primary" }, { "label": "LightOnOCR-3-0.8B checkpoint configuration", "url": "https://huggingface.co/lightonai/LightOnOCR-3-0.8B/blob/36d8636925327c671bd8e479c517a004f7566bd9/config.json", "type": "primary" }, { "label": "Complete BF16 checkpoint tensor header", "url": "https://huggingface.co/lightonai/LightOnOCR-3-0.8B/resolve/36d8636925327c671bd8e479c517a004f7566bd9/model.safetensors", "type": "weights" }, { "label": "Qwen3.5 implementation (Transformers 5.5.4)", "url": "https://github.com/huggingface/transformers/blob/v5.5.4/src/transformers/models/qwen3_5/modeling_qwen3_5.py", "type": "primary" }, { "label": "LightOnOCR-3 release announcement", "url": "https://huggingface.co/blog/lightonai/lightonocr-3", "type": "primary" }, { "label": "Official OCR serving guidance", "url": "https://github.com/lightonai/LightOnOCR/blob/36755d461be079737860a5f03ae0c803501269e9/README.md", "type": "primary" }, { "label": "Official checkpoint metadata, rechecked October 27", "url": "https://huggingface.co/api/models/lightonai/LightOnOCR-3-0.8B", "type": "primary" }, { "label": "LightOnOCR-3 developer parameter-size plot", "url": "https://lightonai-lightonocr-benchmark-plots.static.hf.space/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/lightonai/lightonocr-3-0.8b.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/lightonai/LightOnOCR-3-0.8B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete checkpoint: 852,985,920 parameters; developer nominal size: 0.8B. The count includes text, vision and projection components, with tied embeddings counted once. Active parameters are not separately specified for this variant. Dense Qwen3.5 vision-language architecture. Context 262,144 is configured positional capacity; official OCR serving guidance uses 18,000 total sequence tokens. OCR training length and long-context quality at 262,144 are unestablished. Thinking is disabled for the trained OCR task; supported prompts are empty or grounding.", "parameter_components_b": { "text": 0.752393024, "vision_encoder": 0.088004352, "vision_projector": 0.012588544 }, "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T13:47:47.771911Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/249.json" }, { "id": "anthropic/claude-haiku-5-5", "slug": "anthropic--claude-haiku-5-5", "name": "Claude Haiku 5.5", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude-haiku", "description": "Fast Claude model for responsive assistance, classification, and lightweight agents", "release_date": "2026-10-07", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1000000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.anthropic.com/claude-haiku-5-5", "source_type": "primary", "sources": [ { "label": "Introducing Claude Haiku 5.5", "url": "https://www.anthropic.com/claude-haiku-5-5", "type": "primary" }, { "label": "Claude Haiku 5.5 overview", "url": "https://platform.claude.com/docs/en/models/haiku-5-5/overview", "type": "primary" }, { "label": "System Card: Claude Haiku 5.5", "url": "https://assets.claude.com/e1080d6bf5ae2018ea3c2f414064be03232f5be5.pdf", "type": "primary" }, { "label": "PDF support", "url": "https://platform.claude.com/docs/en/build-with-claude/pdf-support", "type": "primary" }, { "label": "Claude Haiku 5.5 migration guide", "url": "https://platform.claude.com/docs/en/models/haiku-5-5/migration-guide", "type": "primary" }, { "label": "Commercial Terms of Service", "url": "https://www.anthropic.com/legal/commercial-terms", "type": "primary" }, { "label": "GLM-5.3 and the spread of advanced cyber capabilities", "url": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities", "type": "primary" }, { "label": "Claude Haiku 5.5 System Card — fresh parameter review", "url": "https://www.anthropic.com/claude-haiku-5-5-system-card", "type": "primary" }, { "label": "Claude Haiku 4.5 System Card — predecessor lineage check", "url": "https://assets.anthropic.com/m/99128ddd009bdcb/original/Claude-Haiku-4-5-System-Card.pdf", "type": "primary" }, { "label": "Claude Sonnet 5.5 System Card — sibling comparison check", "url": "https://www.anthropic.com/claude-sonnet-5-5-system-card", "type": "primary" }, { "label": "Claude Opus 5.5 System Card — sibling comparison check", "url": "https://www.anthropic.com/claude-opus-5-5-system-card", "type": "primary" }, { "label": "Claude models overview — current comparison table", "url": "https://platform.claude.com/docs/en/models/overview", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/anthropic/claude-haiku-5-5.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T10:27:28.963805+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/099.json", "notes": "Anthropic has not disclosed nominal total or active parameter counts in the checked sources. The Haiku 4.5 predecessor relationship does not establish a sized base; effort settings describe test-time compute." }, { "id": "google/gemini-nano-banana-2.1", "slug": "google--gemini-nano-banana-2.1", "name": "Gemini nano banana 2.1", "developer": "Google", "developer_id": "google", "family": "gemini", "description": null, "release_date": "2026-10-06", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 131072, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text", "image" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.google.dev/gemini-api/docs/changelog", "source_type": "primary", "sources": [ { "label": "Dated publisher API releases", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "documentation" }, { "label": "Publisher interleaved text and image capability", "url": "https://ai.google.dev/gemini-api/docs/image-generation", "type": "documentation" }, { "label": "Publisher model specifications", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-nano-banana-2.1", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:31:17.277Z", "last_seen": "2026-10-29T10:31:17.277Z", "upstream_present": false, "availability": "public", "notes": "Generates interleaved text and images. Context and capabilities describe current published model documentation; preview limits may have differed." }, { "id": "mistral/mistral-large-4", "slug": "mistral--mistral-large-4", "name": "Mistral Large 4 Preview", "developer": "Mistral", "developer_id": "mistral", "family": "mistral-large", "description": "October 6 public API preview of Mistral Large 4.", "release_date": "2026-10-06", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": 1050, "active_parameters_b": 52, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://mistral.ai/news/mistral-large-4/", "source_type": "primary", "sources": [ { "label": "Mistral Large 4 official release", "url": "https://mistral.ai/news/mistral-large-4/", "type": "primary" }, { "label": "Mistral Large 4 official model card", "url": "https://docs.mistral.ai/models/mistral-large-4-0", "type": "primary" }, { "label": "Mistral Large 4 official Markdown identity and modalities", "url": "https://docs.mistral.ai/models/mistral-large-4-0.md", "type": "primary" }, { "label": "Mistral adjustable reasoning documentation", "url": "https://docs.mistral.ai/studio/conversations/reasoning", "type": "primary" }, { "label": "Mistral October 21 active-count reference", "url": "https://mistral.ai/news/mistral-nvidia-ai-for-industrial-engineering/", "type": "primary" }, { "label": "Mistral-hosted route context limits via OpenRouter", "url": "https://openrouter.ai/api/v1/models/mistralai/mistral-large-4-0-20261006/endpoints", "type": "primary" }, { "label": "Publisher distinguishes later trained revision", "url": "https://mistral.ai/news/mistral-large-4-oss/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/mistral-large-4.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Public API preview released October6,2026. Mistral released an improved checkpoint on October29 after three additional weeks of reinforcement learning; that trained revision is listed separately. No downloadable checkpoint matching the October6 preview was verified. The existing nominal1.05T total and52B active count describe the preview; active52B includes embeddings/output layers. The1.6B vision encoder was separately disclosed without explicit total inclusion. Context1048576 was documented for the Mistral hosted default/ZDR route; nativepreview checkpoint context remains unavailable.", "field_status": { "license": "undisclosed" }, "audited_at": "2026-10-26T23:20:05.403141Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/336.json" }, { "id": "amazon/nova-2.5-sonic", "slug": "amazon--nova-2.5-sonic", "name": "Nova 2.5 Sonic", "developer": "Amazon", "developer_id": "amazon", "family": "nova-sonic", "description": "Conversational speech model with text and audio responses and tool use.", "release_date": "2026-10-05", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 256000, "modalities": { "input": [ "text", "audio" ], "output": [ "text", "audio" ] }, "reasoning": null, "tool_call": true, "source_url": "https://aws.amazon.com/about-aws/whats-new/2026/10/amazon-nova-2.5-Sonic/", "source_type": "primary", "sources": [ { "label": "Dated public launch", "url": "https://aws.amazon.com/about-aws/whats-new/2026/10/amazon-nova-2.5-Sonic/", "type": "primary" }, { "label": "Publisher service card and architecture", "url": "https://docs.aws.amazon.com/ai/responsible-ai/nova-2-5-sonic/overview.html", "type": "primary" }, { "label": "Bedrock model identity and limits", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-2-5-sonic.html", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:13:05.399Z", "last_seen": "2026-10-30T07:13:05.399Z", "upstream_present": false, "availability": "public", "notes": "Conversational speech model with a core transformer language model, speech encoder and renderer. The publisher does not disclose total or active parameters or dense/MoE subtype. Available through Bedrock; no downloadable exactcheckpoint was verified. Toolcalling is explicitly documented; a separate extended-thinking mode is unverified. Context256000 expresses the nominal256K in the launch and modelcard; the general Nova models landing page instead says up to1M, so exact binary capacity remains unresolved." }, { "id": "reflection/beam-1.0-501b-a23b", "slug": "reflection--beam-1.0-501b-a23b", "name": "Beam 1.0 501B A23B", "developer": "Reflection", "developer_id": "reflection", "family": "beam", "description": "Reflection text model for coding and tool use, with a sparse mixture-of-experts architecture and reasoning.", "release_date": "2026-10-05", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 501.3, "active_parameters_b": 23.4, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://reflection.ai/blog/introducing-beam", "source_type": "primary", "sources": [ { "label": "Introducing Beam", "url": "https://reflection.ai/blog/introducing-beam", "type": "blog" }, { "label": "Launch announcement", "url": "https://reflection.ai/blog/beam-is-available", "type": "announcement" }, { "label": "Beam 1.0 501B A23B BF16 model card, pinned revision", "url": "https://huggingface.co/reflection/Beam-1.0-501B-A23B/blob/50a18f8ccf28c75cfdfd9cfe602451c5c81acaef/README.md", "type": "primary" }, { "label": "Beam pinned BF16 configuration", "url": "https://huggingface.co/reflection/Beam-1.0-501B-A23B/blob/50a18f8ccf28c75cfdfd9cfe602451c5c81acaef/config.json", "type": "primary" }, { "label": "Complete Beam checkpoint index and all188 pinned shard headers", "url": "https://huggingface.co/reflection/Beam-1.0-501B-A23B/blob/50a18f8ccf28c75cfdfd9cfe602451c5c81acaef/model.safetensors.index.json", "type": "primary" }, { "label": "Beam native implementation in Transformers5.20.1", "url": "https://github.com/huggingface/transformers/blob/v5.20.1/src/transformers/models/beam/modeling_beam.py", "type": "primary" }, { "label": "Beam native configuration defaults", "url": "https://github.com/huggingface/transformers/blob/v5.20.1/src/transformers/models/beam/configuration_beam.py", "type": "primary" }, { "label": "Beam Apache License2.0", "url": "https://huggingface.co/reflection/Beam-1.0-501B-A23B/blob/50a18f8ccf28c75cfdfd9cfe602451c5c81acaef/LICENSE", "type": "primary" }, { "label": "Beam pinned modified-Harmony chat template", "url": "https://huggingface.co/reflection/Beam-1.0-501B-A23B/blob/50a18f8ccf28c75cfdfd9cfe602451c5c81acaef/chat_template.jinja", "type": "primary" }, { "label": "API docs", "url": "https://developers.reflection.ai/models", "type": "docs" }, { "label": "Reflection reasoning API documentation", "url": "https://developers.reflection.ai/reasoning", "type": "primary" }, { "label": "Beam Safety Technical Report", "url": "https://reflection.ai/beam-safety-report", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/reflection/beam-1.0-501b-a23b.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/reflection/Beam-1.0-501B-A23B", "type": "weights" }, { "label": "Hugging Face (FP8)", "url": "https://huggingface.co/reflection/Beam-1.0-501B-A23B-FP8", "type": "weights" }, { "label": "Hugging Face (NVFP4)", "url": "https://huggingface.co/reflection/Beam-1.0-501B-A23B-NVFP4", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "October 5, 2026 was a limited early-group preview with a waitlist; public weights and API beta followed on October 16. Parameters are the model card’s nominal 501.3B total and 23.4B active figures. All 188 pinned checkpoint headers reconcile to 501,267,612,352 stored elements; excluding 12,544 router-correction buffer values leaves 501,267,599,808 learned-weight elements. F32 router weights are learned parameters; correction buffers are not. Both untied vocabulary matrices are included, and the separate speculative drafter is excluded. The exact active embedding convention is not stated, so 23.4B remains a nominal figure. The supported context is 1,048,576 tokens; 131,072 is the API output cap.", "field_status": {}, "audited_at": "2026-10-27T04:16:02.875028Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/477.json" }, { "id": "aleph-alpha/kolibri-1", "slug": "aleph-alpha--kolibri-1", "name": "Kolibri 1", "developer": "Aleph Alpha", "developer_id": "aleph-alpha", "family": "kolibri", "description": "Aleph Alpha's German and English mixture-of-experts reasoning model (78B total, 3.46B active) for chat, RAG and agents", "release_date": "2026-10-03", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 78.10307456, "active_parameters_b": 3.45757312, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/Aleph-Alpha/Kolibri-1/blob/5bb7075ba22db733e3834b1c71d2245537cc14ed/README.md", "source_type": "primary", "sources": [ { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/aleph-alpha/kolibri-1.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/Aleph-Alpha/Kolibri-1", "type": "weights" }, { "label": "Official model card (pinned revision)", "url": "https://huggingface.co/Aleph-Alpha/Kolibri-1/blob/5bb7075ba22db733e3834b1c71d2245537cc14ed/README.md", "type": "primary" }, { "label": "Official model configuration (pinned revision)", "url": "https://huggingface.co/Aleph-Alpha/Kolibri-1/raw/5bb7075ba22db733e3834b1c71d2245537cc14ed/config.json", "type": "primary" }, { "label": "Official technical report", "url": "https://aleph-alpha.com/downloads/tech-report.pdf", "type": "primary" }, { "label": "Official release announcement", "url": "https://aleph-alpha.com/en/blog/kolibri-has-landed-a-sovereign-open-weight-model/", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Active parameter count excludes embeddings. Context is 262,144 tokens natively and 1,048,576 by validated extrapolation; serving beyond 262,144 requires configuration overrides. This repository is the FP8 Kolibri-1 release. Apache-2.0 applies to the published weights and configuration files.", "field_status": {}, "audited_at": "2026-10-26T13:56:43.764936Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/011.json" }, { "id": "nvidia/nemotron-3.5-super-120b-a12b", "slug": "nvidia--nemotron-3.5-super-120b-a12b", "name": "Nemotron 3.5 Super 120B A12B", "developer": "NVIDIA", "developer_id": "nvidia", "family": "nemotron", "description": "NVIDIA reasoning model served through Perplexity’s Agent API with text-only input and a 1,048,576-token context window.", "release_date": "2026-10", "date_precision": "month", "open_weights": false, "license": null, "parameters_b": 120, "active_parameters_b": 12, "parameters_source": "primary_source", "architecture": "hybrid-ssm-transformer-moe", "context_window": 1048576, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://docs.perplexity.ai/docs/agent-api/models", "source_type": "primary", "sources": [ { "label": "Perplexity Agent API model reference", "url": "https://docs.perplexity.ai/docs/agent-api/models", "type": "primary" }, { "label": "Perplexity API changelog", "url": "https://docs.perplexity.ai/docs/resources/changelog", "type": "primary" }, { "label": "October 19 Perplexity product changelog", "url": "https://www.perplexity.ai/changelog/ultrafast-computer-models-decisions-api-and-connector-controls", "type": "primary" }, { "label": "Upstream record and identity comments", "url": "https://raw.githubusercontent.com/anomalyco/models.dev/dev/models/nvidia/nemotron-3.5-super-120b-a12b.toml", "type": "catalog" }, { "label": "Nemotron 3.5 Super training recipe", "url": "https://docs.nvidia.com/nemotron/latest/nemotron/super35/README.html", "type": "primary" }, { "label": "Super text-only conversion guide", "url": "https://docs.nvidia.com/nemo/megatron-bridge/nightly/models/nemotron/nemotron3.5-super-vl-text-only.html", "type": "primary" }, { "label": "Nemotron 3.5 Super and Lightning technical report, October 21 edition", "url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3.5-Super-and-Lightning-Technical-Report.pdf", "type": "primary" }, { "label": "NVIDIA Nemotron 3.5 Super and Lightning research release", "url": "https://research.nvidia.com/labs/nemotron/Nemotron-3.5-Super-and-Lightning/", "type": "primary" }, { "label": "Perplexity exact Nemotron 3.5 Super hosted model", "url": "https://docs.perplexity.ai/docs/agent-api/models.md", "type": "primary" }, { "label": "NVIDIA Nemotron 3.5 Super public checkpoint inventory", "url": "https://huggingface.co/api/models?author=nvidia&search=Nemotron-3.5-Super&limit=100", "type": "primary" }, { "label": "NVIDIA official released-checkpoint and training guide", "url": "https://raw.githubusercontent.com/NVIDIA-NeMo/Nemotron/main/docs/nemotron/super35/README.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/nvidia/nemotron-3.5-super-120b-a12b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Perplexity’s text-only perplexity/nemotron-3.5-super-120b-a12b offer names NVIDIA Nemotron 3.5 Super. Retain the published nominal language-backbone specification of 120B total and 12B active, supported by the exact endpoint designation and NVIDIA’s separate total/active disclosures and decoder scope. These rounded specifications are not an exact inventory of the hosted service. The served checkpoint and auxiliary modules remain unidentified; the separate VL complete-checkpoint count is not applied here. The published hybrid Mamba-Transformer MoE architecture follows the same model identity. Model license and custom-tool support retain their prior unverified assessments. Primary sources establish October 2026 availability without a verified release day; context and reasoning are explicitly documented. Classified closed for the exact hosted text-only offering; related NVIDIA Super-VL weights remain separate public releases.", "field_status": { "license": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-28T08:02:53.397204Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/387.json", "parameters_scope": "language_backbone", "active_parameters_scope": "language_backbone", "parameter_components_b": { "language_backbone": 120 } }, { "id": "tencent/hy-memory-30b-a3b", "slug": "tencent--hy-memory-30b-a3b", "name": "Hy-Memory-30B-A3B", "developer": "Tencent", "developer_id": "tencent", "family": "hy-memory", "description": "Language model trained to write and update long-term agent memory.", "release_date": "2026-10", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 30, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/Tencent-Hunyuan/Hy-Memory", "source_type": "primary", "sources": [ { "label": "Developer release notes", "url": "https://github.com/Tencent-Hunyuan/Hy-Memory", "type": "primary" }, { "label": "Official model card", "url": "https://huggingface.co/tencent/Hy-Memory-30B-A3B/blob/64a460a94d5498b7316fb02395c72d01c47a7347/README.md", "type": "primary" }, { "label": "Released checkpoint configuration", "url": "https://huggingface.co/tencent/Hy-Memory-30B-A3B/blob/64a460a94d5498b7316fb02395c72d01c47a7347/config.json", "type": "primary" }, { "label": "Checkpoint license", "url": "https://huggingface.co/tencent/Hy-Memory-30B-A3B/blob/64a460a94d5498b7316fb02395c72d01c47a7347/LICENSE", "type": "primary" }, { "label": "Original technical report", "url": "https://arxiv.org/abs/2610.22196", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-29T08:32:56.967Z", "last_seen": "2026-10-29T08:32:56.967Z", "upstream_present": false, "notes": "October release is explicitly documented in the developer repository. An exact launch day was not established from primary sources; the October 15 Hugging Face creation date is not treated as release evidence. Total and active sizes are nominal model-card labels; config specifies 128 experts, eight selected per token and a 262144-token context. This is a released text-generation checkpoint, separate from the memory framework and its desktop application." }, { "id": "inclusionai/ling-3.1-flash", "slug": "inclusionai--ling-3.1-flash", "name": "Ling 3.1 Flash", "developer": "InclusionAI", "developer_id": "inclusionai", "family": "ling", "description": "Open-weight hybrid-linear MoE language model for reasoning, tool use, and long-horizon coding and office tasks.", "release_date": "2026-09-29", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 572.12064544, "active_parameters_b": 25, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/inclusionAI/Ling-3.1-flash", "source_type": "primary", "sources": [ { "label": "Ling-3.1-flash official model card", "url": "https://huggingface.co/inclusionAI/Ling-3.1-flash/blob/e5623f6e4c34e9f2ca40bec867b9210db42ac76c/README.md", "type": "primary" }, { "label": "Ling-3.1-flash checkpoint configuration", "url": "https://huggingface.co/inclusionAI/Ling-3.1-flash/blob/e5623f6e4c34e9f2ca40bec867b9210db42ac76c/config.json", "type": "primary" }, { "label": "Ling-3.1-flash official modeling implementation", "url": "https://huggingface.co/inclusionAI/Ling-3.1-flash/blob/e5623f6e4c34e9f2ca40bec867b9210db42ac76c/modeling_bailing_moe_v3.py", "type": "primary" }, { "label": "Ling-3.1-flash checkpoint index and bounded tensor-header audit", "url": "https://huggingface.co/inclusionAI/Ling-3.1-flash/blob/e5623f6e4c34e9f2ca40bec867b9210db42ac76c/model.safetensors.index.json", "type": "primary" }, { "label": "Ling-3.1-flash official chat template", "url": "https://huggingface.co/inclusionAI/Ling-3.1-flash/blob/e5623f6e4c34e9f2ca40bec867b9210db42ac76c/chat_template.jinja", "type": "primary" }, { "label": "Ant Ling model documentation and exact-model comparison table", "url": "https://developer.ant-ling.com/en/docs/models/ling/", "type": "primary" }, { "label": "Ling-3.1-flash: model upgrade for real-world tasks (September 30, 2026)", "url": "https://developer.ant-ling.com/zh-CN/blogs/ling-3.1-flash", "type": "primary" }, { "label": "Ant Ling official dated publication index", "url": "https://developer.ant-ling.com/zh-CN/blogs/", "type": "primary" }, { "label": "Developer-linked SGLang Ling-3.1-flash deployment cookbook, section 3.4", "url": "https://docs.sglang.io/cookbook/autoregressive/InclusionAI/Ling-3.1-flash", "type": "primary" }, { "label": "Ant Ling announcement returned by X official oEmbed", "url": "https://publish.twitter.com/oembed?url=https%3A%2F%2Ftwitter.com%2FAntLingAGI%2Fstatus%2F2105335205741596911", "type": "primary" }, { "label": "Vercel AI Gateway Ling 3.1 Flash provider page", "url": "https://vercel.com/ai-gateway/models/ling-3.1-flash", "type": "primary" }, { "label": "Vercel AI Gateway public models API", "url": "https://ai-gateway.vercel.sh/v1/models", "type": "primary" }, { "label": "Original automated Vercel catalog synchronization", "url": "https://github.com/anomalyco/models.dev/pull/8384", "type": "catalog" }, { "label": "Vercel Ling 3.1 Flash availability announcement", "url": "https://vercel.com/changelog/ling-3-1-flash-is-now-available-on-ai-gateway", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/inclusionai/ling-3.1-flash.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete checkpoint count is 572.12064544B learned parameters: 563.946448704B main model plus 8.174196736B MTP, excluding quantization scales and routing buffers. Publisher figures of 560B total and 25B active are nominal; 25B active is retained as published. Hybrid-linear MoE uses 63 KDA + 9 gated MLA backbone layers, four dense FFNs, and top-8 of 512 routed experts plus one shared expert; MTP adds one MLA/MoE layer. Maximum 1,048,576-token context requires the documented fourfold YaRN extension; checkpoint default and currently documented official API limit are 262,144. Retain September 29, 2026 as Vercel explicit provider-reported release date, corroborated by contemporaneous catalog sync; Ant Ling and Vercel published launch announcements September 30. No historical inference request was tested. Weight repository creation and open-source announcement followed in October.", "parameter_components_b": { "main_model": 563.946448704, "mtp": 8.174196736 }, "field_status": {}, "audited_at": "2026-10-26T21:08:41.383707Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/246.json" }, { "id": "openai/gpt-6.1-sol", "slug": "openai--gpt-6.1-sol", "name": "GPT-6.1 Sol", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-6.1", "description": "GPT-6.1 reasoning model for coding, computer use and professional work, with text and image input.", "release_date": "2026-09-29", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1050000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://developers.openai.com/api/docs/changelog.md", "source_type": "primary", "sources": [ { "label": "GPT-6.1 Sol API model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-6.1-sol.md", "type": "primary" }, { "label": "OpenAI API changelog", "url": "https://developers.openai.com/api/docs/changelog.md", "type": "primary" }, { "label": "Introducing GPT-6.1 Sol", "url": "https://openai.com/index/introducing-gpt-6-1-sol/", "type": "primary" }, { "label": "Addendum to GPT-6 Astra System Card: GPT-6.1 Sol", "url": "https://deploymentsafety.openai.com/gpt-6-1-sol/gpt-6-1-sol.pdf", "type": "primary" }, { "label": "Addendum to GPT-6 Astra System Card: GPT-6.1 Terra", "url": "https://deploymentsafety.openai.com/gpt-6-1-terra/gpt-6-1-terra.pdf", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "All OpenAI models", "url": "https://developers.openai.com/api/docs/models/all.md", "type": "primary" }, { "label": "Using GPT-6: GPT-6.1 Sol", "url": "https://developers.openai.com/api/docs/guides/latest-model#gpt-61-sol", "type": "primary" }, { "label": "Introducing GPT-6.1 Terra", "url": "https://openai.com/index/introducing-gpt-6-1-terra/", "type": "primary" }, { "label": "GPT-6 Astra Public Summary of Training Content", "url": "https://cdn.openai.com/pdf/gpt-6-astra-eu-ai-act-public-summary-of-training-content.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-6.1-sol.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released September 29, 2026 in API, Work and Codex; Chat was excluded at launch. The only listed API snapshot is the undated gpt-6.1-sol identifier. Context is 1,050,000 tokens, with 922,000 maximum input and 128,000 maximum output. Reasoning supports low, medium (default), high, xhigh and max; tool calling requires Responses. PDF input is converted to text and page images. Total and active parameter counts remain undisclosed in the checked sources; no counted base is identified. Architecture is unverified. Ultrafast was added October 8 as a service tier of the same model.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "unverified" }, "audited_at": "2026-10-27T17:14:18.570316+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/445.json" }, { "id": "anthropic/claude-sonnet-5-5", "slug": "anthropic--claude-sonnet-5-5", "name": "Claude Sonnet 5.5", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude-sonnet", "description": "Claude Sonnet model with adaptive thinking and a 1M-token context window.", "release_date": "2026-09-28", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1000000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.anthropic.com/claude-sonnet-5-5", "source_type": "primary", "sources": [ { "label": "Introducing Claude Sonnet 5.5", "url": "https://www.anthropic.com/claude-sonnet-5-5", "type": "primary" }, { "label": "Claude Sonnet 5.5 - Claude Platform Docs", "url": "https://platform.claude.com/docs/en/models/sonnet-5-5/overview", "type": "primary" }, { "label": "Claude Sonnet 5.5 System Card", "url": "https://www.anthropic.com/claude-sonnet-5-5-system-card", "type": "primary" }, { "label": "GLM-5.3 and the spread of advanced cyber capabilities", "url": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities", "type": "primary" }, { "label": "Commercial Terms of Service", "url": "https://www.anthropic.com/legal/commercial-terms", "type": "primary" }, { "label": "PDF support", "url": "https://platform.claude.com/docs/en/build-with-claude/pdf-support", "type": "primary" }, { "label": "What's new in Claude Sonnet 5.5", "url": "https://platform.claude.com/docs/en/models/sonnet-5-5/whats-new-sonnet-5-5", "type": "primary" }, { "label": "Anthropic Transparency Hub - Claude Sonnet 5.5 summary", "url": "https://www.anthropic.com/transparency", "type": "primary" }, { "label": "Claude Haiku 5.5 System Card", "url": "https://www.anthropic.com/claude-haiku-5-5-system-card", "type": "primary" }, { "label": "Model system cards", "url": "https://www.anthropic.com/system-cards", "type": "primary" }, { "label": "Anthropic Sonnet 5.5 migration and predecessor documentation", "url": "https://raw.githubusercontent.com/anthropics/skills/main/skills/claude-api/shared/model-migration.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/anthropic/claude-sonnet-5-5.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Nominal total and active parameter counts remain undisclosed; no size-bearing base relationship is documented. Architecture remains undisclosed. Proprietary describes nonpublic weights accessed under commercial terms. PDF input is document processing through text and vision. Adaptive thinking is enabled by default, and auto tool choice is supported.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T10:51:32.687970+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/116.json" }, { "id": "minimax/MiniMax-M3.1-Flash-Preview", "slug": "minimax--minimax-m3.1-flash-preview", "name": "MiniMax-M3.1-Flash-Preview", "developer": "MiniMax", "developer_id": "minimax", "family": "minimax", "description": "Preview multimodal MoE model for coding and tool use, with a 1M-token context window and adjustable reasoning effort.", "release_date": "2026-09-27", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 428, "active_parameters_b": 23, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://x.com/MiniMaxAgent/status/2104079819881517400", "source_type": "primary", "sources": [ { "label": "MiniMax Code Preview launch announcement", "url": "https://x.com/MiniMaxAgent/status/2104079819881517400", "type": "primary" }, { "label": "MiniMax Preview Token Plan announcement", "url": "https://x.com/MiniMax_AI/status/2104256406786547800", "type": "primary" }, { "label": "Archived MiniMax M3.1-Flash-Preview product guide, October 11", "url": "https://web.archive.org/web/20261011195320id_/https://agent.minimax.io/tools/m3-1-flash-preview", "type": "primary" }, { "label": "Archived MiniMax model invocation documentation, October 1", "url": "https://web.archive.org/web/20261001082842id_/https://platform.minimax.io/docs/guides/text-generation", "type": "primary" }, { "label": "Archived MiniMax Anthropic compatibility documentation, October 10", "url": "https://web.archive.org/web/20261010202425id_/https://platform.minimax.io/docs/api-reference/text-anthropic-api", "type": "primary" }, { "label": "Official MiniMax M3.1 public repository listing", "url": "https://huggingface.co/api/models?author=MiniMaxAI&search=M3.1&limit=100", "type": "weights" }, { "label": "Later MiniMax M3.1 Flash release card", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M3.1-Flash/blob/262ae6830934f3512211081b8a3f16425b93a239/README.md", "type": "weights" }, { "label": "Current MiniMax M3.1 Flash product page", "url": "https://agent.minimax.io/tools/m3.1-flash", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/minimax/MiniMax-M3.1-Flash-Preview.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Preview-specific nominal 428B total and about 23B active are documented in the archived October 11 developer product FAQ; component counting conventions are unspecified. September 27 launch and 1,000,000 context are primary-sourced. This historical hosted Preview is distinct from the later released Flash checkpoint; its final weights, Community License and detailed architecture are not transferred. Archived docs establish last-documented capabilities, not current endpoint operation.", "field_status": {}, "audited_at": "2026-10-26T22:43:12.814610Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/309.json" }, { "id": "meituan/longcat-2.5-preview", "slug": "meituan--longcat-2.5-preview", "name": "LongCat-2.5-Preview", "developer": "Meituan", "developer_id": "meituan", "family": "longcat", "description": "Meituan's multimodal reasoning model for coding and agent tasks, with image understanding and a 1M-token context window", "release_date": "2026-09-25", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 1600, "active_parameters_b": 48, "parameters_source": "primary_source", "architecture": null, "context_window": 1048576, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://longcat.chat/platform/docs/change-log", "source_type": "primary", "sources": [ { "label": "LongCat-2.5-Preview launch announcement", "url": "https://x.com/Meituan_LongCat/status/2103488918788411728", "type": "primary" }, { "label": "X oEmbed for the official LongCat-2.5-Preview launch", "url": "https://publish.twitter.com/oembed?url=https%3A%2F%2Fx.com%2FMeituan_LongCat%2Fstatus%2F2103488918788411728", "type": "primary" }, { "label": "X syndication data for LongCat-2.5-Preview announcement", "url": "https://cdn.syndication.twimg.com/tweet-result?id=2103488918788411728&lang=en&token=1", "type": "primary" }, { "label": "LongCat API Platform Change Log", "url": "https://longcat.chat/platform/docs/change-log", "type": "primary" }, { "label": "LongCat Model Parameters", "url": "https://longcat.chat/platform/docs/api/model-params", "type": "primary" }, { "label": "LongCat Responses API", "url": "https://longcat.chat/platform/docs/api/responses", "type": "primary" }, { "label": "LongCat Retrieve Model reference", "url": "https://longcat.chat/platform/docs/api/model", "type": "primary" }, { "label": "Official Meituan LongCat public model inventory", "url": "https://huggingface.co/api/models?author=meituan-longcat&limit=100", "type": "primary" }, { "label": "Official Meituan LongCat public repository inventory", "url": "https://api.github.com/orgs/meituan-longcat/repos?per_page=100", "type": "primary" }, { "label": "LongCat AI Platform Terms of Use", "url": "https://longcat.chat/platform/private/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/meituan/longcat-2.5-preview.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Exact LongCat-2.5-Preview API release, September 25, 2026. Meituan launch announcement reports nominal 1.6T total parameters and approximately 48B active; neither figure is a complete learned-tensor count. No public exact-preview checkpoint or internal architecture disclosure was located in checked official repositories, API docs and later research; LongCat-2.0 architecture/MIT licensing are not transferred. API context is 1,048,576 total input-plus-output tokens. The parameter reference gives 262,144 maximum output while Quick Start retains 128K. Image input is explicit in the launch and Responses API, despite a text-only Retrieve Model example. API access is proprietary; optional thinking and tool calls are documented, not live-tested.", "field_status": { "architecture": "undisclosed" }, "audited_at": "2026-10-26T21:16:10.070440Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/253.json" }, { "id": "perceptron/perceptron-mk1.5", "slug": "perceptron--perceptron-mk1.5", "name": "Perceptron Mk1.5", "developer": "Perceptron", "developer_id": "perceptron", "family": "perceptron", "description": "Multimodal model for visual grounding, video tracking and audio understanding, with optional reasoning and function calling.", "release_date": "2026-09-25", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 36864, "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.perceptron.inc/blog/introducing-perceptron-mk1-5", "source_type": "primary", "sources": [ { "label": "Introducing Perceptron Mk1.5", "url": "https://www.perceptron.inc/blog/introducing-perceptron-mk1-5", "type": "primary" }, { "label": "Perceptron Mk1.5 model specifications", "url": "https://docs.perceptron.inc/perceptron-mk1.5/models/perceptron-mk1.5.md", "type": "primary" }, { "label": "Mk1.5 release notes", "url": "https://docs.perceptron.inc/perceptron-mk1.5/changelog.md", "type": "primary" }, { "label": "Mk1.5 tokenization guide", "url": "https://docs.perceptron.inc/perceptron-mk1.5/guides/tokenization.md", "type": "primary" }, { "label": "Mk1.5 reasoning controls", "url": "https://docs.perceptron.inc/perceptron-mk1.5/capabilities/thinking.md", "type": "primary" }, { "label": "Mk1.5 tool calling", "url": "https://docs.perceptron.inc/perceptron-mk1.5/capabilities/tool-calling.md", "type": "primary" }, { "label": "Perceptron Terms of Use", "url": "https://www.perceptron.inc/terms-of-use", "type": "primary" }, { "label": "Perceptron home page", "url": "https://www.perceptron.inc/", "type": "primary" }, { "label": "Isaac 0.5 technical report (distinct model)", "url": "https://pub-d90b81cad7254a1aa6b148ac18153c0c.r2.dev/isaac-0.5.pdf", "type": "primary" }, { "label": "Mk1.5 release: architecture changes and audio", "url": "https://x.com/ArmenAgha/status/2103519189864751221", "type": "primary" }, { "label": "Perceptron model family", "url": "https://www.perceptron.inc/about", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/perceptron/perceptron-mk1.5.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Current first-party API documentation specifies a shared 36,864-token context and 8,192-token output maximum; the launch announcement states nominal 32K. Video soundtracks require opt-in. Spatial and temporal outputs are structured annotations; the application executes function calls. Total and active parameters remain undisclosed: the third-party 35B claim is unverified, and no named base model with a published size is documented. Isaac 0.5’s 36B/2.5B counts describe a separate model. Mk1.5 adds audio and architecture changes; its topology remains unspecified.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T17:34:52.827746Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/465.json" }, { "id": "fireworks/ember-1", "slug": "fireworks--ember-1", "name": "Ember-1", "developer": "Fireworks", "developer_id": "fireworks", "family": "ember", "description": "Fireworks reasoning model based on Kimi K3, trained to shorten reasoning traces for coding and agentic tasks.", "release_date": "2026-09-23", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 2780, "active_parameters_b": 104, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1040000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://fireworks.ai/blog/ember-1", "source_type": "primary", "sources": [ { "label": "Ember-1 API & Playground", "url": "https://fireworks.ai/models/fireworks/ember-1", "type": "primary" }, { "label": "Introducing Ember-1", "url": "https://fireworks.ai/blog/ember-1", "type": "primary" }, { "label": "Vercel AI Gateway live model metadata", "url": "https://ai-gateway.vercel.sh/v1/models", "type": "primary" }, { "label": "Ember-1 (Fireworks) - Cloudflare AI documentation", "url": "https://developers.cloudflare.com/ai/models/fireworks/ember-1/", "type": "primary" }, { "label": "Fireworks Terms of Service, July 10, 2026", "url": "https://fireworks.ai/terms-of-service", "type": "primary" }, { "label": "Kimi K3 official model card", "url": "https://huggingface.co/moonshotai/Kimi-K3", "type": "primary" }, { "label": "Kimi K3: Open Frontier Intelligence", "url": "https://arxiv.org/html/2607.24653v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/fireworks/ember-1.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "2.78T is Fireworks’ nominal total. Nominal 104B active is inferred from its documented post-training of Kimi K3 and the base model’s official specification. Context follows Fireworks’ advertised 1040k tokens (1,040,000), also documented by Cloudflare; Vercel lists 1,048,576 for its route.", "field_status": {}, "audited_at": "2026-10-27T12:38:48.415159+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/186.json" }, { "id": "anthropic/claude-opus-5-5", "slug": "anthropic--claude-opus-5-5", "name": "Claude Opus 5.5", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude-opus", "description": "Claude Opus model with always-on adaptive thinking and a 1M-token context window.", "release_date": "2026-09-22", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1000000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.anthropic.com/claude-opus-5-5", "source_type": "primary", "sources": [ { "label": "Introducing Claude Opus 5.5", "url": "https://www.anthropic.com/claude-opus-5-5", "type": "primary" }, { "label": "Claude Opus 5.5 - Claude Platform Docs", "url": "https://platform.claude.com/docs/en/models/opus-5-5/overview", "type": "primary" }, { "label": "Claude Opus 5.5 System Card", "url": "https://www.anthropic.com/claude-opus-5-5-system-card", "type": "primary" }, { "label": "GLM-5.3 and the spread of advanced cyber capabilities", "url": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities", "type": "primary" }, { "label": "Commercial Terms of Service", "url": "https://www.anthropic.com/legal/commercial-terms", "type": "primary" }, { "label": "PDF support", "url": "https://platform.claude.com/docs/en/build-with-claude/pdf-support", "type": "primary" }, { "label": "Tool use with Claude", "url": "https://platform.claude.com/docs/en/agents-and-tools/tool-use/overview", "type": "primary" }, { "label": "Anthropic Transparency Hub - Claude Opus 5.5 summary", "url": "https://www.anthropic.com/transparency", "type": "primary" }, { "label": "Claude Sonnet 5.5 System Card", "url": "https://www.anthropic.com/claude-sonnet-5-5-system-card", "type": "primary" }, { "label": "Claude Haiku 5.5 System Card", "url": "https://www.anthropic.com/claude-haiku-5-5-system-card", "type": "primary" }, { "label": "Model Migration Guide - Migrating to Claude Opus 5.5", "url": "https://raw.githubusercontent.com/anthropics/skills/main/skills/claude-api/shared/model-migration.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/anthropic/claude-opus-5-5.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Nominal total and active parameter counts are undisclosed; published comparisons with Opus 5 and Fable 5.1 do not establish a size-preserving base relationship. Architecture remains undisclosed. Proprietary describes nonpublic weights accessed under commercial service terms. PDF input is document processing through text and vision.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T10:43:17.420964Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/111.json" }, { "id": "openai/gpt-6-luna", "slug": "openai--gpt-6-luna", "name": "GPT-6 Luna", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-6", "description": "GPT-6 reasoning model for focused, high-volume tasks, with text and image input and tool calling.", "release_date": "2026-09-22", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1050000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-gpt-6-sol-and-luna/", "source_type": "primary", "sources": [ { "label": "Introducing GPT-6 Sol and Luna", "url": "https://openai.com/index/introducing-gpt-6-sol-and-luna/", "type": "primary" }, { "label": "GPT-6 Luna API model reference", "url": "https://developers.openai.com/api/docs/models/gpt-6-luna.md", "type": "primary" }, { "label": "OpenAI API changelog", "url": "https://developers.openai.com/api/docs/changelog.md", "type": "primary" }, { "label": "GPT-6 Astra System Card — September 22 Sol/Luna appendix", "url": "https://deploymentsafety.openai.com/gpt-6-astra/gpt-6-astra.pdf", "type": "primary" }, { "label": "GPT-6.1 Sol System Card", "url": "https://deploymentsafety.openai.com/gpt-6-1-sol/gpt-6-1-sol.pdf", "type": "primary" }, { "label": "GPT-6.1 Terra System Card", "url": "https://deploymentsafety.openai.com/gpt-6-1-terra/gpt-6-1-terra.pdf", "type": "primary" }, { "label": "GPT-6 Sol and GPT-6 Luna: October 2026 update", "url": "https://deploymentsafety.openai.com/gpt-6-october/gpt-6-october.pdf", "type": "primary" }, { "label": "Prompt caching", "url": "https://developers.openai.com/api/docs/guides/prompt-caching.md", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "OpenAI model inventory", "url": "https://developers.openai.com/api/docs/models/all.md", "type": "primary" }, { "label": "OpenAI API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "EU AI Act model training-summary directory", "url": "https://help.openai.com/en/articles/12141645-eu-ai-act-openai-resources-and-customer-guidance", "type": "primary" }, { "label": "A model guide for the GPT-6 family", "url": "https://openai.com/index/practical-guide-building-gpt-6/", "type": "primary" }, { "label": "GPT-6 Astra public summary of training content", "url": "https://cdn.openai.com/pdf/gpt-6-astra-eu-ai-act-public-summary-of-training-content.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-6-luna.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released September 22, 2026 in the API, Work and Codex. The model reference lists only the undated gpt-6-luna identifier, not an immutable dated snapshot. Current API limits are 1,050,000 context, 922,000 input and 128,000 output tokens. Reasoning supports none through max with medium as default. Use Responses for built-in tools; Chat Completions function calling requires reasoning_effort=none. PDF denotes API extraction into text/page images. September 25 fixed image encoding; the October ChatGPT versions are separately identified in their system card and are not assumed to replace the production API model. Decisions API support was added October 6 and became generally available October 22. Total and active parameter counts remain undisclosed; shared training methods with Astra do not establish a counted base. Architecture remains unverified. No Luna shutdown date was found; endpoint operation was not tested.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "unverified" }, "audited_at": "2026-10-27T17:11:19.522278Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/443.json" }, { "id": "openai/gpt-6-sol", "slug": "openai--gpt-6-sol", "name": "GPT-6 Sol", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-6", "description": "OpenAI model for complex coding and agentic workflows", "release_date": "2026-09-22", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1050000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-gpt-6-sol-and-luna/", "source_type": "primary", "sources": [ { "label": "Introducing GPT-6 Sol and Luna", "url": "https://openai.com/index/introducing-gpt-6-sol-and-luna/", "type": "primary" }, { "label": "GPT-6 Sol API model reference", "url": "https://developers.openai.com/api/docs/models/gpt-6-sol.md", "type": "primary" }, { "label": "OpenAI API changelog", "url": "https://developers.openai.com/api/docs/changelog.md", "type": "primary" }, { "label": "GPT-6 Astra System Card — Appendix A: GPT-6 Sol, GPT-6 Luna", "url": "https://deploymentsafety.openai.com/gpt-6-astra/gpt-6-astra.pdf", "type": "primary" }, { "label": "Addendum to GPT-6 Astra System Card: GPT-6.1 Sol", "url": "https://deploymentsafety.openai.com/gpt-6-1-sol/gpt-6-1-sol.pdf", "type": "primary" }, { "label": "Addendum to GPT-6 Astra System Card: GPT-6.1 Terra", "url": "https://deploymentsafety.openai.com/gpt-6-1-terra/gpt-6-1-terra.pdf", "type": "primary" }, { "label": "GPT-6 Sol and GPT-6 Luna: October 2026 update", "url": "https://deploymentsafety.openai.com/gpt-6-october/gpt-6-october.pdf", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "Prompt caching", "url": "https://developers.openai.com/api/docs/guides/prompt-caching.md", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201", "type": "primary" }, { "label": "OpenAI API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "OpenAI EU AI Act resources", "url": "https://help.openai.com/en/articles/12141645-eu-ai-act-openai-resources-and-customer-guidance", "type": "primary" }, { "label": "Public Summary of Training Content for GPT-6 Astra", "url": "https://cdn.openai.com/pdf/gpt-6-astra-eu-ai-act-public-summary-of-training-content.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-6-sol.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Public/API release was September 22, 2026. The API reference lists only the undated gpt-6-sol identifier: 1,050,000 context, 922,000 maximum input and 128,000 output. Reasoning defaults to medium and supports none through max. Responses supports reasoning with tools; Chat Completions function calling requires reasoning_effort=none. PDF is converted to text and page images. September 25 corrected image encoding; October 7 introduced a distinct ChatGPT version, while Work and Codex retained September versions. Those events do not establish an identical or replacement production API checkpoint. Nominal total and active parameters remain undisclosed; shared training methods with Astra do not establish a counted base, and later exact Sol comparisons report performance rather than parameter counts. Production architecture remains unverified. No exact Sol API shutdown notice was found; endpoint operation was not tested.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "unverified" }, "audited_at": "2026-10-27T17:12:44.650962Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/444.json" }, { "id": "vivgrid/viv-fast", "slug": "vivgrid--viv-fast", "name": "Viv Fast", "developer": "Vivgrid", "developer_id": "vivgrid", "family": null, "description": "Vivgrid model for coding and general-purpose agent tasks, accepting text and images with reasoning controls and function calling.", "release_date": "2026-09-21", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": null, "context_window": 1000000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/commit/ba1ee716d838a37b985a48b6805e7237e88fe755", "source_type": "primary", "sources": [ { "label": "viv-fast model documentation", "url": "https://vivgrid.com/docs/models/viv-fast", "type": "primary" }, { "label": "Initial Viv Fast contribution, September 21, 2026", "url": "https://github.com/anomalyco/models.dev/pull/7626", "type": "primary" }, { "label": "Initial Viv Fast model and provider specifications", "url": "https://github.com/anomalyco/models.dev/commit/ba1ee716d838a37b985a48b6805e7237e88fe755", "type": "primary" }, { "label": "C.C. author identity", "url": "https://github.com/fanweixiao", "type": "primary" }, { "label": "Vivgrid service terms, Scope of Service and Restrictions", "url": "https://vivgrid.com/terms", "type": "primary" }, { "label": "Vivgrid current and earlier model comparison", "url": "https://vivgrid.com/docs/models", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/vivgrid/viv-fast.toml", "type": "catalog" }, { "label": "Model docs", "url": "https://docs.vivgrid.com/models/viv-fast", "type": "docs" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The September 21, 2026 release date and low/high/max reasoning controls come from Vivgrid founder C.C.’s initial public contribution. Official documentation confirms 1,000,000 context tokens, text/image input and function calling. Total and active parameter counts and network architecture remain undisclosed; no named base model supports a size inference. The SLM label and regional serving do not establish model size.", "field_status": { "family": "undisclosed", "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T18:19:20.116213+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/522.json" }, { "id": "xai/grok-4.7", "slug": "xai--grok-4.7", "name": "Grok 4.7", "developer": "xAI", "developer_id": "xai", "family": "grok", "description": "Grok reasoning model for coding, agentic tasks and knowledge work, with text and image input, text output and configurable reasoning effort.", "release_date": "2026-09-21", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 2100, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": null, "context_window": 500000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://x.ai/news/grok-4-7", "source_type": "primary", "sources": [ { "label": "Introducing Grok 4.7", "url": "https://x.ai/news/grok-4-7", "type": "primary" }, { "label": "Grok 4.7 model specification", "url": "https://docs.x.ai/developers/models/grok-4.7", "type": "primary" }, { "label": "Grok 4.7 Model Card, revision 2026-09-21", "url": "https://media.x.ai/v1/website/card4p7-3a96f40b.pdf", "type": "primary" }, { "label": "Grok 4.7 Model Card, earlier public file", "url": "https://media.x.ai/v1/website/4p7card-5eccc980.pdf", "type": "primary" }, { "label": "Elon Musk original Grok 4.6 / Grok 4.7 scale post, July 28, 2026", "url": "https://cdn.syndication.twimg.com/tweet-result?id=2082123925283041545&lang=en&token=1", "type": "primary" }, { "label": "API release notes: September 21, 2026", "url": "https://docs.x.ai/developers/release-notes", "type": "primary" }, { "label": "Reasoning: Grok 4.7 controls", "url": "https://docs.x.ai/developers/model-capabilities/text/reasoning", "type": "primary" }, { "label": "Files Overview", "url": "https://docs.x.ai/developers/files", "type": "primary" }, { "label": "Enterprise Terms of Service, updated August 14, 2026", "url": "https://x.ai/legal/terms-of-service-enterprise", "type": "primary" }, { "label": "Cursor Grok 4.7 model and training overview", "url": "https://cursor.com/grok", "type": "primary" }, { "label": "Grok 4.7: Cursor model help", "url": "https://prod.cursor.com/help/models-and-usage/grok-4-7", "type": "primary" }, { "label": "Elon Musk Grok 4.7 launch-day post, September 21, 2026", "url": "https://cdn.syndication.twimg.com/tweet-result?id=2102071804495872374&lang=en&token=1", "type": "primary" }, { "label": "Elon Musk Grok 4.7 training update, August 12, 2026", "url": "https://cdn.syndication.twimg.com/tweet-result?id=2087604711767896527&lang=en&token=1", "type": "primary" }, { "label": "Grok 4.7: current Cursor model reference", "url": "https://prod.cursor.com/docs/models/grok-4-7", "type": "primary" }, { "label": "Introducing Grok 4.7: Cursor release post", "url": "https://cursor.com/blog/grok-4-7", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/xai/grok-4.7.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released on the xAI API September 21, 2026. Text and image inputs are native; PDF attachments use search. The nominal 2.1T size comes from Elon Musk's July 28 pre-release announcement for Grok 4.7. Released model cards do not independently confirm the final count, and no superseding figure was found. Active parameters and network architecture remain undisclosed. The new base and Grok 4.6-generated training data do not establish shared model size. Reasoning cannot be disabled and supports low, medium, high and xhigh. Grok 4.7 Fast serves the same model faster. Cursor's standard 256k product tier does not replace the native 500k context maximum.", "field_status": { "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T07:20:33.504Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/534.json" }, { "id": "xiaomi/mimo-v2.6-flash", "slug": "xiaomi--mimo-v2.6-flash", "name": "MiMo-V2.6-Flash", "developer": "Xiaomi", "developer_id": "xiaomi", "family": "mimo", "description": "Multimodal MoE model with hybrid sliding-window attention, a 1M-token context and reasoning and tool-calling support.", "release_date": "2026-09-21", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 310, "active_parameters_b": 15, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "audio", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://vercel.com/changelog/mimo-v2-6-models-now-available-on-ai-gateway", "source_type": "primary", "sources": [ { "label": "Introducing MiMo-V2.6 series — September 22, 2026", "url": "https://mimo.xiaomi.com/mimo-v2-6/article", "type": "primary" }, { "label": "MiMo-V2.6 technical report — architecture section and Table 1", "url": "https://arxiv.org/html/2610.11959v1", "type": "primary" }, { "label": "MiMo-V2.6-Flash-RL model card at inspected revision", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL/blob/dcd3fd83fb181622fbc606fbb1709901982a62c2/README.md", "type": "weights" }, { "label": "MiMo-V2.6-Flash-RL configuration", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL/blob/dcd3fd83fb181622fbc606fbb1709901982a62c2/config.json", "type": "weights" }, { "label": "Separate DFlash draft configuration", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL/blob/dcd3fd83fb181622fbc606fbb1709901982a62c2/dflash/config.json", "type": "weights" }, { "label": "MiMo-V2.6-Flash-RL chat template", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL/blob/dcd3fd83fb181622fbc606fbb1709901982a62c2/chat_template.jinja", "type": "weights" }, { "label": "Official RL checkpoint repository inventory", "url": "https://huggingface.co/api/models/XiaomiMiMo/MiMo-V2.6-Flash-RL", "type": "weights" }, { "label": "Official model release log", "url": "https://mimo.mi.com/docs/en-US/updates/model", "type": "primary" }, { "label": "Diagnosing and Mitigating Tool-Call Repetition in MiMo-V2.6 — September 27, 2026", "url": "https://mimo.xiaomi.com/blog/mimo-v2-6-tool-call-repetition", "type": "primary" }, { "label": "MiMo-V2.6-Flash-MOPD model card", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-MOPD/blob/b12f5573350bb1704513fcf119fca48e498b0423/README.md", "type": "weights" }, { "label": "Native Xiaomi Chat Completions API reference", "url": "https://mimo.mi.com/docs/en-US/api/chat/openai-api", "type": "primary" }, { "label": "MiMo V2.6 models now available on AI Gateway", "url": "https://vercel.com/changelog/mimo-v2-6-models-now-available-on-ai-gateway", "type": "primary" }, { "label": "MiMo Studio V3.4.0 release log", "url": "https://mimo.mi.com/docs/en-US/updates/feature/studio", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/xiaomi/mimo-v2.6-flash.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The technical report gives nominal 310B main-block parameters and 15B active; the model card says 309B total, a 1B nominal discrepancy that is not independently reconciled. Vision (0.681B), audio tokenizer encoder (0.308B) and audio patch encoder (0.127B) are separate; these encoder counts exclude projectors, and the audio tokenizer count excludes EMA codebooks. The five-layer DFlash speculative decoder is also separate, so no exact complete-checkpoint sum or active count for all phases is asserted. Flash was publicly available September21,2026 according to Vercel and Xiaomi Studio; the Xiaomi research/open-sourcing announcement followed September22; Xiaomi replaced the API with Flash-MOPD on September 25 at 06:00 UTC+8 under the unchanged mimo-v2.6-flash name. Both released checkpoint stages are designated MIT.", "parameters_scope": "language_backbone", "parameter_components_b": { "language_backbone": 310, "vision_encoder": 0.681, "audio_tokenizer_encoder": 0.308, "audio_patch_encoder": 0.127 }, "active_parameters_scope": "language_decode", "active_parameters_by_phase_b": { "language_decode": 15 }, "field_status": {}, "audited_at": "2026-10-27T05:54:47.506828Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/543.json" }, { "id": "xiaomi/mimo-v2.6-pro", "slug": "xiaomi--mimo-v2.6-pro", "name": "MiMo-V2.6-Pro", "developer": "Xiaomi", "developer_id": "xiaomi", "family": "mimo", "description": "Multimodal MoE model with text, image, audio and video input, text output, reasoning and tool calling.", "release_date": "2026-09-21", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 1020, "active_parameters_b": 42, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "audio", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://vercel.com/changelog/mimo-v2-6-models-now-available-on-ai-gateway", "source_type": "primary", "sources": [ { "label": "Introducing MiMo-V2.6 series", "url": "https://mimo.xiaomi.com/mimo-v2-6/article", "type": "primary" }, { "label": "MiMo V2.6 models now available on AI Gateway", "url": "https://vercel.com/changelog/mimo-v2-6-models-now-available-on-ai-gateway", "type": "primary" }, { "label": "Xiaomi MiMo Studio changelog", "url": "https://mimo.mi.com/docs/en-US/updates/feature/studio", "type": "primary" }, { "label": "MiMo-V2.6: Scaling Reinforcement Learning Towards Self-Improvement", "url": "https://arxiv.org/html/2610.11959v1", "type": "primary" }, { "label": "MiMo-V2.6-Pro-RL official model card", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL/blob/b61f21852680ee02a29e946e18cf8697f4844061/README.md", "type": "weights" }, { "label": "MiMo-V2.6-Pro-MOPD official model card", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-MOPD/blob/6a7f202e8dbd0e57b3fe1da03917d08d75abc4b8/README.md", "type": "weights" }, { "label": "MiMo-V2.6-Pro-MOPD checkpoint configuration", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-MOPD/blob/6a7f202e8dbd0e57b3fe1da03917d08d75abc4b8/config.json", "type": "weights" }, { "label": "MiMo-V2.6-Pro-MOPD chat template", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-MOPD/blob/6a7f202e8dbd0e57b3fe1da03917d08d75abc4b8/chat_template.jinja", "type": "weights" }, { "label": "MiMo-V2.6-Pro-MOPD model implementation", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-MOPD/blob/6a7f202e8dbd0e57b3fe1da03917d08d75abc4b8/modeling_mimo_v2.py", "type": "weights" }, { "label": "MiMo-V2.6-Pro-MOPD tensor index", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-MOPD/blob/6a7f202e8dbd0e57b3fe1da03917d08d75abc4b8/model.safetensors.index.json", "type": "weights" }, { "label": "MiMo-V2.6-Pro-MOPD DFlash configuration", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-MOPD/blob/6a7f202e8dbd0e57b3fe1da03917d08d75abc4b8/dflash/config.json", "type": "weights" }, { "label": "Diagnosing and Mitigating Tool-Call Repetition in MiMo-V2.6", "url": "https://mimo.xiaomi.com/blog/mimo-v2-6-tool-call-repetition", "type": "primary" }, { "label": "MiMo-V2.6-Pro native model specifications", "url": "https://mimo.mi.com/models/en-US/mimo-v2.6-pro", "type": "primary" }, { "label": "Xiaomi model release changelog", "url": "https://mimo.mi.com/docs/en-US/updates/model", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/xiaomi/mimo-v2.6-pro.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Public availability is documented on September 21, 2026 by Vercel and Xiaomi Studio; Xiaomi research and model launch pages are dated September 22. Nominal 1.02T/42B counts refer to the language backbone. Separately reported encoders: 681M vision, 308M audio tokenizer and 127M audio patch; these exclude projectors and the tokenizer EMA codebooks. MTP and five-layer DFlash accounting are not reconciled into an exact total. Pro-RL and Pro-MOPD weights are MIT; the native API changed to MOPD on September 25 at 06:00 UTC+8 without changing its model name. Native outputs are text; media-generation demos use external tools.", "parameters_scope": "language_backbone", "parameter_components_b": { "language_backbone": 1020, "vision_encoder": 0.681, "audio_tokenizer_encoder": 0.308, "audio_patch_encoder": 0.127 }, "active_parameters_scope": "language_decode", "active_parameters_by_phase_b": { "language_decode": 42 }, "field_status": {}, "audited_at": "2026-10-27T05:57:57.091601Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/544.json" }, { "id": "stepfun/step-5-preview", "slug": "stepfun--step-5-preview", "name": "Step 5 Preview", "developer": "StepFun", "developer_id": "stepfun", "family": "step", "description": "Multimodal sparse MoE model for coding and tool-based tasks, accepting text, images and video and producing text.", "release_date": "2026-09-20", "date_precision": "day", "open_weights": true, "license": "StepFun Model License 1.0", "parameters_b": 602.972894848, "active_parameters_b": 27, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.stepfun.com/news/step-5-preview", "source_type": "primary", "sources": [ { "label": "Dated Step 5 Preview release announcement", "url": "https://www.stepfun.com/news/step-5-preview", "type": "primary" }, { "label": "Exact Step 5 Preview official model card", "url": "https://huggingface.co/stepfun-ai/Step-5-Preview/blob/364b3b03fed2477aa0953fe855a2c4c8c48ee3cc/README.md", "type": "primary" }, { "label": "Current Step 5 Preview API guide", "url": "https://platform.stepfun.ai/docs/en/guides/models/step-5-preview", "type": "primary" }, { "label": "Inside Step 5 Preview: Architecture and Training for Agentic Work, October 15, 2026", "url": "https://platform.stepfun.com/blog/inside-step-5-preview", "type": "primary" }, { "label": "Pinned released BF16 Step 5 Preview configuration", "url": "https://huggingface.co/stepfun-ai/Step-5-Preview-BF16/blob/7734b7ffac118d70d8398bfa3eb0fab007eb45b1/config.json", "type": "primary" }, { "label": "StepFun Model License 1.0 for the released checkpoint", "url": "https://huggingface.co/stepfun-ai/Step-5-Preview-BF16/blob/7734b7ffac118d70d8398bfa3eb0fab007eb45b1/LICENSE.txt", "type": "primary" }, { "label": "Complete BF16 index and bounded tensor headers", "url": "https://huggingface.co/stepfun-ai/Step-5-Preview-BF16/blob/7734b7ffac118d70d8398bfa3eb0fab007eb45b1/model.safetensors.index.json", "type": "primary" }, { "label": "Exact Step 5 standalone text and vision implementation", "url": "https://huggingface.co/stepfun-ai/Step-5-Preview-BF16/blob/7734b7ffac118d70d8398bfa3eb0fab007eb45b1/inference/model.py", "type": "primary" }, { "label": "Exact Step 5 vision encoder implementation", "url": "https://huggingface.co/stepfun-ai/Step-5-Preview-BF16/blob/7734b7ffac118d70d8398bfa3eb0fab007eb45b1/inference/vision_encoder.py", "type": "primary" }, { "label": "StepFun October 15 public weight release announcement", "url": "https://cdn.syndication.twimg.com/tweet-result?id=2110717330657616092&lang=en&token=0", "type": "primary" }, { "label": "Pinned Step 5 decoder and weight-loading implementation", "url": "https://github.com/vllm-project/vllm/blob/5d35ce3934a7804b298559d2d99eec33792e1eb9/vllm/models/step5/nvidia/model.py", "type": "primary" }, { "label": "Pinned Step 5 MTP ownership and loading rules", "url": "https://github.com/vllm-project/vllm/blob/5d35ce3934a7804b298559d2d99eec33792e1eb9/vllm/models/step5/nvidia/mtp.py", "type": "primary" }, { "label": "Pinned Step 5 configuration defaults", "url": "https://github.com/vllm-project/vllm/blob/5d35ce3934a7804b298559d2d99eec33792e1eb9/vllm/transformers_utils/configs/step5.py", "type": "primary" }, { "label": "Pinned Step 5 multimodal wrapper", "url": "https://github.com/vllm-project/vllm/blob/5d35ce3934a7804b298559d2d99eec33792e1eb9/vllm/models/step5/nvidia/vl.py", "type": "primary" }, { "label": "Pinned exact Step 5 training bridge in Miles PR 4105", "url": "https://github.com/radixark/miles/blob/266bec4356f79711a493827169288d8e3952832f/miles_plugins/mbridge/step5.py", "type": "primary" }, { "label": "Pinned Megatron expert-bias buffer implementation", "url": "https://github.com/NVIDIA/Megatron-LM/blob/63f687dc83f44b31346e3775a5e23cf2337f7cbd/megatron/core/transformer/moe/router.py", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/stepfun/step-5-preview.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/stepfun-ai/Step-5-Preview", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete released checkpoint: 602972894848 learned parameters, including three MTP branches, visual encoder and projector. The 2449 unique indexed tensor shapes reconcile; 30976 auxiliary routing-bias values, 1472 unused ssmax_s values and unindexed duplicate storage are excluded. StepFun advertises nominal 600B total and 27B active per token; the active figure is not an exact per-phase count. Public API access began September 20, 2026; official weights followed October 15 under StepFun Model License 1.0. The checkpoint supports 1048576 tokens; API prose advertises nominal 1M. Image/video input produces text output.", "parameter_components_b": { "language_backbone": 598.476021504, "mtp": 2.51585856, "vision_encoder_and_downsamplers": 1.95584896, "vision_projector": 0.025165824 }, "field_status": {}, "audited_at": "2026-10-27T04:47:19.337370Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/497.json" }, { "id": "alibaba/qwen3.8-omni-flash", "slug": "alibaba--qwen3.8-omni-flash", "name": "Qwen3.8 Omni Flash", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Qwen omni model for text, vision, audio, and multimodal agent tasks", "release_date": "2026-09-17", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": 125, "active_parameters_b": 6, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text", "image", "audio", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://help.aliyun.com/en/model-studio/qwen3-8-omni-flash", "source_type": "primary", "sources": [ { "label": "Qwen3.8-Omni technical report", "url": "https://arxiv.org/html/2609.25611v1", "type": "primary" }, { "label": "Alibaba Cloud model documentation", "url": "https://help.aliyun.com/en/model-studio/qwen3-8-omni-flash", "type": "primary" }, { "label": "Alibaba Cloud model release tables", "url": "https://help.aliyun.com/en/model-studio/newly-released-models", "type": "primary" }, { "label": "QwenCloud model release changelog", "url": "https://docs.qwencloud.com/changelog/models", "type": "primary" }, { "label": "Qwen-Omni developer guide", "url": "https://docs.qwencloud.com/developer-guides/speech/multimodal-speech", "type": "primary" }, { "label": "Qwen3.8-Flash-Next named base specifications", "url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next", "type": "primary" }, { "label": "Qwen3.8-Next architecture and parameter table", "url": "https://arxiv.org/html/2608.30320v1", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=qwen3.8-omni-flash&limit=100", "label": "Official Qwen inventory search for the exact hosted model", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=Qwen3.8&limit=1000", "label": "Official Qwen Qwen3.8 weight inventory", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.8-omni-flash.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Nominal 125B main-language parameters and 6B active are inferred from the technical report's explicit Qwen3.8-Next initialization. Both counts cover the main language model only; they exclude encoders and the base's separately counted n-gram/MTP components. Full Omni total and full multimodal active count remain undisclosed. This endpoint outputs text; Realtime handles speech generation. Earliest official release-table date is September 17, 2026; the launch article shows September 18. No public weights for the complete Omni-Flash model were found; its weight license remains undisclosed. Classified as closed weights: no public release of this hosted variant was found in the developer documentation and weight inventory.", "field_status": { "license": "unverified" }, "audited_at": "2026-10-28T08:01:10.207758+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/074.json", "parameters_scope": "main_language_model", "active_parameters_scope": "main_language_model", "parameter_components_b": { "main_language_model": 125 } }, { "id": "google/gemini-3.8-live", "slug": "google--gemini-3.8-live", "name": "Gemini 3.8 live", "developer": "Google", "developer_id": "google", "family": "gemini", "description": null, "release_date": "2026-09-15", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 131072, "modalities": { "input": [ "text", "image", "audio", "video" ], "output": [ "text", "audio" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ai.google.dev/gemini-api/docs/changelog", "source_type": "primary", "sources": [ { "label": "Dated publisher API releases", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "documentation" }, { "label": "Publisher model capability table", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.8-live", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:31:17.277Z", "last_seen": "2026-10-29T10:31:17.277Z", "upstream_present": false, "availability": "public", "notes": "Conversational audio model. Text can be returned as the generated speech transcript; text-only response mode is not implied." }, { "id": "upstage/solar-mini4", "slug": "upstage--solar-mini4", "name": "Solar Mini 4", "developer": "Upstage", "developer_id": "upstage", "family": "solar-mini", "description": "Upstage text model with a mixture-of-experts architecture, optional reasoning, structured outputs and tool calling.", "release_date": "2026-09-13", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 35, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 524288, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.aitimes.com/news/articleView.html?idxno=215245", "source_type": "secondary", "sources": [ { "label": "Solar Mini 4 model documentation", "url": "https://console.upstage.ai/docs/models/solar-mini-4", "type": "primary" }, { "label": "Solar Mini 4: Built for High-Volume Agent Workloads", "url": "https://www.upstage.ai/blog/en/solar-mini-4", "type": "primary" }, { "label": "Solar model history", "url": "https://console.upstage.ai/docs/models/history", "type": "primary" }, { "label": "Reasoning capability documentation", "url": "https://console.upstage.ai/docs/capabilities/generate/reasoning", "type": "primary" }, { "label": "Responses API reference", "url": "https://console.upstage.ai/api/responses", "type": "primary" }, { "label": "Upstage Solar Mini 4 provider endpoints", "url": "https://openrouter.ai/api/v1/models/upstage/solar-mini4/endpoints", "type": "primary" }, { "label": "Upstage terms of service", "url": "https://www.upstage.ai/terms-of-service", "type": "primary" }, { "label": "Upstage opens free Solar Mini 4 preview", "url": "https://www.aitimes.com/news/articleView.html?idxno=215245", "type": "secondary" }, { "label": "External evaluation of Solar Mini 4 Preview", "url": "https://www.aitimes.com/news/articleView.html?idxno=215457", "type": "secondary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/upstage/solar-mini4.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "35B total and 3B active per token are nominal Upstage-reported counts, not exact checkpoint counts. September 13, 2026 is the free preview opening reported by AI Times from the CEO announcement; access required application and approval, and the original Facebook post was not located. The dated solar-mini4-260922 API version became generally available September 22. Technical specifications describe that documented release: MoE, 512K context (524288 on the Upstage provider endpoint), optional reasoning and text input/output. Artificial Analysis lists larger context/output limits that conflict with Upstage and are not adopted. Proprietary describes the service terms; commercial on-premises deployment does not establish public weights.", "field_status": {}, "audited_at": "2026-10-27T05:10:24.454172Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/515.json" }, { "id": "moonshotai/kimi-k2.8-preview", "slug": "moonshotai--kimi-k2.8-preview", "name": "Kimi K2.8 Preview", "developer": "Moonshot AI", "developer_id": "moonshotai", "family": "kimi-k2", "description": "Moonshot AI's Kimi Code preview model for coding and agent tasks, served as kimi-for-coding with adjustable reasoning, image/video input and a 1,048,576-token context window.", "release_date": "2026-09-11", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1048576, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.kimi.com/code/docs/en/kimi-code/whats-new.html", "source_type": "primary", "sources": [ { "label": "Kimi Code K2.8 Preview release", "url": "https://www.kimi.com/code/docs/en/kimi-code/whats-new.html", "type": "primary" }, { "label": "Kimi Code model configuration", "url": "https://www.kimi.com/code/docs/en/kimi-code/models.html", "type": "primary" }, { "label": "Kimi service terms", "url": "https://www.kimi.com/user/agreement/modelUse?version=v2", "type": "primary" }, { "label": "Official Moonshot AI model inventory", "url": "https://huggingface.co/api/models?author=moonshotai&limit=100&full=false", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/moonshotai/kimi-k2.8-preview.toml", "type": "catalog" }, { "label": "Kimi-K3 official model card", "url": "https://huggingface.co/moonshotai/Kimi-K3", "type": "primary" }, { "label": "Kimi-K2.7-Code official model card", "url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "type": "primary" }, { "label": "Introducing Kimi K3.1 Pro", "url": "https://www.kimi.com/en/blog/kimi-k3-1-pro", "type": "primary" }, { "label": "Kimi Code Chinese release history — K2.8 Preview and later updates", "url": "https://www.kimi.com/code/docs/kimi-code/whats-new.html", "type": "primary" }, { "label": "Kimi Research Blog — current research index", "url": "https://www.kimi.com/en/blog/", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "K2.8 Preview is the Kimi Code model behind kimi-for-coding. September 11, 2026 is the explicit full-rollout date and earliest verified release date in this audit. The October 27 parameter re-audit reopened English/Chinese release notes, the current model table, official model inventory, sibling model cards and the newer K3.1 Pro technical blog; targeted English/Chinese searches and the current research index located no K2.8 nominal size or explicit named-base/post-training lineage. Total and active parameters therefore remain undisclosed. K2.7 Code explicitly reports 1T/32B and K3 reports 2.8T/104B, but neither card maps its architecture to K2.8. The unchanged service ID and qualitative comparisons do not establish that experts, attention or vision structure were preserved, so architecture also remains undisclosed. Supported nominal counts or documented lineage would suffice; an exact checkpoint census is not required. Reasoning is optional and adjustable. No K2.8 weights were present in the fresh 19-repository official inventory; sibling weight licenses do not transfer.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T07:14:03.668Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/363.json" }, { "id": "deepseek/deepseek-v4.1-flash", "slug": "deepseek--deepseek-v4.1-flash", "name": "DeepSeek V4.1 Flash", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-flash", "description": "DeepSeek V4.1 Flash model for reasoning and agentic coding", "release_date": "2026-09-10", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 552, "active_parameters_b": 16, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.deepseek.com/en/news/deepseek-v4-1-flash/", "source_type": "primary", "sources": [ { "label": "DeepSeek V4.1 Flash release", "url": "https://www.deepseek.com/en/news/deepseek-v4-1-flash/", "type": "announcement" }, { "label": "Hugging Face", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash", "type": "weights" }, { "label": "DeepSeek V4.1 Flash technical report", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash/resolve/main/DeepSeek_V41_Tech_Report.pdf", "type": "technical_report" }, { "label": "DeepSeek V4.1 Flash configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash/raw/main/config.json", "type": "config" }, { "label": "DeepSeek API change log", "url": "https://api-docs.deepseek.com/updates/", "type": "documentation" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepseek/deepseek-v4.1-flash.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "552B is the nominal backbone count; an additional 196B Engram memory is separately reported. A complete checkpoint total is not established here. Active parameters are 8B for prefill and 16B for decode; the scalar records decode. Original September 10 checkpoint; the live deepseek-flash API was upgraded to V4.1-Flash-1017 on October 17.", "parameters_scope": "backbone", "active_parameters_scope": "decode", "parameter_components_b": { "backbone": 552, "engram": 196 }, "active_parameters_by_phase_b": { "prefill": 8, "decode": 16 }, "field_status": {}, "audited_at": "2026-10-26T13:45:18Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/001.json" }, { "id": "openai/gpt-live-1", "slug": "openai--gpt-live-1", "name": "GPT-live-1", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-live", "description": null, "release_date": "2026-09-10", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text", "audio" ], "output": [ "text", "audio" ] }, "reasoning": null, "tool_call": null, "source_url": "https://developers.openai.com/api/docs/changelog", "source_type": "primary", "sources": [ { "label": "Dated publisher API releases", "url": "https://developers.openai.com/api/docs/changelog", "type": "documentation" }, { "label": "Publisher model documentation and snapshot mapping", "url": "https://developers.openai.com/api/docs/models/gpt-live-1", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:31:17.277Z", "last_seen": "2026-10-29T10:31:17.277Z", "upstream_present": false, "availability": "public", "notes": "Original publisher conversational audio model. " }, { "id": "cohere/north-small-translate-1-0", "slug": "cohere--north-small-translate-1-0", "name": "North Small Translate", "developer": "Cohere", "developer_id": "cohere", "family": "north", "description": "Open-weight translation model for more than 50 languages and locale variants", "release_date": "2026-09-09", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-4.0", "parameters_b": 218, "active_parameters_b": 25, "parameters_source": "primary_source", "architecture": "moe", "context_window": 16000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": false, "source_url": "https://docs.cohere.com/docs/north-small-translate-1.0", "source_type": "primary", "sources": [ { "label": "Cohere North Small Translate release announcement", "url": "https://docs.cohere.com/changelog/north-small-translate-1-0", "type": "primary" }, { "label": "Cohere North Small Translate technical report", "url": "https://arxiv.org/html/2609.13916v1", "type": "primary" }, { "label": "Cohere North Small Translate model documentation", "url": "https://docs.cohere.com/docs/north-small-translate-1.0", "type": "primary" }, { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/North-Small-Translate-1.0", "type": "weights" }, { "label": "Cohere exact capability configuration", "url": "https://github.com/cohere-ai/cohere-developer-experience/blob/main/fern/pages/models/north/north-small-translate-1.0.mdx", "type": "primary" }, { "label": "Cohere capability renderer", "url": "https://raw.githubusercontent.com/cohere-ai/cohere-developer-experience/main/fern/components/model-showcase.tsx", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/cohere/north-small-translate-1-0.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The provider API identifier is north-small-translate-09-2026; the matching weights are North-Small-Translate-1.0. The catalog ID is not a verified API alias. Total and active parameters are nominal developer-reported counts. Context uses the advertised 16K input limit; the card separately lists 16K output. Native reasoning and tool use are disabled in the exact documentation. The optional agentic translation workflow makes multiple calls to the same non-reasoning model.", "field_status": {}, "audited_at": "2026-10-26T18:30:34.221109Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/156.json" }, { "id": "inception/mercury-2.5", "slug": "inception--mercury-2.5", "name": "Mercury 2.5", "developer": "Inception", "developer_id": "inception", "family": "mercury", "description": "Diffusion language model for text chat, with adjustable reasoning effort, parallel tool calling, and structured outputs.", "release_date": "2026-09-08", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "diffusion", "context_window": 260000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.inceptionlabs.ai/blog/introducing-mercury-2-5", "source_type": "primary", "sources": [ { "label": "Inception Mercury 2.5 announcement", "url": "https://www.inceptionlabs.ai/blog/introducing-mercury-2-5", "type": "primary" }, { "label": "Inception dated release index", "url": "https://www.inceptionlabs.ai/blog", "type": "primary" }, { "label": "Inception model specifications", "url": "https://docs.inceptionlabs.ai/get-started/models", "type": "primary" }, { "label": "Inception reasoning controls", "url": "https://docs.inceptionlabs.ai/capabilities/reasoning-efforts", "type": "primary" }, { "label": "Inception Terms of Use", "url": "https://www.inceptionlabs.ai/docs/terms-of-use", "type": "primary" }, { "label": "Inception Mercury scaling comparison", "url": "https://www.inceptionlabs.ai/blog/introducing-mercury-3", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/inception/mercury-2.5.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Inception has not published total or active parameter counts in the checked sources. Its later scaling chart gives relative total sizes without an absolute reference count. Dense versus MoE structure is also unspecified.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T13:39:19.819410+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/239.json" }, { "id": "inclusionai/ling-3.0-flash-sante", "slug": "inclusionai--ling-3.0-flash-sante", "name": "Ling 3.0 Flash Sante", "developer": "InclusionAI", "developer_id": "inclusionai", "family": "ling", "description": "Health and medicine specialization of Ling 3.0 Flash for medical reasoning and evidence-based retrieval", "release_date": "2026-09-04", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 127.486384608, "active_parameters_b": 5.1, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Sante", "source_type": "primary", "sources": [ { "label": "Ling-3.0-flash-Sante official model card", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Sante/blob/f74bd27bd26ffc71a50dd847805f8f7666ff0b43/README.md", "type": "primary" }, { "label": "Ling-3.0-flash-Sante checkpoint configuration", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Sante/blob/f74bd27bd26ffc71a50dd847805f8f7666ff0b43/config.json", "type": "primary" }, { "label": "Ling-3.0-flash-Sante implementation", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Sante/blob/f74bd27bd26ffc71a50dd847805f8f7666ff0b43/modeling_bailing_moe_v3.py", "type": "primary" }, { "label": "Exact Sante checkpoint: bounded safetensors header inventory", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Sante/tree/f74bd27bd26ffc71a50dd847805f8f7666ff0b43", "type": "primary" }, { "label": "Sante checkpoint shard index", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Sante/blob/f74bd27bd26ffc71a50dd847805f8f7666ff0b43/model.safetensors.index.json", "type": "primary" }, { "label": "Ant Ling original Sante announcement, official X syndication response", "url": "https://cdn.syndication.twimg.com/tweet-result?id=2095953758148853892&lang=en&token=0", "type": "primary" }, { "label": "Ling 3.0 Flash Sante is now available on AI Gateway for free", "url": "https://vercel.com/changelog/ling-3-0-flash-sante-is-now-available-on-ai-gateway-for-free", "type": "primary" }, { "label": "Ling 3.0 Flash Sante on Novita AI: Free API, Specs, and Pricing", "url": "https://blogs.novita.ai/ling-3-0-flash-sante-on-novita-ai-launch-pricing/", "type": "primary" }, { "label": "Sante repository metadata", "url": "https://huggingface.co/api/models/inclusionAI/Ling-3.0-flash-Sante", "type": "primary" }, { "label": "Sante native chat template", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Sante/blob/f74bd27bd26ffc71a50dd847805f8f7666ff0b43/chat_template.jinja", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/inclusionai/ling-3.0-flash-sante.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete checkpoint: 127.486384608B learned parameters = 124.414191072B main model + 3.072193536B auxiliary MTP. The official 124B label rounds the main model; 5.1B active per token is the nominal developer figure, whose precise embedding, output-head and MTP inclusion convention is not specified. Header raw inventory includes 20,992 nonlearned expert_bias buffer elements, which are excluded. Main input embeddings and output head are untied; MTP reuses them. Hybrid MoE combines 35 KDA and 7 gated MLA main layers; two initial dense FFNs, then 40 layers selecting 8 of 512 routed experts plus one shared expert. Original API launch September 4, 2026 predates the October public-weight repository; Novita separately records a September 3 catalog addition.", "parameter_components_b": { "main_model": 124.414191072, "mtp": 3.072193536 }, "field_status": {}, "audited_at": "2026-10-26T21:00:26.273264Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/244.json" }, { "id": "inclusionai/ling-3.0-flash-vl", "slug": "inclusionai--ling-3.0-flash-vl", "name": "Ling 3.0 Flash VL", "developer": "InclusionAI", "developer_id": "inclusionai", "family": "ling", "description": "Native multimodal hybrid MoE model for image and video understanding, reasoning, and visual agents", "release_date": "2026-09-04", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 124, "active_parameters_b": 5.5, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-VL", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-VL", "type": "weights" }, { "label": "Official later model comparison", "url": "https://developer.ant-ling.com/en/docs/models/ling/", "type": "primary" }, { "label": "Official release changelog", "url": "https://developer.ant-ling.com/en/docs/getting-started/changelog/", "type": "primary" }, { "label": "Exact checkpoint configuration", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-VL/blob/c22bf252f15b66d18ca00a15989ac7ed46b7f09e/config.json", "type": "primary" }, { "label": "Exact checkpoint tensor index", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-VL/blob/c22bf252f15b66d18ca00a15989ac7ed46b7f09e/model.safetensors.index.json", "type": "weights" }, { "label": "Official multimodal API guide", "url": "https://developer.ant-ling.com/en/docs/tutorials/multimodal-understanding", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/inclusionai/ling-3.0-flash-vl.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "124B total and 5.5B active are developer nominal specifications. The full BF16 checkpoint contains 124.848460496B tensor elements, including a 0.415893744B vision encoder and 0.0183552B projector; 0.00002048B routing correction elements are not conclusively classified as gradient-learned weights. No MTP weights are exported. Supported 262144 context uses YaRN over a 131072 config base. Public API/chat launch was September 4, 2026; open-weight announcement followed September 8.", "field_status": {}, "audited_at": "2026-10-26T21:02:40.480546+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/245.json" }, { "id": "ifm/k2-horizon-0.9b", "slug": "ifm--k2-horizon-0.9b", "name": "K2 Horizon 0.9B", "developer": "Institute of Foundation Models", "developer_id": "ifm", "family": "k2-horizon", "description": null, "release_date": "2026-09-03", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 1.078285824, "active_parameters_b": null, "parameters_source": "https://huggingface.co/api/models/IFM/K2-Horizon-0.9B/revision/d50881f4f8dc6923fb8f301e1c2313fe3a2b6c1d", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ifm.ai/blog/k2/", "source_type": "primary", "sources": [ { "label": "IFM launch announcement", "url": "https://ifm.ai/blog/k2/", "type": "primary" }, { "label": "Official model card", "url": "https://huggingface.co/IFM/K2-Horizon-0.9B/blob/d50881f4f8dc6923fb8f301e1c2313fe3a2b6c1d/README.md", "type": "primary" }, { "label": "Official configuration", "url": "https://huggingface.co/IFM/K2-Horizon-0.9B/blob/d50881f4f8dc6923fb8f301e1c2313fe3a2b6c1d/config.json", "type": "primary" }, { "label": "Official checkpoint tensor metadata", "url": "https://huggingface.co/api/models/IFM/K2-Horizon-0.9B/revision/d50881f4f8dc6923fb8f301e1c2313fe3a2b6c1d", "type": "primary" }, { "label": "Checkpoint weight index", "url": "https://huggingface.co/IFM/K2-Horizon-0.9B/blob/d50881f4f8dc6923fb8f301e1c2313fe3a2b6c1d/model.safetensors.index.json", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:07:44.790Z", "last_seen": "2026-10-30T07:07:44.790Z", "upstream_present": false, "availability": "public", "notes": "Final post-trained checkpoint; intermediate training checkpoints and quantizations are not separate releases. IFM names this variant 0.9B. Total here is the full 1.07829B stored BF16 parameter count in the official repository metadata, corroborated by the weight index byte count; it differs from the nominal model name." }, { "id": "ifm/k2-horizon-3.7b", "slug": "ifm--k2-horizon-3.7b", "name": "K2 Horizon 3.7B", "developer": "Institute of Foundation Models", "developer_id": "ifm", "family": "k2-horizon", "description": null, "release_date": "2026-09-03", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 5.05825536, "active_parameters_b": null, "parameters_source": "https://huggingface.co/api/models/IFM/K2-Horizon-3.7B/revision/4525090a8f0fce1026d861701b9bde2fb726b7a0", "architecture": "dense", "context_window": 524288, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ifm.ai/blog/k2/", "source_type": "primary", "sources": [ { "label": "IFM launch announcement", "url": "https://ifm.ai/blog/k2/", "type": "primary" }, { "label": "Official model card", "url": "https://huggingface.co/IFM/K2-Horizon-3.7B/blob/4525090a8f0fce1026d861701b9bde2fb726b7a0/README.md", "type": "primary" }, { "label": "Official configuration", "url": "https://huggingface.co/IFM/K2-Horizon-3.7B/blob/4525090a8f0fce1026d861701b9bde2fb726b7a0/config.json", "type": "primary" }, { "label": "Official checkpoint tensor metadata", "url": "https://huggingface.co/api/models/IFM/K2-Horizon-3.7B/revision/4525090a8f0fce1026d861701b9bde2fb726b7a0", "type": "primary" }, { "label": "Checkpoint weight index", "url": "https://huggingface.co/IFM/K2-Horizon-3.7B/blob/4525090a8f0fce1026d861701b9bde2fb726b7a0/model.safetensors.index.json", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:07:44.790Z", "last_seen": "2026-10-30T07:07:44.790Z", "upstream_present": false, "availability": "public", "notes": "Final post-trained checkpoint; intermediate training checkpoints and quantizations are not separate releases. IFM names this variant 3.7B. Total here is the full 5.05826B stored BF16 parameter count in the official repository metadata, corroborated by the weight index byte count; it differs from the nominal model name." }, { "id": "ifm/k2-horizon-32b", "slug": "ifm--k2-horizon-32b", "name": "K2 Horizon 32B", "developer": "Institute of Foundation Models", "developer_id": "ifm", "family": "k2-horizon", "description": null, "release_date": "2026-09-03", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 34.77930496, "active_parameters_b": null, "parameters_source": "https://huggingface.co/api/models/IFM/K2-Horizon-32B/revision/4b05c13ad1972e8f470ab93ff90c3e90010ce592", "architecture": "dense", "context_window": 524288, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ifm.ai/blog/k2/", "source_type": "primary", "sources": [ { "label": "IFM launch announcement", "url": "https://ifm.ai/blog/k2/", "type": "primary" }, { "label": "Official model card", "url": "https://huggingface.co/IFM/K2-Horizon-32B/blob/4b05c13ad1972e8f470ab93ff90c3e90010ce592/README.md", "type": "primary" }, { "label": "Official configuration", "url": "https://huggingface.co/IFM/K2-Horizon-32B/blob/4b05c13ad1972e8f470ab93ff90c3e90010ce592/config.json", "type": "primary" }, { "label": "Official checkpoint tensor metadata", "url": "https://huggingface.co/api/models/IFM/K2-Horizon-32B/revision/4b05c13ad1972e8f470ab93ff90c3e90010ce592", "type": "primary" }, { "label": "Checkpoint weight index", "url": "https://huggingface.co/IFM/K2-Horizon-32B/blob/4b05c13ad1972e8f470ab93ff90c3e90010ce592/model.safetensors.index.json", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:07:44.790Z", "last_seen": "2026-10-30T07:07:44.790Z", "upstream_present": false, "availability": "public", "notes": "Final post-trained checkpoint; intermediate training checkpoints and quantizations are not separate releases. IFM names this variant 32B. Total here is the full 34.7793B stored BF16 parameter count in the official repository metadata, corroborated by the weight index byte count; it differs from the nominal model name." }, { "id": "ifm/k2-horizon-375b-a23b", "slug": "ifm--k2-horizon-375b-a23b", "name": "K2 Horizon 375B-A23B", "developer": "Institute of Foundation Models", "developer_id": "ifm", "family": "k2-horizon", "description": null, "release_date": "2026-09-03", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 379.167159168, "active_parameters_b": 23, "parameters_source": "https://huggingface.co/api/models/IFM/K2-Horizon-375B-A23B/revision/78272deca7adc2f90d154288b5161172283c80b3", "architecture": "moe", "context_window": 524288, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ifm.ai/blog/k2/", "source_type": "primary", "sources": [ { "label": "IFM launch announcement", "url": "https://ifm.ai/blog/k2/", "type": "primary" }, { "label": "Official model card", "url": "https://huggingface.co/IFM/K2-Horizon-375B-A23B/blob/78272deca7adc2f90d154288b5161172283c80b3/README.md", "type": "primary" }, { "label": "Official configuration", "url": "https://huggingface.co/IFM/K2-Horizon-375B-A23B/blob/78272deca7adc2f90d154288b5161172283c80b3/config.json", "type": "primary" }, { "label": "Official checkpoint tensor metadata", "url": "https://huggingface.co/api/models/IFM/K2-Horizon-375B-A23B/revision/78272deca7adc2f90d154288b5161172283c80b3", "type": "primary" }, { "label": "Checkpoint weight index", "url": "https://huggingface.co/IFM/K2-Horizon-375B-A23B/blob/78272deca7adc2f90d154288b5161172283c80b3/model.safetensors.index.json", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:07:44.790Z", "last_seen": "2026-10-30T07:07:44.790Z", "upstream_present": false, "availability": "public", "notes": "Final post-trained checkpoint; intermediate training checkpoints and quantizations are not separate releases. IFM names this variant 375B. Total here is the full 379.167B stored BF16 parameter count in the official repository metadata, corroborated by the weight index byte count; it differs from the nominal model name. Active 23B is the publisher's reported per-token count; the announcement does not specify whether its counting convention matches the stored total." }, { "id": "ifm/k2-horizon-7b", "slug": "ifm--k2-horizon-7b", "name": "K2 Horizon 7B", "developer": "Institute of Foundation Models", "developer_id": "ifm", "family": "k2-horizon", "description": null, "release_date": "2026-09-03", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 8.99917824, "active_parameters_b": null, "parameters_source": "https://huggingface.co/api/models/IFM/K2-Horizon-7B/revision/4622f94a6632384784471fd674d5b78f8d527798", "architecture": "dense", "context_window": 524288, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ifm.ai/blog/k2/", "source_type": "primary", "sources": [ { "label": "IFM launch announcement", "url": "https://ifm.ai/blog/k2/", "type": "primary" }, { "label": "Official model card", "url": "https://huggingface.co/IFM/K2-Horizon-7B/blob/4622f94a6632384784471fd674d5b78f8d527798/README.md", "type": "primary" }, { "label": "Official configuration", "url": "https://huggingface.co/IFM/K2-Horizon-7B/blob/4622f94a6632384784471fd674d5b78f8d527798/config.json", "type": "primary" }, { "label": "Official checkpoint tensor metadata", "url": "https://huggingface.co/api/models/IFM/K2-Horizon-7B/revision/4622f94a6632384784471fd674d5b78f8d527798", "type": "primary" }, { "label": "Checkpoint weight index", "url": "https://huggingface.co/IFM/K2-Horizon-7B/blob/4622f94a6632384784471fd674d5b78f8d527798/model.safetensors.index.json", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:07:44.790Z", "last_seen": "2026-10-30T07:07:44.790Z", "upstream_present": false, "availability": "public", "notes": "Final post-trained checkpoint; intermediate training checkpoints and quantizations are not separate releases. IFM names this variant 7B. Total here is the full 8.99918B stored BF16 parameter count in the official repository metadata, corroborated by the weight index byte count; it differs from the nominal model name." }, { "id": "ifm/k2-horizon-mova-36b-a4b", "slug": "ifm--k2-horizon-mova-36b-a4b", "name": "K2 Horizon MoVA 36B-A4B", "developer": "Institute of Foundation Models", "developer_id": "ifm", "family": "k2-horizon", "description": null, "release_date": "2026-09-03", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 37.44479202, "active_parameters_b": 4, "parameters_source": "https://huggingface.co/api/models/IFM/K2-Horizon-MoVA-36B-A4B/revision/e9ce717642b2074082875be233c02c729ba47960", "architecture": "moe", "context_window": 524288, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ifm.ai/blog/k2/", "source_type": "primary", "sources": [ { "label": "IFM launch announcement", "url": "https://ifm.ai/blog/k2/", "type": "primary" }, { "label": "Official model card", "url": "https://huggingface.co/IFM/K2-Horizon-MoVA-36B-A4B/blob/e9ce717642b2074082875be233c02c729ba47960/README.md", "type": "primary" }, { "label": "Official configuration", "url": "https://huggingface.co/IFM/K2-Horizon-MoVA-36B-A4B/blob/e9ce717642b2074082875be233c02c729ba47960/config.json", "type": "primary" }, { "label": "Official checkpoint tensor metadata", "url": "https://huggingface.co/api/models/IFM/K2-Horizon-MoVA-36B-A4B/revision/e9ce717642b2074082875be233c02c729ba47960", "type": "primary" }, { "label": "Checkpoint weight index", "url": "https://huggingface.co/IFM/K2-Horizon-MoVA-36B-A4B/blob/e9ce717642b2074082875be233c02c729ba47960/model.safetensors.index.json", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:07:44.790Z", "last_seen": "2026-10-30T07:07:44.790Z", "upstream_present": false, "availability": "public", "notes": "Final post-trained checkpoint; intermediate training checkpoints and quantizations are not separate releases. IFM names this variant 36B. Total here is the full 37.4448B stored BF16 parameter count in the official repository metadata, corroborated by the weight index byte count; it differs from the nominal model name. Active 4B is the publisher's reported per-token count; the announcement does not specify whether its counting convention matches the stored total." }, { "id": "openai/gpt-6-astra", "slug": "openai--gpt-6-astra", "name": "GPT-6 Astra", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-6", "description": "Reasoning model for coding, computer use, research and document creation, with text and image input.", "release_date": "2026-09-03", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1050000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/safety-overview-gpt-6-astra/", "source_type": "primary", "sources": [ { "label": "GPT-6 Astra: A new generation of intelligence", "url": "https://openai.com/index/gpt-6-astra/", "type": "primary" }, { "label": "Safety overview: GPT-6 Astra", "url": "https://openai.com/index/safety-overview-gpt-6-astra/", "type": "primary" }, { "label": "OpenAI API changelog", "url": "https://developers.openai.com/api/docs/changelog.md", "type": "primary" }, { "label": "GPT-6 Astra API model reference", "url": "https://developers.openai.com/api/docs/models/gpt-6-astra.md", "type": "primary" }, { "label": "GPT-6 Astra System Card", "url": "https://deploymentsafety.openai.com/gpt-6-astra/gpt-6-astra.pdf", "type": "primary" }, { "label": "Public Summary of Training Content for GPT-6 Astra", "url": "https://cdn.openai.com/pdf/gpt-6-astra-eu-ai-act-public-summary-of-training-content.pdf", "type": "primary" }, { "label": "GPT-6.1 Sol System Card", "url": "https://deploymentsafety.openai.com/gpt-6-1-sol/gpt-6-1-sol.pdf", "type": "primary" }, { "label": "GPT-6.1 Terra System Card", "url": "https://deploymentsafety.openai.com/gpt-6-1-terra/gpt-6-1-terra.pdf", "type": "primary" }, { "label": "Using GPT-6", "url": "https://developers.openai.com/api/docs/guides/latest-model.md", "type": "primary" }, { "label": "Prompt caching", "url": "https://developers.openai.com/api/docs/guides/prompt-caching.md", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "OpenAI model inventory", "url": "https://developers.openai.com/api/docs/models/all.md", "type": "primary" }, { "label": "OpenAI API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "Fast mode", "url": "https://developers.openai.com/api/docs/guides/fast-mode.md", "type": "primary" }, { "label": "Jalapeño’s first results show industry-leading speed and efficiency in AI inference", "url": "https://openai.com/index/jalapeno-first-results/", "type": "primary" }, { "label": "Suppressing misalignment monitoring alerts and disabling misalignment monitoring", "url": "https://alignment.openai.com/misalignment-reports/suppressing-misalignment-monitoring-alerts-and-disabling-misalignment-monitoring/", "type": "primary" }, { "label": "ChatGPT release notes", "url": "https://help.openai.com/en/articles/6825453-chatgpt-release-notes", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-6-astra.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Initial release was September 3, 2026, with access limited to selected organizations before broader rollout over subsequent days. Only the undated gpt-6-astra identifier is listed in the current API reference; no immutable date-suffixed snapshot is inferred. Documented context is 1,050,000 tokens with 922,000 maximum input and 128,000 maximum output. Reasoning supports low, medium, high, xhigh and max, but not none. Tool calling requires Responses despite Chat Completions support. PDF is an API document format converted into text and page images. Fast and Ultrafast are processing modes, not evidence of separate model weights. Total and active parameter counts remain undisclosed. The EU training summary identifies Astra as a new model, so no prior-model count is inherited. Production network architecture remains unverified; KV caching and activation classifiers do not establish topology. No exact API retirement entry was found; endpoint operation was not tested.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "unverified" }, "audited_at": "2026-10-27T17:10:09.395798Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/441.json" }, { "id": "alibaba/qwen3.8-max-0902", "slug": "alibaba--qwen3.8-max-0902", "name": "Qwen3.8 Max 0902", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "September 2026 snapshot of Qwen3.8 Max with a one-million-token context, thinking mode, tool calling, and text, image, video and PDF input.", "release_date": "2026-09-02", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": 2400, "active_parameters_b": 95, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text", "image", "video", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-8-max", "source_type": "primary", "sources": [ { "label": "Alibaba Max model and snapshot specifications", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-8-max", "type": "primary" }, { "label": "Alibaba explicit Max-to-0902 endpoint mapping", "url": "https://www.alibabacloud.com/en/notice/model_studio_update_notice_for_qwen38max_models_863?_p_lc=1", "type": "primary" }, { "label": "Alibaba model release chronology", "url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "type": "primary" }, { "label": "Alibaba hosted Max parameter disclosure", "url": "https://www.alibabacloud.com/en/press-room/alibaba-unveils-qwen3-8-max?_p_lc=1", "type": "primary" }, { "label": "QwenCloud exact 0902 model page", "url": "https://www.qwencloud.com/models/qwen3.8-max-0902", "type": "primary" }, { "label": "Alibaba exact-snapshot PDF support", "url": "https://www.alibabacloud.com/help/en/model-studio/pdf-understanding", "type": "primary" }, { "label": "Qwen hosted Max nominal total and active parameters", "url": "https://qwen.ai/blog?id=qwen3.8", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=qwen3.8-max-0902&limit=100", "label": "Official Qwen inventory search for the exact hosted model", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=Qwen3.8&limit=1000", "label": "Official Qwen Qwen3.8 weight inventory", "type": "primary" }, { "url": "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B", "label": "Original developer documentation checked for model availability", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.8-max-0902.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Alibaba publishes a nominal 2.4T total and 95B active for hosted Qwen3.8-Max. The 95B active count is inherited by inference from the developer-documented 0902 snapshot relationship; no structural change is disclosed. PDF input is supported in Beijing/Singapore through Chat Completions, not Responses. No public weights for the complete 0902 hosted revision were found; its weight license remains undisclosed. Classified as closed weights: no public release of this hosted variant was found in the developer documentation and weight inventory.", "field_status": { "license": "unverified" }, "audited_at": "2026-10-28T08:13:49.514266+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/071.json" }, { "id": "google/gemini-3.8-flash", "slug": "google--gemini-3.8-flash", "name": "Gemini 3.8 Flash", "developer": "Google", "developer_id": "google", "family": "gemini-flash", "description": "Google's multimodal reasoning model based on Gemini 3.7 Flash, with text output, configurable thinking levels, and function calling.", "release_date": "2026-09-02", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ai.google.dev/gemini-api/docs/changelog#september-2-2026", "source_type": "primary", "sources": [ { "label": "Gemini 3.8 Flash model card", "url": "https://deepmind.google/models/model-cards/gemini-3-8-flash/", "type": "primary" }, { "label": "Gemini 3.8 Flash API model reference", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.8-flash", "type": "primary" }, { "label": "Introducing Gemini 3.8 Flash and 3.8 Flash Cyber", "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/", "type": "primary" }, { "label": "Gemini 3.7 Flash model card", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-7-Flash-Model-Card.pdf", "type": "primary" }, { "label": "Gemini 3.6 Flash model card", "url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "type": "primary" }, { "label": "Gemini 3.5 Flash model card", "url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "type": "primary" }, { "label": "Gemini 3 Flash model card", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-Flash-Model-Card.pdf", "type": "primary" }, { "label": "Gemini 3 Pro model card, updated May 2026", "url": "https://deepmind.google/models/model-cards/gemini-3-pro/", "type": "primary" }, { "label": "Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Gemini API release notes — September 2, 2026", "url": "https://ai.google.dev/gemini-api/docs/changelog#september-2-2026", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemini-3.8-flash.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Sparse MoE Transformer based on Gemini 3.7 Flash. Total and active parameter counts are undisclosed for this model and its named bases in the reviewed sources. The context figure is the documented input limit.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T13:07:02.817323+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/212.json" }, { "id": "meta/muse-spark-1.3", "slug": "meta--muse-spark-1.3", "name": "Muse Spark 1.3", "developer": "Meta", "developer_id": "meta", "family": "muse", "description": "Muse Spark 1.3 is Meta's hosted multimodal reasoning model for agentic and coding workflows. Available through Meta Model API and Muse Code, it supports tool calling and a 1,048,576-token context window. Meta reports improved instruction following and coding efficiency relative to Muse Spark 1.2.", "release_date": "2026-09-02", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 816, "active_parameters_b": 42, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "video", "pdf", "audio" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://research.meta.ai/blog/introducing-muse-spark-1-3", "source_type": "primary", "sources": [ { "label": "Meta Muse Spark 1.3 announcement", "url": "https://research.meta.ai/blog/introducing-muse-spark-1-3", "type": "primary" }, { "label": "Meta Model API models", "url": "https://dev.meta.ai/docs/models", "type": "primary" }, { "label": "Meta API terms", "url": "https://dev.meta.ai/legal/terms-of-service", "type": "primary" }, { "label": "Meta model designation policy", "url": "https://dev.meta.ai/legal/geographic-use-policy", "type": "primary" }, { "label": "Meta video and audio documentation", "url": "https://dev.meta.ai/docs/video-understanding", "type": "primary" }, { "label": "Muse Spark 1.3 evaluation report", "url": "https://research.meta.ai/static/muse-spark-1-3-evaluation-report", "type": "primary" }, { "label": "Meta reasoning documentation", "url": "https://dev.meta.ai/docs/reasoning", "type": "primary" }, { "label": "Meta tool calling documentation", "url": "https://dev.meta.ai/docs/tool-calling", "type": "primary" }, { "label": "Meta PyTorch Spark Open and Spark 1.3 shared-base statement", "url": "https://pytorch.org/blog/muse-spark-open/", "type": "primary" }, { "label": "Meta Spark Open parameter convention", "url": "https://dev.meta.ai/resources/blog/build-with-muse-spark-open", "type": "primary" }, { "label": "Meta Spark Open model card", "url": "https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B", "type": "primary" }, { "label": "Meta Spark Open release", "url": "https://research.meta.ai/blog/muse-spark-open", "type": "primary" }, { "label": "Muse Spark 1.3 evaluation methodology", "url": "https://research.meta.ai/static/muse-spark-1-3-multimodal-evaluation-methodology", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/meta/muse-spark-1.3.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Nominal 816B total, roughly 42B active per token and MoE architecture are inferred from Meta's explicit statement that Spark 1.3 and Spark Open share the same base model; no count-changing structural modification is documented. These are not exact hosted-deployment or modality-specific counts. Audio input is documented but understanding is not fully supported and response quality may be degraded.", "field_status": {}, "audited_at": "2026-10-27T14:26:20Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/284.json" }, { "id": "anthropic/claude-fable-5-1", "slug": "anthropic--claude-fable-5-1", "name": "Claude Fable 5.1", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude-fable", "description": "Claude model for demanding reasoning and long-horizon agentic work", "release_date": "2026-09-01", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1000000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://platform.claude.com/docs/en/models/fable-5-1/overview", "source_type": "primary", "sources": [ { "label": "Claude Fable 5.1 official specifications", "url": "https://platform.claude.com/docs/en/models/fable-5-1/overview", "type": "primary" }, { "label": "Anthropic models overview", "url": "https://platform.claude.com/docs/en/models/overview", "type": "primary" }, { "label": "Anthropic PDF support", "url": "https://platform.claude.com/docs/en/build-with-claude/pdf-support", "type": "primary" }, { "label": "Claude Fable product page", "url": "https://www.anthropic.com/claude/fable", "type": "primary" }, { "label": "Anthropic Commercial Terms", "url": "https://www.anthropic.com/legal/commercial-terms", "type": "primary" }, { "label": "Anthropic Transparency Hub", "url": "https://www.anthropic.com/transparency", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/anthropic/claude-fable-5-1.toml", "type": "catalog" }, { "label": "Claude Fable 5.1 and Mythos 5.1 System Card", "url": "https://www-cdn.anthropic.com/0339e6a7c5c7b87f5c07798616dc32c215d14235/Claude%20Fable%205.1%20%26%20Claude%20Mythos%205.1%20System%20Card.pdf", "type": "primary" }, { "label": "Anthropic research on model-weight availability", "url": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities", "type": "primary" }, { "label": "Introducing Claude Fable 5.1 and Claude Mythos 5.1", "url": "https://www.anthropic.com/claude-fable-and-mythos-5-1", "type": "primary" }, { "label": "Claude Fable 5.1 & Claude Mythos 5.1 System Card (official endpoint, fresh parameter review)", "url": "https://www.anthropic.com/claude-fable-5-1-mythos-5-1-system-card", "type": "primary" }, { "label": "Claude Fable 5 & Claude Mythos 5 System Card", "url": "https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf", "type": "primary" }, { "label": "Claude Mythos 5.1 official specifications", "url": "https://platform.claude.com/docs/en/models/mythos-5-1/overview", "type": "primary" }, { "label": "Claude Opus 5.5 System Card (official endpoint, fresh comparison review)", "url": "https://www.anthropic.com/claude-opus-5-5-system-card", "type": "primary" }, { "label": "Claude Sonnet 5.5 System Card", "url": "https://www.anthropic.com/claude-sonnet-5-5-system-card", "type": "primary" }, { "label": "Claude Haiku 5.5 System Card", "url": "https://www.anthropic.com/claude-haiku-5-5-system-card", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T10:24:57.941760Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/097.json", "notes": "Nominal total and active parameter counts remain undisclosed in the checked Anthropic sources. Fable 5.1 and Mythos 5.1 share identical model weights, but neither has a published size." }, { "id": "anthropic/claude-mythos-5-1", "slug": "anthropic--claude-mythos-5-1", "name": "Claude Mythos 5.1", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude-mythos", "description": "Restricted Claude model for advanced cybersecurity and biology research workflows", "release_date": "2026-09-01", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1000000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://platform.claude.com/docs/en/models/mythos-5-1/overview", "source_type": "primary", "sources": [ { "label": "Anthropic exact-model specifications", "url": "https://platform.claude.com/docs/en/models/mythos-5-1/overview", "type": "primary" }, { "label": "Anthropic system card", "url": "https://www.anthropic.com/claude-fable-5-1-mythos-5-1-system-card", "type": "primary" }, { "label": "Anthropic commercial terms", "url": "https://www.anthropic.com/legal/commercial-terms", "type": "primary" }, { "label": "Anthropic release notes", "url": "https://platform.claude.com/docs/en/release-notes/overview", "type": "primary" }, { "label": "Anthropic PDF support", "url": "https://platform.claude.com/docs/en/build-with-claude/pdf-support", "type": "primary" }, { "label": "Anthropic statement on Claude weight availability", "url": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities", "type": "research" }, { "label": "Introducing Claude Fable 5.1 and Claude Mythos 5.1", "url": "https://www.anthropic.com/claude-fable-and-mythos-5-1", "type": "primary" }, { "label": "System Card: Claude Opus 5.5", "url": "https://www.anthropic.com/claude-opus-5-5-system-card", "type": "primary" }, { "label": "System Card: Claude Sonnet 5.5", "url": "https://www.anthropic.com/claude-sonnet-5-5-system-card", "type": "primary" }, { "label": "System Card: Claude Haiku 5.5", "url": "https://www.anthropic.com/claude-haiku-5-5-system-card", "type": "primary" }, { "label": "System Card: Claude Fable 5 & Claude Mythos 5", "url": "https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf", "type": "primary" }, { "label": "Claude Fable 5.1 model overview", "url": "https://platform.claude.com/docs/en/models/fable-5-1/overview", "type": "primary" }, { "label": "An alignment assessment of the UK AISI cybertesting incident", "url": "https://www.anthropic.com/research/alignment-assessment-ukaisi", "type": "research" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/anthropic/claude-mythos-5-1.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T10:32:08.468260Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/103.json", "notes": "Nominal total and active parameter counts remain undisclosed in the checked Anthropic sources. Mythos 5.1 and Fable 5.1 share identical model weights, but neither has a published size." }, { "id": "openbmb/minicpm5-2b", "slug": "openbmb--minicpm5-2b", "name": "MiniCPM5-2B", "developer": "OpenBMB", "developer_id": "openbmb", "family": "minicpm5", "description": "Dense text language model with 2.517B parameters, native 131,072-token context, optional thinking, and XML-style tool calling.", "release_date": "2026-09", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 2.51675648, "active_parameters_b": 2.51675648, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openbmb.cn/docs/en/changelog", "source_type": "primary", "sources": [ { "label": "Official MiniCPM5-2B model card (pinned)", "url": "https://huggingface.co/openbmb/MiniCPM5-2B/blob/f97400052a43d642bbc6e9975e2397e3ae6a6b52/README.md", "type": "primary" }, { "label": "Official configuration (pinned)", "url": "https://huggingface.co/openbmb/MiniCPM5-2B/blob/f97400052a43d642bbc6e9975e2397e3ae6a6b52/config.json", "type": "primary" }, { "label": "Complete checkpoint index (pinned)", "url": "https://huggingface.co/openbmb/MiniCPM5-2B/blob/f97400052a43d642bbc6e9975e2397e3ae6a6b52/model.safetensors.index.json", "type": "weights" }, { "label": "Official checkpoint (bounded tensor-header verification)", "url": "https://huggingface.co/openbmb/MiniCPM5-2B/resolve/f97400052a43d642bbc6e9975e2397e3ae6a6b52/model-00000-of-00001.safetensors", "type": "weights" }, { "label": "OpenBMB release changelog", "url": "https://openbmb.cn/docs/en/changelog", "type": "primary" }, { "label": "ModelBest announcement archive", "url": "https://www.modelbest.cn/en", "type": "primary" }, { "label": "ModelBest September press release", "url": "https://en.prnasia.com/releases/global/china-s-latest-ai-model-brings-general-purpose-agentic-capability-to-edge-devices-at-a-lower-cost-547140.shtml", "type": "primary" }, { "label": "Official repository history", "url": "https://huggingface.co/api/models/openbmb/MiniCPM5-2B/commits/main", "type": "primary" }, { "label": "EliovpAI/Paiton original total/active comparison convention", "url": "https://huggingface.co/EliovpAI/MiniCPM5-2B-W4A16-Paiton-RDNA4/blob/4bcf80e1bcf1bc8f5c2d657086121958cfe145cb/SELECTION.md", "type": "primary" }, { "label": "Paiton original runtime benchmark methodology", "url": "https://huggingface.co/EliovpAI/MiniCPM5-2B-W4A16-Paiton-RDNA4/blob/4bcf80e1bcf1bc8f5c2d657086121958cfe145cb/BENCHMARKS.md", "type": "primary" }, { "label": "Official comparison-revision checkpoint metadata", "url": "https://huggingface.co/api/models/openbmb/MiniCPM5-2B/revision/bca24102e83b4a03a5d7e9d281e9a6790cc6e3a3?blobs=true", "type": "primary" }, { "label": "Official catalog-revision checkpoint metadata", "url": "https://huggingface.co/api/models/openbmb/MiniCPM5-2B/revision/f97400052a43d642bbc6e9975e2397e3ae6a6b52?blobs=true", "type": "primary" }, { "label": "OpenBMB current parameter and VRAM reference", "url": "https://openbmb.cn/docs/en/models/vram-and-performance", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openbmb/minicpm5-2b.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/openbmb/MiniCPM5-2B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Total parameters are 2,516,756,480 including separate input embeddings and output head; the card's 1,981,982,720 non-embedding count excludes both. Active parameters use EliovpAI/Paiton's published 2.51675648B convention counting the complete dense model for the identical checkpoint; this is not OpenBMB disclosure or literal per-token embedding participation. Public-weight release is recorded at September 2026 precision: OpenBMB says September 7, ModelBest says September 8, and its press release conflicts internally on September 8/9. HF creation/upload on September 6 is not proof of public availability; July 19 was an earlier unveiling in the accessed official excerpt.", "field_status": {}, "audited_at": "2026-10-27T17:29:33.890534Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/462.json", "parameter_components_b": { "input_embedding": 0.26738688, "output_head": 0.26738688, "attention": 0.396361728, "mlp": 1.585446912, "norms": 0.00017408 } }, { "id": "quiverai/arrow-2", "slug": "quiverai--arrow-2", "name": "Arrow 2", "developer": "QuiverAI", "developer_id": "quiverai", "family": "arrow", "description": "QuiverAI model that generates, edits, vectorizes and animates SVG code from text and image inputs.", "release_date": "2026-09", "date_precision": "month", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://quiver.ai/blog/introducing-arrow-2-0", "source_type": "primary", "sources": [ { "label": "Arrow 2 — exact model documentation", "url": "https://docs.quiver.ai/developers/models/arrow-2", "type": "primary" }, { "label": "Arrow model comparison table", "url": "https://docs.quiver.ai/developers/models", "type": "primary" }, { "label": "Introducing Arrow 2 and Arrow 2 Telos", "url": "https://quiver.ai/blog/introducing-arrow-2-0", "type": "primary" }, { "label": "App and API Platform changelog", "url": "https://quiver.ai/changelog", "type": "primary" }, { "label": "Official Arrow 2 introduction on X", "url": "https://x.com/QuiverAI/status/2100295136261349802", "type": "primary" }, { "label": "Migrate to the Responses API", "url": "https://docs.quiver.ai/developers/guides/migrate-to-responses", "type": "primary" }, { "label": "Text to SVG", "url": "https://docs.quiver.ai/developers/models/text-to-svg", "type": "primary" }, { "label": "Terms of Service — sections 5 and 6", "url": "https://quiver.ai/legal/terms", "type": "primary" }, { "label": "Arrow models", "url": "https://quiver.ai/models", "type": "primary" }, { "label": "TIDES: Test-time Inference Drift Exploitation via Scaling — author project and extended results", "url": "https://haorandai.com/tides-paper/", "type": "primary" }, { "label": "Arrow 2 — current product model page", "url": "https://quiver.ai/models/arrow-2", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/quiverai/arrow-2.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Release month is September 2026: the official blog dated September 7 states App availability, while the official September 16 X introduction and changelog describe App/API availability and platform changes. The earliest day is unresolved. Context is 131,072 tokens, with a separate 65,536-token output cap. The model generates SVG source code, including animation, rather than raster images or video; exported previews do not add native output modalities. Responses supports caller-executed function/custom tools when enabled for the key. Total and active parameters remain undisclosed; no counted base lineage is documented for Arrow 2. Neural topology is also undisclosed. Sizes from the contributors' separate research models do not establish Arrow 2's size.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T17:38:15.954378Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/474.json" }, { "id": "quiverai/arrow-2-telos", "slug": "quiverai--arrow-2-telos", "name": "Arrow 2 Telos", "developer": "QuiverAI", "developer_id": "quiverai", "family": "arrow", "description": "QuiverAI model that generates and refines editable SVG code from text, images and source SVGs.", "release_date": "2026-09", "date_precision": "month", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1050000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://quiver.ai/blog/introducing-arrow-2-0", "source_type": "primary", "sources": [ { "label": "Arrow 2 Telos model reference", "url": "https://docs.quiver.ai/developers/models/arrow-2-telos", "type": "primary" }, { "label": "Arrow model comparison", "url": "https://docs.quiver.ai/developers/models", "type": "primary" }, { "label": "Arrow 2 Telos product model page", "url": "https://quiver.ai/models/arrow-2-telos", "type": "primary" }, { "label": "Introducing Arrow 2 and Arrow 2 Telos", "url": "https://quiver.ai/blog/introducing-arrow-2-0", "type": "primary" }, { "label": "QuiverAI changelog", "url": "https://quiver.ai/changelog", "type": "primary" }, { "label": "Arrow 2 app and API launch announcement", "url": "https://x.com/QuiverAI/status/2100295136261349802", "type": "primary" }, { "label": "Text to SVG model guide", "url": "https://docs.quiver.ai/developers/models/text-to-svg", "type": "primary" }, { "label": "Migrate to the Responses API", "url": "https://docs.quiver.ai/developers/guides/migrate-to-responses", "type": "primary" }, { "label": "Create OpenResponse API reference", "url": "https://docs.quiver.ai/api-reference/openresponses/create-open-response", "type": "primary" }, { "label": "QuiverAI Terms of Service", "url": "https://quiver.ai/legal/terms", "type": "primary" }, { "label": "QuiverAI company and research announcement", "url": "https://quiver.ai/blog/announcing-our-seed-round", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/quiverai/arrow-2-telos.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Release month is verified as September 2026; the September 7 blog and September 16 app/API launch records leave the earliest Telos release day unresolved. The published context limit is 1,050,000 tokens, with a separate 65,536-token output limit. Outputs are editable SVG source code, including coded animations, not native raster images or video. Reasoning is documented; caller-executed function/custom tools require a key/model catalog that enables Responses. Total and active parameter counts remain undisclosed: QuiverAI identifies neither a sized base model nor the frontier models used for refinement. Neural architecture is also undisclosed. No live inference or authenticated model-catalog request was performed.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T17:41:25.284Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/475.json" }, { "id": "celeris/celeris-1-magnus", "slug": "celeris--celeris-1-magnus", "name": "Celeris-1 Magnus", "developer": "Celeris", "developer_id": "celeris", "family": "celeris", "description": null, "release_date": "2026-08-31", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "diffusion", "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://celeris.ai/introducing-celeris-1-magnus", "source_type": "primary", "sources": [ { "label": "Publisher release evidence", "url": "https://celeris.ai/introducing-celeris-1-magnus", "type": "primary" }, { "label": "Celeris model documentation", "url": "https://docs.celeris.ai/models", "type": "primary" }, { "label": "Celeris reasoning and tools", "url": "https://docs.celeris.ai/making-requests", "type": "primary" }, { "label": "Company and API identity", "url": "https://celeris.ai/privacy", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:07:44.790Z", "last_seen": "2026-10-30T07:07:44.790Z", "upstream_present": false, "availability": "public", "notes": "Hosted hybrid diffusion language model. Total and active parameter counts are not disclosed in the reviewed primary documentation; no downloadable checkpoint is documented. August31 announcement identifies Magnus as a separate, larger second model available to all API customers that day." }, { "id": "tencent/hy4-preview", "slug": "tencent--hy4-preview", "name": "Hy4 preview", "developer": "Tencent", "developer_id": "tencent", "family": "Hy", "description": "Tencent Hy preview text-generation MoE model with optional reasoning and tool calling, designed for coding and productivity tasks.", "release_date": "2026-08-28", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 770, "active_parameters_b": 49, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.tencent.com/tencent-releases-and-open-sources-tencent-hy4-preview/", "source_type": "primary", "sources": [ { "label": "Tencent Releases and Open-Sources Tencent Hy4 preview", "url": "https://www.tencent.com/tencent-releases-and-open-sources-tencent-hy4-preview/", "type": "primary" }, { "label": "Hy4 preview official model card (pinned)", "url": "https://huggingface.co/tencent/Hy4-preview/blob/705d81ee51566a186d645b74c974d642ef2828fe/README.md", "type": "primary" }, { "label": "Hy4 preview checkpoint configuration (pinned)", "url": "https://huggingface.co/tencent/Hy4-preview/blob/705d81ee51566a186d645b74c974d642ef2828fe/config.json", "type": "primary" }, { "label": "Hy4 preview license (pinned)", "url": "https://huggingface.co/tencent/Hy4-preview/blob/705d81ee51566a186d645b74c974d642ef2828fe/LICENSE", "type": "primary" }, { "label": "Hy4 preview chat template (pinned)", "url": "https://huggingface.co/tencent/Hy4-preview/blob/705d81ee51566a186d645b74c974d642ef2828fe/chat_template.jinja", "type": "primary" }, { "label": "Tencent Cloud ADP introduces Hy4 Preview", "url": "https://adp.tencent.com/zh/blog/adp-hunyuan-hy4-preview-support", "type": "primary" }, { "label": "Hugging Face weights", "url": "https://huggingface.co/tencent/Hy4-preview", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/tencent/hy4-preview.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Tencent reports nominal backbone-only counts of 770B total and 49B active parameters. A separate native MTP layer is reported as 10B total and 0.7B active for speculative decoding; no exact complete-checkpoint count or combined active count is claimed. The original BF16 checkpoint is distinct from Hy4 preview-FP8 and has no configured vision module. Its configuration permits 1,048,576 positions; Tencent ADP advertises a hosted 1,024,000-token context. The August 28, 2026 release announcement explicitly establishes public availability.", "parameters_scope": "language_backbone", "active_parameters_scope": "language_backbone", "parameter_components_b": { "language_backbone": 770, "mtp": 10 }, "active_parameters_by_phase_b": { "language_backbone": 49, "mtp": 0.7 }, "field_status": {}, "audited_at": "2026-10-27T04:53:38.741962Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/505.json" }, { "id": "inclusionai/ling-3.0-flash-fin", "slug": "inclusionai--ling-3.0-flash-fin", "name": "Ling 3.0 Flash Fin", "developer": "InclusionAI", "developer_id": "inclusionai", "family": "ling", "description": "Finance-enhanced model for financial research, multi-step investment workflows, and long-horizon planning and execution", "release_date": "2026-08-27", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 127.486384608, "active_parameters_b": 5.1, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Fin", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Fin", "type": "weights" }, { "label": "Official Fin configuration", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Fin/blob/main/config.json", "type": "primary" }, { "label": "Official Fin chat template", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Fin/blob/main/chat_template.jinja", "type": "primary" }, { "label": "Official Fin launch announcement", "url": "https://developer.ant-ling.com/zh-CN/blogs/ling-3.0-flash-fin", "type": "primary" }, { "label": "Official Fin initial announcement timestamp", "url": "https://cdn.syndication.twimg.com/tweet-result?id=2093022087069958492&token=0", "type": "primary" }, { "label": "Official Fin open-weight release", "url": "https://developer.ant-ling.com/zh-CN/blogs/ling-3.0-flash-fin-finfirst", "type": "primary" }, { "label": "Fin learned parameters and buffer implementation", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Fin/blob/a5b4afb48f526175991724472c179e5a065604de/modeling_bailing_moe_v3.py", "type": "primary" }, { "label": "Fin complete tensor index", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Fin/blob/a5b4afb48f526175991724472c179e5a065604de/model.safetensors.index.json", "type": "primary" }, { "label": "Fin auxiliary MTP checkpoint header", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Fin/resolve/a5b4afb48f526175991724472c179e5a065604de/model-mtp-00001-of-00001.safetensors", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/inclusionai/ling-3.0-flash-fin.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The full BF16 checkpoint contains 127,486,384,608 unique learned parameters: 124,414,191,072 backbone plus 3,072,193,536 auxiliary MTP. All 65 shard headers were counted; 20,992 registered routing expert_bias buffer elements were excluded. F32 KDA A_log and dt_bias are learned parameters and remain included. Input embeddings and lm_head are explicitly untied, each counted once; MTP reuses both without extra copies. The official 124B total is a nominal backbone label. The official 5.1B activated figure is retained as nominal, with no inferred MTP/decoding-phase count. Architecture is hybrid-linear MoE with KDA and gated MLA in a 5:1 arrangement. Initial public introduction was August 27, 2026 UTC; checkpoint uploads followed September 3, with the open-weight announcement September 4.", "parameter_components_b": { "backbone": 124.414191072, "mtp": 3.072193536 }, "field_status": {}, "audited_at": "2026-10-26T20:58:07.399140Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/243.json" }, { "id": "alibaba/qwen3.8-flash", "slug": "alibaba--qwen3.8-flash", "name": "Qwen3.8 Flash", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Qwen vision-language model for visual reasoning, documents, and agent tasks", "release_date": "2026-08-26", "date_precision": "day", "open_weights": true, "license": "qwen-community-1.0", "parameters_b": 125, "active_parameters_b": 6, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text", "image", "video", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://help.aliyun.com/en/model-studio/qwen3-8-flash", "source_type": "primary", "sources": [ { "label": "Qwen3.8-Flash API specifications", "url": "https://help.aliyun.com/en/model-studio/qwen3-8-flash", "type": "primary" }, { "label": "QwenCloud Qwen3.8-Flash guide", "url": "https://docs.qwencloud.com/developer-guides/getting-started/latest-model", "type": "primary" }, { "label": "Alibaba Qwen3.8-Flash announcement", "url": "https://www.alibabacloud.com/blog/alibaba-releases-qwen3-8-flash-with-innovative-model-architecture-delivering-optimal-price-performance_603503", "type": "primary" }, { "label": "Flash-Next model card and hosted-version mapping", "url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next", "type": "primary" }, { "label": "Model Studio release history", "url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "type": "primary" }, { "label": "Released weight counterpart license", "url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next/raw/de4b8e4d43b917e7706784d8bb445c9af86a3540/LICENSE", "type": "primary" }, { "label": "Alibaba exact-model PDF input support", "url": "https://www.alibabacloud.com/help/en/model-studio/pdf-understanding", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.8-flash.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "125B is the nominal main model count, with 51B N-gram embeddings separately disclosed and 6B nominal active main-model parameters. Full hosted deployment total is unspecified. Alibaba calls Flash open-weight and explicitly links Flash-Next as the released weight counterpart under Qwen Community License 1.0; its card maps hosted Flash as the production version with added features. Exact hosted internals and the released checkpoint are not asserted identical; Flash-Next MTP counts are not imported. API service terms are separate from the weight license. PDF input is supported in Beijing/Singapore via Chat Completions/DashScope, not Responses.", "parameters_scope": "main_model", "parameter_components_b": { "main_model": 125, "ngram_embeddings": 51 }, "active_parameters_scope": "main_model", "field_status": {}, "audited_at": "2026-10-26T15:53:40.034719+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/068.json" }, { "id": "alibaba/qwen3.8-flash-next", "slug": "alibaba--qwen3.8-flash-next", "name": "Qwen3.8 Flash Next", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Open-weight multimodal MoE preview of Qwen4, combining Gated DeltaNet and Qwen Sparse Attention with a vision encoder, n-gram embeddings, and MTP. The complete BF16 checkpoint contains about 180B parameters; the main language model activates a nominal 6B per token.", "release_date": "2026-08-26", "date_precision": "day", "open_weights": true, "license": "qwen-community-1.0", "parameters_b": 179.999981424, "active_parameters_b": 6, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/QwenLM/Qwen3.8-Flash-Next", "source_type": "primary", "sources": [ { "label": "Qwen3.8-Flash-Next official repository and release news", "url": "https://github.com/QwenLM/Qwen3.8-Flash-Next", "type": "primary" }, { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next", "type": "weights" }, { "label": "Qwen3.8-Flash-Next checkpoint configuration", "url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next/raw/de4b8e4d43b917e7706784d8bb445c9af86a3540/config.json", "type": "weights" }, { "label": "Official BF16 checkpoint index and indexed safetensors headers", "url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next/resolve/de4b8e4d43b917e7706784d8bb445c9af86a3540/model.safetensors.index.json", "type": "weights" }, { "label": "Hugging Face official repository metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen3.8-Flash-Next", "type": "weights" }, { "label": "Qwen Community License 1.0", "url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next/raw/main/LICENSE", "type": "weights" }, { "label": "Qwen3.8-Flash-Next official chat template", "url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next/blob/main/chat_template.jinja", "type": "weights" }, { "label": "Qwen3.8-Flash-Next: A New Architecture, Towards Ultimate Cost-Efficiency", "url": "https://www.alibabacloud.com/blog/qwen3-8-flash-next-a-new-architecture-towardsultimate-cost-efficiency_603501", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.8-flash-next.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete checkpoint count is 179,999,981,424 BF16 parameters, excluding 35 I64 indexing values. Official nominal 125B main + 51B n-gram + 4B MTP labels are component descriptions, not an exact independent-tensor breakdown. Stored MTP tensors total 2.607150848B and configuration disables dedicated MTP embeddings, so shared tensors are counted once. The nominal 6B active count concerns language-model computation and excludes n-gram lookup storage. Native context is 262144; hosted Qwen3.8-Flash adds a default 1M context. August 26 is the explicit official release date; August 27 is the Cloud article date.", "parameter_components_b": { "main_language_model": 125.71081408, "ngram_path": 51.23308544, "mtp_stored": 2.607150848, "vision": 0.448931056 }, "active_parameters_scope": "main_language_model", "field_status": {}, "audited_at": "2026-10-26T15:39:06.267709Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/069.json" }, { "id": "ibm/granite-4.2-8b", "slug": "ibm--granite-4.2-8b", "name": "Granite 4.2 8B", "developer": "IBM", "developer_id": "ibm", "family": "granite", "description": "IBM text model with a dense decoder-only Transformer, configurable thinking modes, and native tool calling.", "release_date": "2026-08-25", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 8.79159296, "active_parameters_b": 8, "parameters_source": "primary_source", "architecture": "dense", "context_window": 524288, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/ibm-granite/granite-4.2-8b", "source_type": "primary", "sources": [ { "label": "IBM Granite 4.2 8B official model card", "url": "https://huggingface.co/ibm-granite/granite-4.2-8b", "type": "primary" }, { "label": "Granite 4.2 8B structured model disclosure", "url": "https://raw.githubusercontent.com/ibm-granite/granite-4.2-language-models/main/disclosures/granite_4.2_8b_disclosure.json", "type": "primary" }, { "label": "Granite 4.2 8B checkpoint configuration", "url": "https://huggingface.co/ibm-granite/granite-4.2-8b/raw/a7042144125ce5ae64f89610e8289647fc000b9a/config.json", "type": "primary" }, { "label": "Granite 4.2 brings native reasoning to enterprise agents", "url": "https://research.ibm.com/blog/introducing-granite-4-2", "type": "primary" }, { "label": "Granite 4.2 LLMs: How They're Built", "url": "https://huggingface.co/blog/ibm-granite/granite-4-2", "type": "primary" }, { "label": "Granite 4.2 8B complete checkpoint tensor index", "url": "https://huggingface.co/ibm-granite/granite-4.2-8b/raw/a7042144125ce5ae64f89610e8289647fc000b9a/model.safetensors.index.json", "type": "primary" }, { "label": "Official IBM checkpoint Hub metadata", "url": "https://huggingface.co/api/models/ibm-granite/granite-4.2-8b", "type": "primary" }, { "label": "IBM Granite clarification of 128K versus 512K context", "url": "https://huggingface.co/ibm-granite/granite-4.2-3b/discussions/3", "type": "primary" }, { "label": "Granite 4.2 8B model-00001-of-00004.safetensors tensor header", "url": "https://huggingface.co/ibm-granite/granite-4.2-8b/resolve/a7042144125ce5ae64f89610e8289647fc000b9a/model-00001-of-00004.safetensors", "type": "primary" }, { "label": "Granite 4.2 8B model-00002-of-00004.safetensors tensor header", "url": "https://huggingface.co/ibm-granite/granite-4.2-8b/resolve/a7042144125ce5ae64f89610e8289647fc000b9a/model-00002-of-00004.safetensors", "type": "primary" }, { "label": "Granite 4.2 8B model-00003-of-00004.safetensors tensor header", "url": "https://huggingface.co/ibm-granite/granite-4.2-8b/resolve/a7042144125ce5ae64f89610e8289647fc000b9a/model-00003-of-00004.safetensors", "type": "primary" }, { "label": "Granite 4.2 8B model-00004-of-00004.safetensors tensor header", "url": "https://huggingface.co/ibm-granite/granite-4.2-8b/resolve/a7042144125ce5ae64f89610e8289647fc000b9a/model-00004-of-00004.safetensors", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/ibm/granite-4.2-8b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete BF16 checkpoint: 8,791,592,960 learned parameters, including separate input/output embeddings. IBM publishes approximate total and active counts of 8B; active 8B is approximate, not an exact equality with the complete count. The index byte total is inconsistent; all four tensor headers and Hub metadata agree on the exact count. IBM supports a 524288-token context extension with a serving-flag change and no RoPE scaling; default and evaluated length remain 131072, with evaluation beyond 128K not established. Dense Transformer, without Mamba or MoE.", "field_status": {}, "audited_at": "2026-10-26T20:48:02.467632Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/237.json" }, { "id": "apodex/apodex-1.1-mini", "slug": "apodex--apodex-1.1-mini", "name": "Apodex 1.1 Mini", "developer": "Apodex", "developer_id": "apodex", "family": "apodex", "description": "Open-weight Qwen3.5-based MoE model with thinking, function calling and a 262,144-token context window.", "release_date": "2026-08-24", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 35.951822704, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/apodex/Apodex-1.1-mini", "source_type": "primary", "sources": [ { "label": "Apodex 1.1: Scaling Agentic Intelligence for Complex Work", "url": "https://www.apodex.com/blog/apodex-1.1-scaling-agentic-intelligence-for-complex-work", "type": "primary" }, { "label": "Apodex-1.1-mini model card", "url": "https://huggingface.co/apodex/Apodex-1.1-mini/blob/4e4f109a5e32d1306dc6c75e4995a09c8b881811/README.md", "type": "primary" }, { "label": "Apodex-1.1-mini checkpoint configuration", "url": "https://huggingface.co/apodex/Apodex-1.1-mini/blob/4e4f109a5e32d1306dc6c75e4995a09c8b881811/config.json", "type": "primary" }, { "label": "Apodex-1.1-mini Apache License", "url": "https://huggingface.co/apodex/Apodex-1.1-mini/blob/4e4f109a5e32d1306dc6c75e4995a09c8b881811/LICENSE", "type": "primary" }, { "label": "Apodex-1.1-mini repository metadata", "url": "https://huggingface.co/api/models/apodex/Apodex-1.1-mini", "type": "primary" }, { "label": "Apodex-1.1-mini weight index and all 15 safetensors headers", "url": "https://huggingface.co/apodex/Apodex-1.1-mini/blob/4e4f109a5e32d1306dc6c75e4995a09c8b881811/model.safetensors.index.json", "type": "primary" }, { "label": "Apodex-1.1-mini chat template", "url": "https://huggingface.co/apodex/Apodex-1.1-mini/blob/4e4f109a5e32d1306dc6c75e4995a09c8b881811/chat_template.jinja", "type": "primary" }, { "label": "Core Models (LLM API)", "url": "https://platform.apodex.ai/docs/models", "type": "primary" }, { "label": "Apodex-1.1-mini repository history", "url": "https://huggingface.co/api/models/apodex/Apodex-1.1-mini/commits/main", "type": "primary" }, { "label": "Apodex 1.1 technical report", "url": "https://arxiv.org/pdf/2608.23283", "type": "primary" }, { "label": "Apodex-1.1-mini image processor configuration", "url": "https://huggingface.co/apodex/Apodex-1.1-mini/blob/4e4f109a5e32d1306dc6c75e4995a09c8b881811/preprocessor_config.json", "type": "primary" }, { "label": "Apodex-1.1-mini video processor configuration", "url": "https://huggingface.co/apodex/Apodex-1.1-mini/blob/4e4f109a5e32d1306dc6c75e4995a09c8b881811/video_preprocessor_config.json", "type": "primary" }, { "label": "Apodex official Mini GGUF deployment instructions", "url": "https://huggingface.co/apodex/Apodex-1.1-mini-GGUF/blob/main/README.md", "type": "primary" }, { "label": "Qwen3.5-35B-A3B model card", "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B/blob/59d61f3ce65a6d9863b86d2e96597125219dc754/README.md", "type": "primary" }, { "label": "Qwen3.5-35B-A3B configuration", "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B/blob/59d61f3ce65a6d9863b86d2e96597125219dc754/config.json", "type": "primary" }, { "label": "Qwen3.5-35B-A3B index and all shard headers", "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B/blob/59d61f3ce65a6d9863b86d2e96597125219dc754/model.safetensors.index.json", "type": "primary" }, { "label": "Qwen3.5 MoE reference implementation", "url": "https://github.com/huggingface/transformers/blob/main/src/transformers/models/qwen3_5_moe/modeling_qwen3_5_moe.py", "type": "primary" }, { "label": "Qwen3.5 MTP implementation", "url": "https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/models/qwen3_5_mtp.py", "type": "primary" }, { "label": "Official Apodex 1.1 release post (X syndication)", "url": "https://cdn.syndication.twimg.com/tweet-result?id=2091916791308313018&lang=en&token=0", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/apodex/apodex-1.1-mini.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/apodex/Apodex-1.1-mini", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Full checkpoint has 35.951822704B parameters, including vision and MTP. Active 3B is the base developer's nominal language-model count, supported by identical configuration and tensor shapes; it excludes vision and MTP. The local model accepts images, while the hosted API is text-only. August 24 is the public introduction date; weights were uploaded August 25, and exact public download timing was not independently established.", "parameter_components_b": { "language_model": 34.660610688, "vision_encoder": 0.446571248, "mtp": 0.844640768 }, "active_parameters_scope": "language_model", "field_status": {}, "audited_at": "2026-10-26T17:22:18.752268Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/117.json" }, { "id": "deepseek/deepseek-v4-flash-vision-exp", "slug": "deepseek--deepseek-v4-flash-vision-exp", "name": "DeepSeek V4 Flash Vision Exp", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-flash", "description": "Experimental multimodal DeepSeek V4 Flash model for image understanding, coding, and agentic work", "release_date": "2026-08-21", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 304.644497086, "active_parameters_b": 13, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://api-docs.deepseek.com/news/news260821/", "source_type": "primary", "sources": [ { "label": "DeepSeek-V4-Flash-Vision-Exp model card", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp/blob/6821d6ad3681a4b137b066b76094fa82ebd0a380/README.md", "type": "weights" }, { "label": "August 21 experimental vision API announcement", "url": "https://api-docs.deepseek.com/news/news260821/", "type": "primary" }, { "label": "DeepSeek API changelog", "url": "https://api-docs.deepseek.com/updates/", "type": "primary" }, { "label": "Exact checkpoint configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp/blob/6821d6ad3681a4b137b066b76094fa82ebd0a380/config.json", "type": "weights" }, { "label": "Exact checkpoint index and bounded safetensors headers", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp/blob/6821d6ad3681a4b137b066b76094fa82ebd0a380/model.safetensors.index.json", "type": "weights" }, { "label": "Official vision model and DSpark reference implementation", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp/blob/6821d6ad3681a4b137b066b76094fa82ebd0a380/inference/model.py", "type": "weights" }, { "label": "Official quantized checkpoint conversion", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp/blob/6821d6ad3681a4b137b066b76094fa82ebd0a380/inference/convert.py", "type": "weights" }, { "label": "Official vision encoder and aligner", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp/blob/6821d6ad3681a4b137b066b76094fa82ebd0a380/inference/vision.py", "type": "weights" }, { "label": "Exact checkpoint MIT license", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp/blob/6821d6ad3681a4b137b066b76094fa82ebd0a380/LICENSE", "type": "weights" }, { "label": "Original Flash language configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash/blob/60d8d70770c6776ff598c94bb586a859a38244f1/config.json", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepseek/deepseek-v4-flash-vision-exp.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released on the API August 21; open weights followed August 31. The complete count includes language, vision and DSpark components, expands packed FP4, excludes quantization scales and fixed routing indices, and counts shared weights once. The nominal 13B active language count is inferred from the matching V4 Flash architecture; it excludes vision encoding and DSpark drafting. Context is the checkpoint’s configured positional limit. The original API identifier now redirects to V4.1 Flash; this record preserves the original checkpoint.", "active_parameters_scope": "language_backbone", "parameter_components_b": { "language_backbone": 284.332252247, "dspark": 19.845851751, "vision_encoder": 0.41184256, "vision_aligner": 0.054534144, "image_embeddings": 0.000016384 }, "field_status": {}, "audited_at": "2026-10-26T19:16:39.554392Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/180.json" }, { "id": "zhipuai/glm-5.3-flash", "slug": "zhipuai--glm-5.3-flash", "name": "GLM-5.3-Flash", "developer": "Z.ai", "developer_id": "zhipuai", "family": "glm-flash", "description": "Multimodal MoE model combining linear and sparse attention, with reasoning and tool calling.", "release_date": "2026-08-20", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 320, "active_parameters_b": 18, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://docs.z.ai/guides/vlm/glm-5.3-flash", "source_type": "primary", "sources": [ { "label": "GLM-5.3-Flash exact official model card", "url": "https://huggingface.co/zai-org/GLM-5.3-Flash/blob/46a050bec13002ad29ee795c16e61cf4fa3c4a9c/README.md", "type": "weights" }, { "label": "GLM-5.3-Flash pinned checkpoint configuration", "url": "https://huggingface.co/zai-org/GLM-5.3-Flash/blob/46a050bec13002ad29ee795c16e61cf4fa3c4a9c/config.json", "type": "weights" }, { "label": "GLM-5.3-Flash tokenizer configuration", "url": "https://huggingface.co/zai-org/GLM-5.3-Flash/blob/46a050bec13002ad29ee795c16e61cf4fa3c4a9c/tokenizer_config.json", "type": "docs" }, { "label": "GLM-5.3-Flash checkpoint MIT license", "url": "https://huggingface.co/zai-org/GLM-5.3-Flash/blob/46a050bec13002ad29ee795c16e61cf4fa3c4a9c/LICENSE", "type": "weights" }, { "label": "GLM-5.3-Flash exact chat template", "url": "https://huggingface.co/zai-org/GLM-5.3-Flash/blob/46a050bec13002ad29ee795c16e61cf4fa3c4a9c/chat_template.jinja", "type": "docs" }, { "label": "GLM-5.3-Flash native API guide", "url": "https://docs.z.ai/guides/vlm/glm-5.3-flash", "type": "docs" }, { "label": "Ox Alpha historical public preview page", "url": "https://openrouter.ai/stealth/ox-alpha", "type": "announcement" }, { "label": "Z.ai dated GLM-5.3-Flash release notes", "url": "https://docs.z.ai/release-notes/new-released.md", "type": "announcement" }, { "label": "Pinned official checkpoint metadata", "url": "https://huggingface.co/api/models/zai-org/GLM-5.3-Flash/revision/46a050bec13002ad29ee795c16e61cf4fa3c4a9c", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/zhipuai/glm-5.3-flash.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/GLM-5.3-Flash", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "First public preview: Ox Alpha on August 20, 2026, directly mapped by Z.ai and the original host. Named release followed August 26; final checkpoint specifications do not imply identical preview weights. Published 320B total and 18B active counts are nominal, with vision, embedding/output and MTP inclusion unspecified. The checkpoint combines MoE with 34 linear and 11 sparse-attention layers, mHC and a vision encoder. Exact model/tokenizer context is 1048576, advertised as 1M. Native thinking is always enabled with low/high/max effort. PDF is treated as a file container for text and visual content, not a separate modality.", "field_status": {}, "audited_at": "2026-10-27T06:22:38.631501+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/562.json" }, { "id": "deepreinforce/ornith-1.5-35b-a3b", "slug": "deepreinforce--ornith-1.5-35b-a3b", "name": "Ornith 1.5 35B A3B", "developer": "DeepReinforce", "developer_id": "deepreinforce", "family": "ornith", "description": "Mixture-of-experts model with hybrid linear/full attention, trained through self-improvement for agentic coding, reasoning and tool use; accepts text, images and video.", "release_date": "2026-08-18", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 35.951822704, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B/resolve/10fbf86fed7ecee4a061f8b499a618f46001cac1/README.md", "source_type": "primary", "sources": [ { "label": "Ornith-1.5-35B-A3B model card at audited revision", "url": "https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B/resolve/10fbf86fed7ecee4a061f8b499a618f46001cac1/README.md", "type": "primary" }, { "label": "Ornith-1.5: From Self-Scaffolding to Self-Improvement", "url": "https://ornith.ai/ornith_1_5.html", "type": "primary" }, { "label": "Exact Ornith-1.5-35B-A3B checkpoint configuration", "url": "https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B/resolve/10fbf86fed7ecee4a061f8b499a618f46001cac1/config.json", "type": "weights" }, { "label": "Exact checkpoint index and all 16 indexed safetensors headers", "url": "https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B/resolve/10fbf86fed7ecee4a061f8b499a618f46001cac1/model.safetensors.index.json", "type": "weights" }, { "label": "Exact checkpoint image and video processor configuration", "url": "https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B/resolve/10fbf86fed7ecee4a061f8b499a618f46001cac1/processor_config.json", "type": "weights" }, { "label": "Exact checkpoint chat template", "url": "https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B/resolve/10fbf86fed7ecee4a061f8b499a618f46001cac1/chat_template.jinja", "type": "weights" }, { "label": "Official model repository commit chronology", "url": "https://huggingface.co/api/models/ornith-ai/Ornith-1.5-35B-A3B/commits/main", "type": "primary" }, { "label": "First complete weight-upload tree", "url": "https://huggingface.co/api/models/ornith-ai/Ornith-1.5-35B-A3B/tree/a111f308b7a9fc99010afd1d412ae92969f854ad?recursive=true", "type": "weights" }, { "label": "Official Ornith organization profile", "url": "https://huggingface.co/ornith-ai", "type": "primary" }, { "label": "Transformers implementation for the declared architecture", "url": "https://raw.githubusercontent.com/huggingface/transformers/v5.8.1/src/transformers/models/qwen3_5_moe/modeling_qwen3_5_moe.py", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepreinforce/ornith-1.5-35b-a3b.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete BF16 checkpoint contains 35,951,822,704 learned parameters: 34,660,610,688 language/output-head, 446,571,248 vision and 844,640,768 auxiliary MTP. The ordinary language-plus-vision model excluding MTP has 35,107,181,936 parameters; standard Transformers ignores MTP checkpoint keys. Publisher 35B total and 3B active are nominal; 3B is not an exact all-component activation count. Architecture combines MoE feed-forward blocks with linear/full attention. Native context is 262144; larger windows require YaRN scaling.", "field_status": {}, "audited_at": "2026-10-26T18:51:18.105211+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/165.json" }, { "id": "alibaba/qwen3.8-27b", "slug": "alibaba--qwen3.8-27b", "name": "Qwen3.8 27B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Dense Qwen3.8 vision-language model with hybrid Gated DeltaNet and gated attention, configurable reasoning, and tool calling.", "release_date": "2026-08-14", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 27.781427952, "active_parameters_b": 27, "parameters_source": "primary_source", "architecture": "dense", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.8-27b.toml", "source_type": "catalog", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3.8-27B", "type": "weights" }, { "label": "Qwen3.8 official repository and release news", "url": "https://github.com/QwenLM/Qwen3.8", "type": "primary" }, { "label": "Qwen3.8-27B LICENSE", "url": "https://huggingface.co/Qwen/Qwen3.8-27B/raw/main/LICENSE", "type": "primary" }, { "label": "Qwen3.8-27B configuration", "url": "https://huggingface.co/Qwen/Qwen3.8-27B/raw/main/config.json", "type": "primary" }, { "label": "Qwen3.8-27B chat template", "url": "https://huggingface.co/Qwen/Qwen3.8-27B/raw/main/chat_template.jinja", "type": "primary" }, { "label": "Pinned original BF16 checkpoint index and its 18 referenced tensor headers", "url": "https://huggingface.co/Qwen/Qwen3.8-27B/resolve/1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0/model.safetensors.index.json", "type": "primary" }, { "label": "Hugging Face metadata for the official Qwen checkpoint", "url": "https://huggingface.co/api/models/Qwen/Qwen3.8-27B", "type": "primary" }, { "label": "Qwen3.5 dense architecture implementation used by Qwen3.8-27B", "url": "https://raw.githubusercontent.com/huggingface/transformers/main/src/transformers/models/qwen3_5/modeling_qwen3_5.py", "type": "primary" }, { "label": "Qwen official parameter comparison table", "url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next/raw/de4b8e4d43b917e7706784d8bb445c9af86a3540/README.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.8-27b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete BF16 checkpoint: 27.781427952B parameters, including 26.895998464B language model plus output head, 0.460730096B vision, and 0.424699392B MTP. The published 27B is nominal. A later official comparison table explicitly reports 27B nominal active language-model parameters; this excludes an asserted all-modality/MTP execution total. Native context is 262,144 tokens; extension to 1,000,000 is documented.", "parameter_components_b": { "language_model": 26.895998464, "vision": 0.460730096, "mtp": 0.424699392 }, "active_parameters_scope": "language_model", "field_status": {}, "audited_at": "2026-10-26T15:44:56.658841+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/067.json" }, { "id": "zhipuai/glm-5.3", "slug": "zhipuai--glm-5.3", "name": "GLM-5.3", "developer": "Z.ai", "developer_id": "zhipuai", "family": "glm", "description": "Text MoE model with sparse attention, adjustable reasoning effort and function calling.", "release_date": "2026-08-14", "date_precision": "day", "open_weights": true, "license": "GLM-5.3 License", "parameters_b": 744, "active_parameters_b": 40, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://z.ai/blog/glm-5.3", "source_type": "primary", "sources": [ { "label": "Original GLM-5.3 release announcement", "url": "https://z.ai/blog/glm-5.3", "type": "announcement" }, { "label": "GLM-5.3 pinned model card", "url": "https://huggingface.co/zai-org/GLM-5.3/blob/ba32f8c19c8738e2c57c71798c601d477a7bf844/README.md", "type": "docs" }, { "label": "GLM-5.3 pinned checkpoint configuration", "url": "https://huggingface.co/zai-org/GLM-5.3/blob/ba32f8c19c8738e2c57c71798c601d477a7bf844/config.json", "type": "docs" }, { "label": "GLM-5.3 License", "url": "https://huggingface.co/zai-org/GLM-5.3/blob/ba32f8c19c8738e2c57c71798c601d477a7bf844/LICENSE", "type": "docs" }, { "label": "Original developer model download table", "url": "https://github.com/zai-org/GLM-5/blob/83c03d317378f86f0a012f7c4cd48b6243c7e69b/README.md", "type": "docs" }, { "label": "GLM-5.3 native API guide", "url": "https://docs.z.ai/guides/llm/glm-5.3", "type": "docs" }, { "label": "GLM-5.3 pinned chat template", "url": "https://huggingface.co/zai-org/GLM-5.3/blob/ba32f8c19c8738e2c57c71798c601d477a7bf844/chat_template.jinja", "type": "docs" }, { "label": "Original vLLM GLM-5.3 serving recipe", "url": "https://recipes.vllm.ai/zai-org/GLM-5.3", "type": "docs" }, { "label": "Native model release notes", "url": "https://docs.z.ai/release-notes/new-released.md", "type": "docs" }, { "label": "Pinned GLM-5.3 public repository metadata", "url": "https://huggingface.co/api/models/zai-org/GLM-5.3/revision/ba32f8c19c8738e2c57c71798c601d477a7bf844", "type": "docs" }, { "label": "GLM-5.3 pinned tokenizer configuration", "url": "https://huggingface.co/zai-org/GLM-5.3/blob/ba32f8c19c8738e2c57c71798c601d477a7bf844/tokenizer_config.json", "type": "docs" }, { "label": "GLM-5 technical report referenced by the current card", "url": "https://arxiv.org/html/2602.15763v2", "type": "docs" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/zhipuai/glm-5.3.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/GLM-5.3", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released through GLM Coding Plan and ZCode on August 14, 2026; native API release notes are dated August 18 and public weights followed later. The developer lists nominal 744B total and 40B active; the linked vLLM recipe instead gives approximately 743B/39B. The developer figures are retained without claiming exact learned-tensor totals or importing an earlier model's embedding/MTP accounting. The checkpoint uses MoE with latent/sparse attention and shared sparse indexers. Context capacity is 1048576 in the checkpoint and tokenizer, while the hosted guide advertises 1M. Reasoning is always enabled with low/high/max effort. Weights use the custom GLM-5.3 License, not MIT.", "field_status": {}, "audited_at": "2026-10-27T06:22:57.335509+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/561.json" }, { "id": "deepseek/deepseek-v4-pro-0813", "slug": "deepseek--deepseek-v4-pro-0813", "name": "DeepSeek V4 Pro 0813", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-thinking", "description": "August 2026 general-availability DeepSeek V4 Pro MoE release with DSpark speculative decoding, million-token context, and thinking and non-thinking modes.", "release_date": "2026-08-13", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 1650.495609866, "active_parameters_b": 49, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813", "source_type": "primary", "sources": [ { "label": "DeepSeek-V4-Pro-0813 official model card", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813/blob/72e1d3230f6c080a530b0a1d46f8eb4602340597/README.md", "type": "primary" }, { "label": "DeepSeek API change log", "url": "https://api-docs.deepseek.com/updates/", "type": "primary" }, { "label": "0813 checkpoint index and indexed safetensors headers", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813/resolve/72e1d3230f6c080a530b0a1d46f8eb4602340597/model.safetensors.index.json", "type": "weights" }, { "label": "0813 reference Transformer and DSpark implementation", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813/blob/72e1d3230f6c080a530b0a1d46f8eb4602340597/inference/model.py", "type": "primary" }, { "label": "0813 checkpoint conversion and FP4 unpacking", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813/blob/72e1d3230f6c080a530b0a1d46f8eb4602340597/inference/convert.py", "type": "primary" }, { "label": "0813 model configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813/blob/72e1d3230f6c080a530b0a1d46f8eb4602340597/config.json", "type": "weights" }, { "label": "0813 reference inference configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813/blob/72e1d3230f6c080a530b0a1d46f8eb4602340597/inference/config.json", "type": "weights" }, { "label": "DeepSeek-V4 technical report", "url": "https://arxiv.org/html/2606.19348v1", "type": "primary" }, { "label": "DeepSeek API models and pricing", "url": "https://api-docs.deepseek.com/quick_start/pricing/?model_version=DeepSeek-V4-Pro-0813", "type": "primary" }, { "label": "0813 prompt and output encoding", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813/blob/72e1d3230f6c080a530b0a1d46f8eb4602340597/encoding/README.md", "type": "primary" }, { "label": "0813 MIT License", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813/blob/72e1d3230f6c080a530b0a1d46f8eb4602340597/LICENSE", "type": "primary" }, { "label": "August13 GA announcement", "url": "https://api-docs.deepseek.com/news/news260813/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepseek/deepseek-v4-pro-0813.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete checkpoint total is 1,650.495609866B logical parameters, including 1,572.997201763B main model and 77.498408103B DSpark. Quantization scales and 2,327,040 integer hash-routing entries are excluded; FP4 packed weights are expanded logically and the shared main embedding/head are counted once. The published 49B activated count is nominal for the main model, not total DSpark-assisted execution. Advertised context is 1M; checkpoint configuration allows 1,048,576 positions.", "parameter_components_b": { "main": 1572.997201763, "dspark": 77.498408103 }, "active_parameters_scope": "main", "field_status": {}, "audited_at": "2026-10-26T19:22:08.860479Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/183.json" }, { "id": "google/gemini-3.7-flash", "slug": "google--gemini-3.7-flash", "name": "Gemini 3.7 Flash", "developer": "Google", "developer_id": "google", "family": "gemini-flash", "description": "Gemini 3-series model with text, image, video, audio and PDF input, text output, configurable thinking, and function calling.", "release_date": "2026-08-13", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-3-7-flash/", "source_type": "primary", "sources": [ { "label": "Gemini 3.7 Flash API model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.7-flash", "type": "primary" }, { "label": "Gemini 3.7 Flash model card", "url": "https://deepmind.google/models/model-cards/gemini-3-7-flash/", "type": "primary" }, { "label": "Introducing Gemini 3.7 Flash", "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-3-7-flash/", "type": "primary" }, { "label": "Gemini 3.6 Flash model card", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-6-Flash-Model-Card.pdf", "type": "primary" }, { "label": "Gemini 3.5 Flash model card", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-5-Flash-Model-Card.pdf", "type": "primary" }, { "label": "Gemini 3 Flash model card", "url": "https://deepmind.google/models/model-cards/gemini-3-flash/", "type": "primary" }, { "label": "Gemini 3 Pro model card, updated May 2026", "url": "https://deepmind.google/models/model-cards/gemini-3-pro/", "type": "primary" }, { "label": "Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Gemini API release notes", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "primary" }, { "label": "Developer's guide to Gemini 3.7 Flash", "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/guides/gemini-3-7-flash", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemini-3.7-flash.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Sparse MoE Transformer based on Gemini 3.6 Flash, through an explicit lineage to Gemini 3 Pro. Total and active parameter counts are undisclosed for this model and its named bases. Google deprecated gemini-3.7-flash on October 8, 2026 and routes this ID to gemini-3.8-flash; listed capabilities describe the original 3.7 Flash model.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T13:06:40.682772Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/211.json" }, { "id": "mixedbread/toast-1", "slug": "mixedbread--toast-1", "name": "Toast 1", "developer": "Mixedbread", "developer_id": "mixedbread", "family": null, "description": "Specialized search model for knowledge-intensive questions, multi-step retrieval, and evidence synthesis", "release_date": "2026-08-13", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://www.mixedbread.com/blog/toast-1", "source_type": "primary", "sources": [ { "label": "Toast 1 release announcement", "url": "https://www.mixedbread.com/blog/toast-1", "type": "primary" }, { "label": "Toast 1 model card", "url": "https://www.mixedbread.com/docs/agent/models", "type": "primary" }, { "label": "Toast 1 exact context and thinking policy", "url": "https://www.mixedbread.com/docs/agent/build-your-own-harness", "type": "primary" }, { "label": "Toast 1 Chat Completions documentation", "url": "https://www.mixedbread.com/docs/agent/chat-completions", "type": "primary" }, { "label": "Mixedbread service terms", "url": "https://www.mixedbread.com/pages/terms", "type": "primary" }, { "label": "Mixedbread public model inventory", "url": "https://huggingface.co/api/models?author=mixedbread-ai&limit=100", "type": "primary" }, { "label": "Mixedbread research index", "url": "https://www.mixedbread.com/research", "type": "primary" }, { "label": "Separate Toast retrieval harness", "url": "https://github.com/mixedbread-ai/toast-harness", "type": "primary" }, { "label": "Toast 1 first-party comparison", "url": "https://www.mixedbread.com/evals/toast-1-browsecomp-plus", "type": "primary" }, { "label": "Toast 1 release changelog", "url": "https://www.mixedbread.com/docs/changelog", "type": "primary" }, { "label": "OfficeQA-Pro benchmark · Toast 1", "url": "https://www.mixedbread.com/evals/toast-1-officeqa-pro", "type": "primary" }, { "label": "LongSeal benchmark · Toast 1", "url": "https://www.mixedbread.com/evals/toast-1-longseal", "type": "primary" }, { "label": "OfficeQA-Pro V2 benchmark · Toast 1", "url": "https://www.mixedbread.com/evals/toast-1-officeqa-pro-v2", "type": "primary" }, { "label": "Harvey LAB Firm Knowledge benchmark · Toast 1", "url": "https://www.mixedbread.com/evals/toast-1-harvey-lab", "type": "primary" }, { "label": "Toast 1 API, Pricing & Playground", "url": "https://vercel.com/ai-gateway/models/toast-1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mixedbread/toast-1.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Toast 1 is a hosted proprietary model; its Apache-2.0 retrieval harness is separate software. The served context is 131,072 tokens. Mixedbread says it was trained and is served without thinking. Total and active parameters, neural architecture and named base-model lineage remain undisclosed in the reviewed sources.", "field_status": { "family": "undisclosed", "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T15:42:49.962378Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/353.json" }, { "id": "alibaba/qwen3.8-2.4t-a95b", "slug": "alibaba--qwen3.8-2.4t-a95b", "name": "Qwen3.8 2.4T A95B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Open-weight text-only reasoning model with hybrid Gated DeltaNet/gated attention and sparse MoE layers. The full checkpoint includes a multi-token prediction module; the hosted Qwen3.8-Max version adds further features.", "release_date": "2026-08-12", "date_precision": "day", "open_weights": true, "license": "qwen3.8-max", "parameters_b": 2446.182725504, "active_parameters_b": 95, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B/raw/207bd685a7e3696cfaff12ded7c6a7ea0f88c996/README.md", "type": "primary" }, { "label": "Official release announcement", "url": "https://github.com/QwenLM/Qwen3.8", "type": "primary" }, { "label": "Official configuration", "url": "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B/raw/207bd685a7e3696cfaff12ded7c6a7ea0f88c996/config.json", "type": "primary" }, { "label": "Official checkpoint index", "url": "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B/raw/207bd685a7e3696cfaff12ded7c6a7ea0f88c996/model.safetensors.index.json", "type": "weights" }, { "label": "Official Hub parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen3.8-2.4T-A95B", "type": "weights" }, { "label": "Official chat template", "url": "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B/raw/207bd685a7e3696cfaff12ded7c6a7ea0f88c996/chat_template.jinja", "type": "primary" }, { "label": "Model-specific license", "url": "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B/raw/207bd685a7e3696cfaff12ded7c6a7ea0f88c996/LICENSE", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.8-2.4t-a95b.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Full BF16 checkpoint: 2446.182725504B parameters, comprising decoder 2419.804697984B and auxiliary MTP 26.37802752B. The official 2.4T/95B labels are nominal language-model figures; 95B active applies to the decoder, not an asserted MTP-inclusive inference total. Hybrid Gated DeltaNet/gated-attention MoE; 262,144 native context, extendable to 1,010,000.", "active_parameters_scope": "decoder", "parameter_components_b": { "decoder": 2419.804697984, "mtp": 26.37802752 }, "field_status": {}, "audited_at": "2026-10-26T15:20:32.892738Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/066.json" }, { "id": "xai/grok-4.6", "slug": "xai--grok-4.6", "name": "Grok 4.6", "developer": "xAI", "developer_id": "xai", "family": "grok", "description": "Reasoning model developed by xAI with Cursor for coding and other knowledge work, with text and image inputs and text output.", "release_date": "2026-08-12", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 1500, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "moe", "context_window": 500000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://x.ai/news/grok-4-6", "source_type": "primary", "sources": [ { "label": "Grok 4.6 release announcement", "url": "https://x.ai/news/grok-4-6", "type": "primary" }, { "label": "Grok 4.6 model card", "url": "https://media.x.ai/v1/website/card-4p6-4cd2dc57.pdf", "type": "primary" }, { "label": "Grok 4.6 API specification", "url": "https://docs.x.ai/developers/models/grok-4.6", "type": "primary" }, { "label": "Musk exact-model size disclosure (original X payload)", "url": "https://cdn.syndication.twimg.com/tweet-result?id=2082123925283041545&lang=en&token=1", "type": "primary" }, { "label": "Cursor 4.6 same-base disclosure", "url": "https://cursor.com/help/models-and-usage/grok-4-6", "type": "primary" }, { "label": "Cursor 4.5 base architecture disclosure", "url": "https://cursor.com/blog/grok-4-5", "type": "primary" }, { "label": "Cursor 4.6 context limit", "url": "https://cursor.com/docs/models/grok-4-6", "type": "provider" }, { "label": "Enterprise model ownership terms", "url": "https://x.ai/legal/terms-of-service-enterprise", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/xai/grok-4.6.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "1500B is the nominal 1.5T scale stated in the released model card and corroborated by the founder's pre-release statement; active parameters remain undisclosed. MoE follows Cursor's explicit same-base relationship to Grok 4.5 and its architectural disclosure; expert configuration is undisclosed. Native API context is 500000 tokens; Cursor documents 256k. Released August 12, superseding the earlier August 7 target. Native text/image inputs and text output exclude file-search and generation tools.", "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T18:32:01.306757+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/533.json" }, { "id": "microsoft/mai-code-1.1-flash", "slug": "microsoft--mai-code-1.1-flash", "name": "MAI-Code-1.1-Flash", "developer": "Microsoft", "developer_id": "microsoft", "family": "mai", "description": "Microsoft's coding model with text and image input, adaptive reasoning and tool use.", "release_date": "2026-08-11", "date_precision": "day", "open_weights": true, "license": "proprietary", "parameters_b": 138, "active_parameters_b": 5, "parameters_source": "primary_source", "architecture": "moe", "context_window": 256000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://microsoft.ai/pdf/MAI-Code-1.1-Flash-Model-Card.PDF", "source_type": "primary", "sources": [ { "label": "Exact 1.1 model card", "url": "https://microsoft.ai/pdf/MAI-Code-1.1-Flash-Model-Card.PDF", "type": "primary" }, { "label": "GitHub launch announcement", "url": "https://github.blog/changelog/2026-08-11-mai-code-1-1-flash-available-in-github-copilot/", "type": "primary" }, { "label": "Local model engineering disclosure", "url": "https://commandline.microsoft.com/local-models-sandboxed-tools-github-windows/", "type": "primary" }, { "label": "VS Code on-device model documentation", "url": "https://github.com/microsoft/vscode-docs/blob/main/docs/agent-customization/language-models.md", "type": "primary" }, { "label": "Microsoft Foundry model listing", "url": "https://ai.azure.com/catalog/models/mai-code-1.1-flash", "type": "primary" }, { "label": "Microsoft instructions for MAI Code 1.1 Flash local weight download", "url": "https://raw.githubusercontent.com/microsoft/vscode-docs/main/docs/agent-customization/language-models.md", "type": "primary" }, { "label": "Microsoft public Azure ML MAI Code 1.1 Flash on-device weight registry", "url": "https://api.catalog.azureml.ms/modelregistry/v1.0/registry/models/nonazureaccount?assetId=azureml://registries/azureml/models/MAI-Code-1.1-Flash-On-Device/versions/1", "type": "primary" }, { "label": "Official MAI-Code repository discussion linking public distribution", "url": "https://github.com/microsoft/MAI-Code/issues/10", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/microsoft/mai-code-1.1-flash.toml", "type": "catalog" }, { "label": "Announcement", "url": "https://microsoft.ai/news/mai-code-1-1-flash-br-better-faster-at-a-quarter-of-the-cost/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "138B total / 5B active are the exact hosted 1.1 card figures; the later local version is reported as 137B / 6.8B, an unresolved discrepancy. The 1.1 landing page links the predecessor card. Context is nominal 256K, normalized to 256000. Commercial service terms apply to hosted use. Microsoft publicly distributes the on-device GGUF through Azure ML; the registry labels it MIT. Direct unauthenticated registry and file checks verify the downloadable quantized weights, without asserting identity to the hosted full-precision checkpoint. Public Microsoft on-device GGUF distribution is verified; open_weights is true.", "field_status": {}, "audited_at": "2026-10-28T07:59:43.648773Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/294.json" }, { "id": "nvidia/nemotron-3.5-lightning", "slug": "nvidia--nemotron-3.5-lightning", "name": "Nemotron 3.5 Lightning 30B A3B", "developer": "NVIDIA", "developer_id": "nvidia", "family": "nemotron", "description": "Text model with Mamba-2, attention and sparse MoE layers, configurable reasoning, tool calling and native multi-token prediction; supports a 1M-token context.", "release_date": "2026-08-11", "date_precision": "day", "open_weights": true, "license": "OpenMDW-1.1", "parameters_b": 32.913263168, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "hybrid-ssm-transformer-moe", "context_window": 1048576, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/blob/a9904d24bcc1d289a1950fa9d2b978c47cf903b9/README.md", "source_type": "primary", "sources": [ { "label": "Pinned Nemotron 3.5 Lightning BF16 model card", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/blob/a9904d24bcc1d289a1950fa9d2b978c47cf903b9/README.md", "type": "primary" }, { "label": "August 11 Lightning launch", "url": "https://blogs.nvidia.com/blog/nemotron-lightning-switchyard-rtx-dgx/", "type": "primary" }, { "label": "Nemotron 3.5 Super and Lightning technical report, October 21, 2026", "url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3.5-Super-and-Lightning-Technical-Report.pdf", "type": "primary" }, { "label": "Pinned Lightning checkpoint configuration", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/blob/a9904d24bcc1d289a1950fa9d2b978c47cf903b9/config.json", "type": "weights" }, { "label": "Complete BF16 tensor index", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/blob/a9904d24bcc1d289a1950fa9d2b978c47cf903b9/model.safetensors.index.json", "type": "weights" }, { "label": "Bounded safetensors headers, first of 14 indexed shards", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/resolve/a9904d24bcc1d289a1950fa9d2b978c47cf903b9/model-00001-of-00014.safetensors", "type": "weights" }, { "label": "Nemotron-H MTP implementation", "url": "https://github.com/vllm-project/vllm/blob/ed1a7307950ab5e6593d25123579ed7ca998b350/vllm/model_executor/models/nemotron_h_mtp.py", "type": "weights" }, { "label": "Released OpenMDW-1.1 agreement", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16/blob/a9904d24bcc1d289a1950fa9d2b978c47cf903b9/LICENSE", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/nvidia/nemotron-3.5-lightning.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete BF16 checkpoint has 32.913263168B learned parameters: 31.577937344B main model plus 1.335325824B native MTP. Excludes 3072 router-buffer elements and separate optional DSpark/DFlash checkpoints. The advertised 3B active count is nominal for the main forward pass; drafting activity depends on speculative-decoding settings. Supported context is 1048576; 262144 is the lower-memory default. Index total_parameters omits MTP and its total_size disagrees with actual payload bytes, so the exact total uses tensor shapes and implementation.", "active_parameters_scope": "main_model", "parameter_components_b": { "main_model": 31.577937344, "mtp": 1.335325824 }, "field_status": {}, "audited_at": "2026-10-27T00:48:45.882052+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/386.json" }, { "id": "meta/muse-glimmer-30b", "slug": "meta--muse-glimmer-30b", "name": "Muse Glimmer 30B", "developer": "Meta", "developer_id": "meta", "family": "muse", "description": "Muse Glimmer 30B is a dense multimodal model from Meta Superintelligence Labs, distilled from Muse Spark for local agents, tool use, coding, and image understanding, with weights released under Apache 2.0.", "release_date": "2026-08-10", "date_precision": "day", "open_weights": true, "license": "Apache 2.0", "parameters_b": 29.776626688, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model", "source_type": "primary", "sources": [ { "label": "Announcement", "url": "https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model", "type": "announcement" }, { "label": "Hugging Face", "url": "https://huggingface.co/meta-models/Muse-Glimmer-30B", "type": "weights" }, { "label": "Technical report, October 23 revision", "url": "https://arxiv.org/html/2610.24628v3", "type": "primary" }, { "label": "Checkpoint parameter metadata", "url": "https://huggingface.co/meta-models/Muse-Glimmer-30B/resolve/cec94fcfd6512f2cfe708138c32533b346ea4f69/model.safetensors.index.json", "type": "weights" }, { "label": "Checkpoint configuration", "url": "https://huggingface.co/meta-models/Muse-Glimmer-30B/resolve/cec94fcfd6512f2cfe708138c32533b346ea4f69/config.json", "type": "primary" }, { "label": "Model license", "url": "https://huggingface.co/meta-models/Muse-Glimmer-30B/resolve/cec94fcfd6512f2cfe708138c32533b346ea4f69/LICENSE", "type": "primary" }, { "label": "Current developer documentation", "url": "https://dev.meta.ai/docs/muse-glimmer", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/meta/muse-glimmer-30b.toml", "type": "catalog" }, { "label": "Developer docs", "url": "https://developer.meta.com/ai/models/muse-glimmer/", "type": "docs" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The published BF16 checkpoint index reports 29.776626688B parameters, including the perception encoder and vision adapter/projection; the separate DFlash drafter is excluded. Meta's card and report give approximately 29.6B, while 30B is the nominal model designation. Active parameters are not separately specified; dense architecture does not establish an all-components per-token count. The configured default context is 131,072 tokens; Meta also advertises longer contexts without an exact upper limit. Meta lists text/image inputs; the report also describes video ingestion as a sequence of frames, without explicit video optimization.", "parameter_components_b": { "language_model": 27.854780928, "perception_encoder": 1.852639744, "vision_adapter": 0.04194304, "vision_projection": 0.027262976 }, "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T14:23:31.080Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/281.json" }, { "id": "openai/gpt-5.6-cyber", "slug": "openai--gpt-5.6-cyber", "name": "GPT-5.6 Cyber", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5.6", "description": "Cybersecurity model for authorized vulnerability research and security testing", "release_date": "2026-08-07", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 400000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://developers.openai.com/api/docs/changelog.md", "source_type": "primary", "sources": [ { "label": "GPT-5.6 Cyber — API model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.6-cyber.md", "type": "primary" }, { "label": "API changelog — August 7, 2026", "url": "https://developers.openai.com/api/docs/changelog.md", "type": "primary" }, { "label": "Expanding Daybreak as the cyber defense window narrows", "url": "https://openai.com/index/expanding-daybreak-as-the-cyber-defense-window-narrows/", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "All models", "url": "https://developers.openai.com/api/docs/models/all.md", "type": "primary" }, { "label": "GPT-5.6 System Card", "url": "https://deploymentsafety.openai.com/gpt-5-6/gpt-5-6.pdf", "type": "primary" }, { "label": "GPT-6 Astra System Card — Table 21", "url": "https://deploymentsafety.openai.com/gpt-6-astra/gpt-6-astra.pdf", "type": "primary" }, { "label": "Prompt caching", "url": "https://developers.openai.com/api/docs/guides/prompt-caching.md", "type": "primary" }, { "label": "How GPT-5.6 fuses frontier intelligence with frontier efficiency", "url": "https://openai.com/index/gpt-5-6-frontier-intelligence-efficiency/", "type": "primary" }, { "label": "Deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "GPT-5.6 Sol — API model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.6-sol.md", "type": "primary" }, { "label": "GPT-5.6: Frontier intelligence that scales with your ambition", "url": "https://openai.com/index/gpt-5-6/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5.6-cyber.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released through separately approved Daybreak Red access on August 7, 2026; the announcement followed on August 10. GPT-5.6 Cyber is an undated alias for purpose-trained cybersecurity models, built on GPT-5.6 Sol, and requires separate approval and provisioning. Current documentation lists 400,000 total context, 272,000 maximum input and 128,000 maximum output, with reasoning tokens, function calling and Responses tools; Chat Completions is unsupported. PDF is an API input format converted to extracted text and page images, not evidence of a native PDF encoder. Total and active parameter counts remain undisclosed after the October 27 recheck. Cyber has explicit Sol lineage, but no supported nominal Sol count is available to inherit; the later Astra comparison reports performance only. Production architecture remains unverified. Only the undated model alias is listed, no exact deprecation was found in the current deprecation document, and endpoint operation was not tested.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "unverified" }, "audited_at": "2026-10-27T17:05:35.895966+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/437.json" }, { "id": "upstage/solar-pro4", "slug": "upstage--solar-pro4", "name": "Solar Pro 4", "developer": "Upstage", "developer_id": "upstage", "family": "solar-pro", "description": "Text model for document analysis, coding and tool-using workflows, with optional reasoning and a 512K-token context window.", "release_date": "2026-08-06", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 524288, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://console.upstage.ai/docs/models/solar-pro-4", "source_type": "primary", "sources": [ { "label": "Solar Pro 4 model specifications and versions", "url": "https://console.upstage.ai/docs/models/solar-pro-4", "type": "primary" }, { "label": "Upstage Console changelog", "url": "https://console.upstage.ai/docs/resources/changelog", "type": "primary" }, { "label": "Upstage AI unveils Solar Pro 4", "url": "https://www.upstage.ai/news/upstage-ai-unveils-solar-pro-4-scoring-42-on-artificial-analysis-index-to-rank", "type": "primary" }, { "label": "Solar Pro 4 launch article", "url": "https://www.upstage.ai/blog/en/solar-pro-4", "type": "primary" }, { "label": "Responses API reference", "url": "https://console.upstage.ai/api/responses", "type": "primary" }, { "label": "Generative model comparison", "url": "https://console.upstage.ai/docs/models/compare/generative-intelligence", "type": "primary" }, { "label": "Solar Pro 4 tokenizer model card", "url": "https://huggingface.co/upstage/solar-pro4-tokenizer", "type": "primary" }, { "label": "Solar Pro 4 hosted model specifications", "url": "https://openrouter.ai/upstage/solar-pro4", "type": "provider" }, { "label": "Solar Pro 4: official Korean launch article", "url": "https://www.upstage.ai/blog/ko/solar-pro-4", "type": "primary" }, { "label": "Solar Open 2 official model card", "url": "https://huggingface.co/upstage/Solar-Open2-250B", "type": "primary" }, { "label": "Solar Open 2 Technical Report version history and abstract", "url": "https://arxiv.org/abs/2607.20062", "type": "primary" }, { "label": "Upstage Solar Mini 4 private-beta graphic reproduced by AI Times", "url": "https://cdn.aitimes.com/news/photo/202609/215245_219045_645.jpeg", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/upstage/solar-pro4.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Upstage documents release on August 6, 2026; paid API usage began August 11. An unaffiliated August 3 testing announcement does not identify the served snapshot. October 27 parameter re-audit: exact-model docs still explicitly withhold total size, and no active count is published. English/Korean launch articles share the OfficeVerse training-task pipeline with Solar Open 2 but do not establish base-weight lineage, so Open 2's nominal 250B/15B cannot be inherited. AI Times' Pro 4 250B claim remains unconfirmed by primary material; its Upstage-supplied image gives only Mini 4 35B/3B. Architecture remains undisclosed. The tokenizer license does not apply to the closed model weights.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T07:22:35.167Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/518.json" }, { "id": "meta/muse-spark-1.2", "slug": "meta--muse-spark-1.2", "name": "Muse Spark 1.2", "developer": "Meta", "developer_id": "meta", "family": "muse", "description": "Muse Spark 1.2 is a coding-focused update to Muse Spark 1.1 with improvements in code generation, complex debugging, codebase understanding, and end-to-end developer workflows.", "release_date": "2026-08-05", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 816, "active_parameters_b": 42, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "video", "pdf", "audio" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://research.meta.ai/blog/introducing-muse-code-and-muse-spark-1-2", "source_type": "primary", "sources": [ { "label": "Meta Muse Spark 1.2 launch", "url": "https://research.meta.ai/blog/introducing-muse-code-and-muse-spark-1-2", "type": "primary" }, { "label": "Meta hosted model specifications", "url": "https://dev.meta.ai/docs/models", "type": "primary" }, { "label": "Muse Spark 1.2 evaluation methodology", "url": "https://research.meta.ai/static/muse-spark-1-2-methodology", "type": "primary" }, { "label": "Meta Model API Terms of Service", "url": "https://dev.meta.ai/legal/terms-of-service", "type": "primary" }, { "label": "Meta model designations", "url": "https://dev.meta.ai/legal/geographic-use-policy", "type": "primary" }, { "label": "Separate Muse Spark Open model card", "url": "https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B/blob/main/README.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/meta/muse-spark-1.2.toml", "type": "catalog" }, { "label": "Meta Spark Open lineage", "url": "https://dev.meta.ai/models/muse-spark-open", "type": "primary" }, { "label": "Meta PyTorch team Open release and 1.2 lineage", "url": "https://pytorch.org/blog/muse-spark-open/", "type": "primary" }, { "label": "Meta TorchTitan explicit Spark 1.2 Open identity", "url": "https://pytorch.org/blog/torchtitan-update-new-sota-models-performance-at-scale-titanrl/", "type": "primary" }, { "label": "Meta Spark Open nominal total and per-token active size", "url": "https://dev.meta.ai/resources/blog/build-with-muse-spark-open", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Nominal 816B total, approximately 42B active per token, and MoE architecture are inferred from Meta's explicit identification of Spark Open as the August 1.2 release with additional safety post-training. These specifications are directly published for Open; no structural change is disclosed in that lineage. The exact hosted 1.2 weights and Open's weight license remain distinct.", "field_status": {}, "audited_at": "2026-10-27T14:25:26.081968Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/283.json" }, { "id": "liquid/lfm2.5-2.6b", "slug": "liquid--lfm2.5-2.6b", "name": "LFM2.5-2.6B", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm", "description": null, "release_date": "2026-08-04", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 2.69, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "hybrid", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B/blob/b212e75975bbda9676a94d3cb7b2ca25b65fed48/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B/blob/b212e75975bbda9676a94d3cb7b2ca25b65fed48/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:23.459Z", "last_seen": "2026-10-29T10:17:23.459Z", "upstream_present": false, "availability": "public", "notes": "Post-trained checkpoint explicitly included in the developer release. Parameter and context values follow the exact model card; MoE variants also use the LFM convolution/attention hybrid backbone." }, { "id": "liquid/lfm2.5-2.6b-base", "slug": "liquid--lfm2.5-2.6b-base", "name": "LFM2.5-2.6B-Base", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm", "description": null, "release_date": "2026-08-04", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 2.69, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "hybrid", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/lfm2-5-2-6b", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B-Base/blob/c57bdaed1ef166fe3095dda07f4a5e789ad5321e/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B-Base/blob/c57bdaed1ef166fe3095dda07f4a5e789ad5321e/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:23.459Z", "last_seen": "2026-10-29T10:17:23.459Z", "upstream_present": false, "availability": "public", "notes": "Pretrained checkpoint explicitly included in the developer release. Parameter and context values follow the exact model card; MoE variants also use the LFM convolution/attention hybrid backbone." }, { "id": "alibaba/qwen3.8-max", "slug": "alibaba--qwen3.8-max", "name": "Qwen3.8 Max", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Qwen3.8 model with 2.4 trillion nominal parameters, sparse MoE architecture, a one-million-token context, and text, image, video and PDF input.", "release_date": "2026-08-03", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": 2400, "active_parameters_b": 95, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text", "image", "video", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.alibabacloud.com/en/press-room/alibaba-unveils-qwen3-8-max?_p_lc=1", "source_type": "primary", "sources": [ { "label": "Alibaba Qwen3.8-Max launch and nominal counts", "url": "https://www.alibabacloud.com/en/press-room/alibaba-unveils-qwen3-8-max?_p_lc=1", "type": "primary" }, { "label": "QwenCloud exact model overview", "url": "https://www.qwencloud.com/models/qwen3.8-max", "type": "primary" }, { "label": "Official open checkpoint and hosted model distinction", "url": "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B/raw/main/README.md", "type": "primary" }, { "label": "Official Max alias upgrade notice", "url": "https://docs.qwencloud.com/changelog/model-updates/qwen3.8-max-upgrade", "type": "primary" }, { "label": "Official PDF input support", "url": "https://docs.qwencloud.com/developer-guides/tool-calling/pdf-understanding", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=qwen3.8-max&limit=100", "label": "Official Qwen inventory search for the exact hosted model", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=Qwen3.8&limit=1000", "label": "Official Qwen Qwen3.8 weight inventory", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.8-max.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Alibaba directly reports nominal 2.4T total / 95B active for hosted Qwen3.8-Max, with sparse MoE and hybrid attention; these are not exact checkpoint tensor counts. Since September 5, 2026 the endpoint maps to qwen3.8-max-0902; August 3 remains the original non-preview release. The open Qwen3.8-2.4T-A95B checkpoint is text-only and the hosted model adds vision and non-thinking support; the complete hosted variant has no identified public weight release, and its weight license remains undisclosed. PDF input is supported through Chat Completions/DashScope, not Responses. API callability was not tested. Classified as closed weights: no public release of this hosted variant was found in the developer documentation and weight inventory.", "field_status": { "license": "unverified" }, "audited_at": "2026-10-28T08:01:10.206859+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/070.json" }, { "id": "sakana/sakana-namazu", "slug": "sakana--sakana-namazu", "name": "Sakana Namazu", "developer": "Sakana AI", "developer_id": "sakana", "family": "sakana-namazu", "description": "Japanese-language adaptation of Kimi K2.6, trained for Japanese language and business contexts with thinking and tool use.", "release_date": "2026-08-03", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": 1000, "active_parameters_b": 32, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://sakana.ai/namazu-api/", "source_type": "primary", "sources": [ { "label": "Sakana Namazu API release", "url": "https://sakana.ai/namazu-api/", "type": "primary" }, { "label": "Official product page", "url": "https://sakana.ai/namazu/", "type": "announcement" }, { "label": "Official model documentation", "url": "https://console.sakana.ai/models?model=sakana-namazu", "type": "docs" }, { "label": "Pinned Kimi K2.6 base-model card", "url": "https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md", "type": "primary" }, { "label": "Pinned Kimi K2.6 configuration", "url": "https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/config.json", "type": "primary" }, { "label": "Namazu alpha prototypes and Sakana Chat launch", "url": "https://sakana.ai/namazu-alpha/", "type": "primary" }, { "label": "Sakana Translate adopts new-generation Namazu", "url": "https://sakana.ai/translate-update/", "type": "primary" }, { "label": "Namazu adoption in Evidence Finder", "url": "https://sakana.ai/namazu-aillis/", "type": "primary" }, { "label": "Sakana API platform terms", "url": "https://console.sakana.ai/terms-of-service", "type": "primary" }, { "label": "Exact Sakana-hosted Namazu endpoint metadata", "url": "https://openrouter.ai/api/v1/models/sakana/sakana-namazu/endpoints", "type": "provider" }, { "label": "Sakana AI Engineer Open House 2026", "url": "https://sakana.ai/open-house-2026/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/sakana/sakana-namazu.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Kimi K2.6-based Sakana Namazu API, released August 3, 2026; current model ID sakana-namazu-v1.0. Nominal 1T total and 32B activated parameters are inferred from Sakana’s explicit additional-training lineage and Moonshot’s base specification; reviewed updates describe no structural change. These are rounded published counts, not a Namazu tensor census or exact modality-specific active count. March alpha prototypes used different bases. The 262,144-token context is the advertised hosted limit. Neural modalities are text/image input and text output; files, search and code execution are API/tool facilities.", "field_status": {}, "audited_at": "2026-10-27T17:50:31.205399Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/486.json" }, { "id": "motif-technologies/motif-3", "slug": "motif-technologies--motif-3", "name": "Motif 3", "developer": "Motif Technologies", "developer_id": "motif-technologies", "family": "motif", "description": "Text MoE model with a 262,144-token context window, reasoning, and tool calling.", "release_date": "2026-08", "date_precision": "month", "open_weights": true, "license": "MIT", "parameters_b": 314.84177575, "active_parameters_b": 13.2, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/motif-technologies/motif-3.toml", "source_type": "catalog", "sources": [ { "label": "Motif 3 model card", "url": "https://huggingface.co/Motif-Technologies/Motif-3/blob/2ed2ed5cfabffa10fdabb2fc0d0288f8e6de893a/README.md", "type": "primary" }, { "label": "Motif 3: Technical Report", "url": "https://arxiv.org/html/2608.09119v1", "type": "primary" }, { "label": "Motif 3 released configuration", "url": "https://huggingface.co/Motif-Technologies/Motif-3/blob/2ed2ed5cfabffa10fdabb2fc0d0288f8e6de893a/config.json", "type": "primary" }, { "label": "Motif 3 pinned checkpoint index and indexed tensor headers", "url": "https://huggingface.co/Motif-Technologies/Motif-3/blob/2ed2ed5cfabffa10fdabb2fc0d0288f8e6de893a/model.safetensors.index.json", "type": "weights" }, { "label": "Motif 3 implementation", "url": "https://huggingface.co/Motif-Technologies/Motif-3/blob/2ed2ed5cfabffa10fdabb2fc0d0288f8e6de893a/modeling_motif.py", "type": "primary" }, { "label": "Motif 3 chat template", "url": "https://huggingface.co/Motif-Technologies/Motif-3/blob/2ed2ed5cfabffa10fdabb2fc0d0288f8e6de893a/chat_template.jinja", "type": "primary" }, { "label": "Motif 3 MIT license", "url": "https://huggingface.co/Motif-Technologies/Motif-3/blob/2ed2ed5cfabffa10fdabb2fc0d0288f8e6de893a/LICENSE.md", "type": "primary" }, { "label": "Motif 3 repository commit history", "url": "https://huggingface.co/api/models/Motif-Technologies/Motif-3/commits/main", "type": "primary" }, { "label": "Motif 3 initial repository tree", "url": "https://huggingface.co/api/models/Motif-Technologies/Motif-3/tree/1e90a1a2d6104945de4c9e6f8c450f567416d20d", "type": "weights" }, { "label": "Chat Motif release notes", "url": "https://chat.motiftech.io/release", "type": "primary" }, { "label": "Motif Technologies model catalog", "url": "https://motiftech.io/en/resources/models/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/motif-technologies/motif-3.toml", "type": "catalog" }, { "label": "Model weights", "url": "https://huggingface.co/Motif-Technologies/Motif-3", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Exact complete BF16 checkpoint: 314.84177575B, including a 0.251151876B MTP head and 19584 routing-bias parameters. All 155 pinned headers reconcile with configuration/code arithmetic; separate NVFP4 calibration scales are excluded. Published 314B total and 13.2B active figures are nominal. Release is retained at August 2026 precision: August 7 is empty-repository creation, while the final-model Chat Motif rollout is August 13; earliest public weight availability within the month is unresolved.", "field_status": {}, "audited_at": "2026-10-27T00:13:41.853328Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/368.json" }, { "id": "deepseek/deepseek-v4-flash-0731", "slug": "deepseek--deepseek-v4-flash-0731", "name": "DeepSeek V4 Flash 0731", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-flash", "description": "Official DeepSeek V4 Flash release with enhanced agentic capabilities and integrated DSpark speculative decoding", "release_date": "2026-07-31", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 304.178091454, "active_parameters_b": 13, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731/blob/7872f01b1d1fe23eabc4c98b48bffcef5a386062/README.md", "source_type": "primary", "sources": [ { "label": "DeepSeek API change log: July 31 release and subsequent alias lifecycle", "url": "https://api-docs.deepseek.com/updates/", "type": "primary" }, { "label": "DeepSeek-V4-Flash-0731 official model card", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731/blob/7872f01b1d1fe23eabc4c98b48bffcef5a386062/README.md", "type": "primary" }, { "label": "Pinned DeepSeek-V4-Flash-0731 model configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731/blob/7872f01b1d1fe23eabc4c98b48bffcef5a386062/config.json", "type": "primary" }, { "label": "Pinned 0731 inference configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731/blob/7872f01b1d1fe23eabc4c98b48bffcef5a386062/inference/config.json", "type": "primary" }, { "label": "Official 0731 inference implementation", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731/blob/7872f01b1d1fe23eabc4c98b48bffcef5a386062/inference/model.py", "type": "primary" }, { "label": "Official 0731 FP4 conversion", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731/blob/7872f01b1d1fe23eabc4c98b48bffcef5a386062/inference/convert.py", "type": "primary" }, { "label": "Pinned 0731 index and all 48 bounded safetensors headers", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731/resolve/7872f01b1d1fe23eabc4c98b48bffcef5a386062/model.safetensors.index.json", "type": "weights" }, { "label": "DeepSeek-V4 technical report", "url": "https://arxiv.org/html/2606.19348v1", "type": "primary" }, { "label": "Pinned 0731 text, reasoning and tool encoding", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731/blob/7872f01b1d1fe23eabc4c98b48bffcef5a386062/encoding/README.md", "type": "primary" }, { "label": "DeepSeek-V4-Flash-0731 MIT license", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731/blob/7872f01b1d1fe23eabc4c98b48bffcef5a386062/LICENSE", "type": "primary" }, { "label": "DeepSeek V4 Preview Release, April 24, 2026", "url": "https://api-docs.deepseek.com/news/news260424/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepseek/deepseek-v4-flash-0731.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "July 31, 2026 dated V4 Flash release. Core advertised 284B/13B architecture is retained from the April preview, with new post-training and a DSpark attachment. Complete logical weights total 304.178091454B: 284.332240471B main plus 19.845850983B unique DSpark; packed FP4 values are expanded, quantization scales/fixed routing indices excluded, and shared embeddings/head counted once. Nominal 13B active applies to main model only. Actual weights and inference configuration contain three DSpark stages, despite root config num_nextn_predict_layers=1. Advertised context is 1M; configured positional ceiling is 1048576. The September 10 redirect of deepseek-v4-flash to V4.1 Flash and October 17 successor update do not change this checkpoint identity. No endpoint execution was tested.", "active_parameters_scope": "main", "parameter_components_b": { "main": 284.332240471, "dspark": 19.845850983 }, "field_status": {}, "audited_at": "2026-10-26T19:13:46.885217+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/179.json" }, { "id": "google/gemini-robotics-er-2-preview", "slug": "google--gemini-robotics-er-2-preview", "name": "Gemini robotics er 2 preview", "developer": "Google", "developer_id": "google", "family": "gemini", "description": null, "release_date": "2026-07-30", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 131072, "modalities": { "input": [ "text", "image", "audio", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ai.google.dev/gemini-api/docs/changelog", "source_type": "primary", "sources": [ { "label": "Dated publisher API releases", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "documentation" }, { "label": "Publisher model capability table", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-robotics-er-2-preview", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:31:17.277Z", "last_seen": "2026-10-29T10:31:17.277Z", "upstream_present": false, "availability": "public", "notes": "" }, { "id": "thinkingmachines/inkling-small", "slug": "thinkingmachines--inkling-small", "name": "Inkling-Small", "developer": "Thinking Machines Lab", "developer_id": "thinkingmachines", "family": "inkling", "description": "Open-weight multimodal MoE model with adjustable reasoning effort, tool calling, and text output from text, image, or audio input.", "release_date": "2026-07-30", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 276, "active_parameters_b": 12, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "audio" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://thinkingmachines.ai/news/inkling-small/", "source_type": "primary", "sources": [ { "label": "Introducing Inkling-Small (July 30, 2026)", "url": "https://thinkingmachines.ai/news/inkling-small/", "type": "primary" }, { "label": "Inkling-Small Model Card", "url": "https://thinkingmachines.ai/model-card/inkling-small/", "type": "primary" }, { "label": "Inkling-Small pinned checkpoint configuration", "url": "https://huggingface.co/thinkingmachines/Inkling-Small/blob/8cc5877b44d343f88b92086aa1fb72897950f06a/config.json", "type": "primary" }, { "label": "Inkling-Small pinned chat template", "url": "https://huggingface.co/thinkingmachines/Inkling-Small/blob/8cc5877b44d343f88b92086aa1fb72897950f06a/chat_template.jinja", "type": "primary" }, { "label": "Inkling: Our Open-Weights Model (July 15, 2026)", "url": "https://thinkingmachines.ai/news/introducing-inkling/", "type": "primary" }, { "label": "Inkling model family", "url": "https://thinkingmachines.ai/inkling/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/thinkingmachines/inkling-small.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/thinkingmachines/Inkling-Small", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "276B total and 12B active are developer-reported nominal figures. The HF automated storage total is approximately 266B; this audit does not reconcile the difference or claim an exact learned parameter count. The 1,048,576-token limit is the checkpoint context; hosted Tinker limits differ. July 15 preview results preceded the July 30 full release.", "field_status": {}, "audited_at": "2026-10-27T05:01:05.577032Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/510.json" }, { "id": "xai/grok-voice-2.0", "slug": "xai--grok-voice-2.0", "name": "Grok Voice 2.0", "developer": "xAI", "developer_id": "xai", "family": "grok-voice", "description": null, "release_date": "2026-07-29", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text", "audio" ], "output": [ "text", "audio" ] }, "reasoning": null, "tool_call": null, "source_url": "https://docs.x.ai/developers/release-notes", "source_type": "primary", "sources": [ { "label": "Dated publisher release notes", "url": "https://docs.x.ai/developers/release-notes", "type": "documentation" }, { "label": "Conversational audio capability", "url": "https://docs.x.ai/developers/model-capabilities/audio/speech-to-speech", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:31:17.277Z", "last_seen": "2026-10-29T10:31:17.277Z", "upstream_present": false, "availability": "public", "notes": "Conversational speech model; separate from the dedicated transcription service. Deprecated October 14, 2026." }, { "id": "anthropic/claude-opus-5", "slug": "anthropic--claude-opus-5", "name": "Claude Opus 5", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude-opus", "description": "Claude Opus model for complex agentic coding and enterprise work, with adaptive thinking and a 1M-token context window.", "release_date": "2026-07-24", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1000000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.anthropic.com/news/claude-opus-5", "source_type": "primary", "sources": [ { "label": "Introducing Claude Opus 5", "url": "https://www.anthropic.com/news/claude-opus-5", "type": "primary" }, { "label": "Claude Opus 5 - Claude Platform Docs", "url": "https://platform.claude.com/docs/en/models/opus-5/overview", "type": "primary" }, { "label": "System Card: Claude Opus 5", "url": "https://anthropic.com/claude-opus-5-system-card", "type": "primary" }, { "label": "System Card: Claude Opus 5.5", "url": "https://www.anthropic.com/claude-opus-5-5-system-card", "type": "primary" }, { "label": "System Card: Claude Fable 5.1 & Claude Mythos 5.1", "url": "https://www.anthropic.com/claude-fable-5-1-mythos-5-1-system-card", "type": "primary" }, { "label": "System Card: Claude Sonnet 5.5", "url": "https://www.anthropic.com/claude-sonnet-5-5-system-card", "type": "primary" }, { "label": "System Card: Claude Haiku 5.5", "url": "https://www.anthropic.com/claude-haiku-5-5-system-card", "type": "primary" }, { "label": "GLM-5.3 and the spread of advanced cyber capabilities", "url": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities", "type": "primary" }, { "label": "Commercial Terms of Service", "url": "https://www.anthropic.com/legal/commercial-terms", "type": "primary" }, { "label": "PDF support - Claude Platform Docs", "url": "https://platform.claude.com/docs/en/build-with-claude/pdf-support", "type": "primary" }, { "label": "Prompting Claude Opus 5", "url": "https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-opus-5", "type": "primary" }, { "label": "Anthropic Transparency Hub - Claude Opus 5 Summary Table", "url": "https://www.anthropic.com/transparency", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/anthropic/claude-opus-5.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Anthropic has not disclosed nominal total or active parameter counts or structural architecture. PDF inputs are processed as text and page images.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T10:43:13.147306Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/110.json" }, { "id": "swiss-ai/apertus-v1.5-70b", "slug": "swiss-ai--apertus-v1.5-70b", "name": "Apertus 1.5 70B", "developer": "Swiss AI Initiative", "developer_id": "swiss-ai", "family": "apertus", "description": null, "release_date": "2026-07-24", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 72.005819139, "active_parameters_b": null, "parameters_source": "https://huggingface.co/api/models/swiss-ai/Apertus-v1.5-70B/revision/18af3fe329469e7ed0060e82dc0229bdb99a3af0", "architecture": "transformer", "context_window": 262144, "modalities": { "input": [ "text", "image", "audio" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/swiss-ai/Apertus-v1.5-70B", "source_type": "primary", "sources": [ { "label": "Publisher release notes", "url": "https://huggingface.co/blog/swiss-ai/apertus-v15-release-notes", "type": "primary" }, { "label": "Official model card", "url": "https://huggingface.co/swiss-ai/Apertus-v1.5-70B/blob/18af3fe329469e7ed0060e82dc0229bdb99a3af0/README.md", "type": "primary" }, { "label": "Official checkpoint parameter metadata", "url": "https://huggingface.co/api/models/swiss-ai/Apertus-v1.5-70B/revision/18af3fe329469e7ed0060e82dc0229bdb99a3af0", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:11:51.983Z", "last_seen": "2026-10-30T07:11:51.983Z", "upstream_present": false, "availability": "public", "notes": "Total is the official safetensors parameter count, including released multimodal components, rather than the nominal size in the name. Model card confirms 262144 context and text/image/audio inputs. Downloads require accepting the publisher's license and usage policy; raw config retrieval returned401 during review. Publisher release notes explicitly date these post-trained weights July24,2026; the October15 article is a retrospective technical release note. Thinking and tool calling are separately supported; the current serving instructions do not support combining them." }, { "id": "swiss-ai/apertus-v1.5-8b", "slug": "swiss-ai--apertus-v1.5-8b", "name": "Apertus 1.5 8B", "developer": "Swiss AI Initiative", "developer_id": "swiss-ai", "family": "apertus", "description": null, "release_date": "2026-07-24", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 8.903547555, "active_parameters_b": null, "parameters_source": "https://huggingface.co/api/models/swiss-ai/Apertus-v1.5-8B/revision/0be762a6a58dad2a3b3f71b2e0f46d34339fa836", "architecture": "transformer", "context_window": 262144, "modalities": { "input": [ "text", "image", "audio" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/swiss-ai/Apertus-v1.5-8B", "source_type": "primary", "sources": [ { "label": "Publisher release notes", "url": "https://huggingface.co/blog/swiss-ai/apertus-v15-release-notes", "type": "primary" }, { "label": "Official model card", "url": "https://huggingface.co/swiss-ai/Apertus-v1.5-8B/blob/0be762a6a58dad2a3b3f71b2e0f46d34339fa836/README.md", "type": "primary" }, { "label": "Official checkpoint parameter metadata", "url": "https://huggingface.co/api/models/swiss-ai/Apertus-v1.5-8B/revision/0be762a6a58dad2a3b3f71b2e0f46d34339fa836", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:11:51.983Z", "last_seen": "2026-10-30T07:11:51.983Z", "upstream_present": false, "availability": "public", "notes": "Total is the official safetensors parameter count, including released multimodal components, rather than the nominal size in the name. Model card confirms 262144 context and text/image/audio inputs. Downloads require accepting the publisher's license and usage policy; raw config retrieval returned401 during review. Publisher release notes explicitly date these post-trained weights July24,2026; the October15 article is a retrospective technical release note. Thinking and tool calling are separately supported; the current serving instructions do not support combining them." }, { "id": "inclusionai/ling-3.0-flash", "slug": "inclusionai--ling-3.0-flash", "name": "Ling 3.0 Flash", "developer": "InclusionAI", "developer_id": "inclusionai", "family": "ling", "description": "Text-only hybrid-linear MoE model with a 42-layer KDA/MLA backbone, top-8 routing among 512 experts plus one shared expert, optional thinking, and tool use.", "release_date": "2026-07-23", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 127.486384608, "active_parameters_b": 5.1, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/inclusionai/ling-3.0-flash.toml", "source_type": "catalog", "sources": [ { "label": "Ling-3.0-flash official model card", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash/blob/fcc60a52b34c69c6271d0ac8a8cb0815020b6a43/README.md", "type": "primary" }, { "label": "Ling-3.0-flash checkpoint configuration", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash/blob/fcc60a52b34c69c6271d0ac8a8cb0815020b6a43/config.json", "type": "primary" }, { "label": "Ling-3.0-flash official model implementation", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash/blob/fcc60a52b34c69c6271d0ac8a8cb0815020b6a43/modeling_bailing_moe_v3.py", "type": "primary" }, { "label": "Ling-3.0-flash complete checkpoint weight index", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash/blob/fcc60a52b34c69c6271d0ac8a8cb0815020b6a43/model.safetensors.index.json", "type": "primary" }, { "label": "Ling-3.0-flash BF16 checkpoint tensor headers, all 24 shards", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash/tree/fcc60a52b34c69c6271d0ac8a8cb0815020b6a43", "type": "primary" }, { "label": "Ling-3.0-flash official chat template", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash/blob/fcc60a52b34c69c6271d0ac8a8cb0815020b6a43/chat_template.jinja", "type": "primary" }, { "label": "Ant Ling platform release changelog", "url": "https://developer.ant-ling.com/en/docs/getting-started/changelog/", "type": "primary" }, { "label": "Official Ling model family and comparison table", "url": "https://developer.ant-ling.com/en/docs/models/ling", "type": "primary" }, { "label": "Official Ling-3.0 training-checkpoint documentation", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-base/blob/main/README.md", "type": "primary" }, { "label": "Ling-3.0-flash: More Useful Work per Token", "url": "https://developer.ant-ling.com/en/blogs/ling-3.0-flash-release/", "type": "primary" }, { "label": "SGLang exact-model Ling-3.0-flash deployment cookbook", "url": "https://docs.sglang.io/cookbook/autoregressive/InclusionAI/Ling-3.0-flash", "type": "primary" }, { "label": "vLLM exact-model Ling-3.0-flash deployment recipe", "url": "https://recipes.vllm.ai/inclusionAI/Ling-3.0-flash", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/inclusionai/ling-3.0-flash.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/inclusionAI/Ling-3.0-flash", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete checkpoint: 127.486384608B learned parameters = 124.414191072B backbone + 3.072193536B auxiliary MTP; excludes 20,992 routing-buffer elements. Official 124B is a nominal backbone count. Official 5.1B active is nominal; its exact accounting scope is not defined. Excluding only the input embedding produces a matching rounded figure but is an inference. Context: advertised extension up to 1M (nominal), versus 262144 native/configured and documented in deployment recipes; an exact 1M cap, extension recipe and evaluation were not established. MIT weights followed the July 23, 2026 chat/API launch.", "parameter_components_b": { "backbone": 124.414191072, "mtp": 3.072193536 }, "field_status": {}, "audited_at": "2026-10-26T21:00:39.829497Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/242.json" }, { "id": "alibaba/qwen3.7-flash", "slug": "alibaba--qwen3.7-flash", "name": "Qwen3.7 Flash", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Qwen3.7 vision-language model supporting text, image, and video inputs, text output, thinking mode, and function calling, with a 1,000,000-token context window.", "release_date": "2026-07-21", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1000000, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.7-flash.toml", "source_type": "catalog", "sources": [ { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.7-flash.toml", "type": "catalog" }, { "label": "Alibaba Cloud model information", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-7-flash", "type": "primary" }, { "label": "Alibaba Cloud model lifecycle and updates", "url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "type": "primary" }, { "label": "Alibaba Cloud Beijing model release log", "url": "https://help.aliyun.com/zh/model-studio/newly-released-models", "type": "primary" }, { "label": "Alibaba Cloud snapshot definition", "url": "https://help.aliyun.com/zh/model-studio/qwen3-7-flash", "type": "primary" }, { "label": "QwenCloud exact model page", "url": "https://www.qwencloud.com/models/qwen3.7-flash", "type": "primary" }, { "label": "Alibaba Cloud snapshot capability mapping", "url": "https://help.aliyun.com/zh/model-studio/model-pricing", "type": "primary" }, { "label": "Qwen later parameter comparison table", "url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next", "type": "primary" }, { "label": "Alibaba Cloud text-generation comparison", "url": "https://www.alibabacloud.com/help/en/model-studio/text-generation-model", "type": "primary" }, { "label": "QwenCloud model release notes", "url": "https://docs.qwencloud.com/changelog/models", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=qwen3.7-flash&limit=100", "label": "Official Qwen inventory search for the exact hosted model", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=Qwen3.7&limit=1000", "label": "Official Qwen Qwen3.7 weight inventory", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "July 21, 2026 is the earliest explicit Model Studio release-table date found in both Beijing and Singapore; July 15 is the named snapshot date. A partially indexed console shows an unlabeled July 20 date, whose meaning/timezone remains unresolved. Total and active parameters remain undisclosed: the capability comparison with 3.6-Flash establishes no size-bearing base, and later Qwen3.7-Plus counts do not apply to Flash. Dense/MoE architecture and weight license remain unestablished. No public Flash weight release is available. Official pricing maps the alias to the July 15 API snapshot. Classified as closed weights: no public release of this hosted variant was found in the developer documentation and weight inventory.", "field_status": { "license": "undisclosed", "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-28T08:01:10.206144+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/063.json" }, { "id": "google/gemini-3.5-flash-lite", "slug": "google--gemini-3.5-flash-lite", "name": "Gemini 3.5 Flash-Lite", "developer": "Google", "developer_id": "google", "family": "gemini-flash-lite", "description": "Gemini Flash-Lite model with multimodal input, text output, thinking and function calling.", "release_date": "2026-07-21", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber/", "source_type": "primary", "sources": [ { "label": "Gemini 3.5 Flash-Lite model card", "url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "type": "primary" }, { "label": "Gemini 3.5 Flash-Lite API specifications", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite", "type": "primary" }, { "label": "Introducing Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber", "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber/", "type": "primary" }, { "label": "Gemini 3.1 Flash-Lite model card", "url": "https://deepmind.google/models/model-cards/gemini-3-1-flash-lite/", "type": "primary" }, { "label": "Gemini 3 Pro model card, updated May 2026", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-Pro-Model-Card.pdf", "type": "primary" }, { "label": "Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemini-3.5-flash-lite.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Sparse MoE Transformer based on Gemini 3.1 Flash-Lite, whose card names Gemini 3 Pro as its base. Total and active parameter counts are undisclosed for this model and both named bases in the reviewed sources.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T13:03:55.511333+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/208.json" }, { "id": "google/gemini-3.6-flash", "slug": "google--gemini-3.6-flash", "name": "Gemini 3.6 Flash", "developer": "Google", "developer_id": "google", "family": "gemini-flash", "description": "Gemini Flash model for coding and multimodal tasks, with text output, thinking and function calling.", "release_date": "2026-07-21", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber/", "source_type": "primary", "sources": [ { "label": "Gemini 3.6 Flash model card", "url": "https://deepmind.google/models/model-cards/gemini-3-6-flash/", "type": "primary" }, { "label": "Gemini 3.6 Flash API model specification", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.6-flash", "type": "primary" }, { "label": "Introducing Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber", "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber/", "type": "primary" }, { "label": "Gemini 3.5 Flash model card", "url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "type": "primary" }, { "label": "Gemini 3 Flash model card, December 2025", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-Flash-Model-Card.pdf", "type": "primary" }, { "label": "Gemini 3 Pro model card, updated May 2026", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-Pro-Model-Card.pdf", "type": "primary" }, { "label": "Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemini-3.6-flash.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Sparse MoE Transformer based on Gemini 3.5 Flash, whose documented base chain leads through Gemini 3 Flash to Gemini 3 Pro. Total and active parameter counts are undisclosed for this model and its named bases in the reviewed sources.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T13:05:01.126517Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/210.json" }, { "id": "mindlab/macaron-v1-tall", "slug": "mindlab--macaron-v1-tall", "name": "Macaron V1 Tall", "developer": "MindLab", "developer_id": "mindlab", "family": "macaron-v1", "description": "Qwen3.6-based Mixture-of-LoRA model with specialists for chat, personal agents, coding, and generative UI", "release_date": "2026-07-21", "date_precision": "day", "open_weights": true, "license": "MIT + Apache-2.0", "parameters_b": 51.054430064, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/mindlab-research/Macaron-V1-Tall/raw/main/README.md", "source_type": "primary", "sources": [ { "label": "Macaron-V1-Tall official model card", "url": "https://huggingface.co/mindlab-research/Macaron-V1-Tall/raw/main/README.md", "type": "primary" }, { "label": "Introducing Macaron-V1", "url": "https://www.mindlab.im/updates/introducing-macaron-v1", "type": "primary" }, { "label": "Macaron-V1 technical report, revision 2", "url": "https://arxiv.org/html/2608.09819v2", "type": "primary" }, { "label": "Released Tall checkpoint configuration", "url": "https://huggingface.co/mindlab-research/Macaron-V1-Tall/raw/main/config.json", "type": "weights" }, { "label": "Tall bundled license file", "url": "https://huggingface.co/mindlab-research/Macaron-V1-Tall/raw/main/LICENSE", "type": "primary" }, { "label": "Qwen3.6-35B-A3B base-model card", "url": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B/raw/main/README.md", "type": "primary" }, { "label": "Tall native chat template", "url": "https://huggingface.co/mindlab-research/Macaron-V1-Tall/raw/main/chat_template.jinja", "type": "weights" }, { "label": "Tall video processor configuration", "url": "https://huggingface.co/mindlab-research/Macaron-V1-Tall/raw/main/video_preprocessor_config.json", "type": "weights" }, { "label": "Official Tall serving profile", "url": "https://raw.githubusercontent.com/MindLab-Research/Mixture-of-LoRA-Harness/main/mol_harness/scripts/start_vllm_tall.sh", "type": "primary" }, { "label": "Official Macaron hosted-model lineup", "url": "https://mint-doc.macaron.im/model-api/models/", "type": "primary" }, { "label": "Official Tall repository metadata and file listing", "url": "https://huggingface.co/api/models/mindlab-research/Macaron-V1-Tall", "type": "weights" }, { "label": "Tall pinned base checkpoint index and bounded shard headers", "url": "https://huggingface.co/mindlab-research/Macaron-V1-Tall/resolve/e64d7387f78f2bb20c729b836319b12db6a3a553/model.safetensors.index.json", "type": "weights" }, { "label": "Tall pinned L0 adapter header", "url": "https://huggingface.co/mindlab-research/Macaron-V1-Tall/resolve/e64d7387f78f2bb20c729b836319b12db6a3a553/loras/L0/adapter_model.safetensors", "type": "weights" }, { "label": "Tall pinned L1 adapter header", "url": "https://huggingface.co/mindlab-research/Macaron-V1-Tall/resolve/e64d7387f78f2bb20c729b836319b12db6a3a553/loras/L1/adapter_model.safetensors", "type": "weights" }, { "label": "Tall pinned L2 adapter header", "url": "https://huggingface.co/mindlab-research/Macaron-V1-Tall/resolve/e64d7387f78f2bb20c729b836319b12db6a3a553/loras/L2/adapter_model.safetensors", "type": "weights" }, { "label": "Tall pinned L3 adapter header", "url": "https://huggingface.co/mindlab-research/Macaron-V1-Tall/resolve/e64d7387f78f2bb20c729b836319b12db6a3a553/loras/L3/adapter_model.safetensors", "type": "weights" }, { "label": "Transformers implementation of the declared checkpoint architecture", "url": "https://raw.githubusercontent.com/huggingface/transformers/main/src/transformers/models/qwen3_5_moe/modeling_qwen3_5_moe.py", "type": "primary" }, { "label": "vLLM Qwen3.5 MoE multi-token predictor implementation", "url": "https://raw.githubusercontent.com/vllm-project/vllm/main/vllm/model_executor/models/qwen3_5_mtp.py", "type": "primary" }, { "label": "PEFT LoRA and parameter-wrapper implementation", "url": "https://raw.githubusercontent.com/huggingface/peft/main/src/peft/tuners/lora/layer.py", "type": "primary" }, { "label": "Official Macaron-V1 family architecture and parameter accounting", "url": "https://raw.githubusercontent.com/MindLab-Research/Macaron-V1/main/README.md", "type": "primary" }, { "label": "Official Mixture-of-LoRA harness model profiles", "url": "https://raw.githubusercontent.com/MindLab-Research/Mixture-of-LoRA-Harness/main/README.md", "type": "primary" }, { "label": "Macaron-V1 technical report revision history", "url": "https://arxiv.org/abs/2608.09819", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mindlab/macaron-v1-tall.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/mindlab-research/Macaron-V1-Tall", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released checkpoint totals 51.054430064B learned parameters including four 3.775651840B LoRA adapters and 0.844640768B auxiliary MTP; excluding MTP gives 50.209789296B. MindLab labels the release 50B; its ~50.1B calculation uses a nominal 35B base. Active 3B is Qwen's nominal language-backbone count, retained through MindLab's explicit frozen Qwen3.6-35B-A3B lineage; it excludes Macaron adapters, and complete active parameters remain undisclosed. Stored adapter sizes are not token-active counts. Card declares MIT with inherited Apache-2.0 base terms. Vision/video is inherited; specialist training is text-only. Hosted docs conflict with the Tall card/report on adapter sizes.", "parameter_components_b": { "base_language_model": 34.660610688, "mtp": 0.844640768, "lora_l0": 3.77565184, "lora_l1": 3.77565184, "lora_l2": 3.77565184, "lora_l3": 3.77565184, "vision_encoder_projector": 0.446571248 }, "field_status": {}, "audited_at": "2026-10-27T15:25:34.955643Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/298.json", "active_parameters_scope": "base_language_model" }, { "id": "mindlab/macaron-v1-venti", "slug": "mindlab--macaron-v1-venti", "name": "Macaron V1 Venti", "developer": "MindLab", "developer_id": "mindlab", "family": "macaron-v1", "description": "GLM-5.2-based Mixture-of-LoRA model with specialists for chat, personal agents, coding, and generative UI", "release_date": "2026-07-21", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 784.082091008, "active_parameters_b": 40, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.mindlab.im/updates/introducing-macaron-v1", "source_type": "primary", "sources": [ { "label": "Mind Lab release announcement", "url": "https://www.mindlab.im/updates/introducing-macaron-v1", "type": "primary" }, { "label": "Hugging Face", "url": "https://huggingface.co/mindlab-research/Macaron-V1-Venti", "type": "weights" }, { "label": "Macaron-V1 technical report v2", "url": "https://arxiv.org/html/2608.09819v2", "type": "primary" }, { "label": "Venti configuration", "url": "https://huggingface.co/mindlab-research/Macaron-V1-Venti/raw/main/config.json", "type": "primary" }, { "label": "Venti chat template", "url": "https://huggingface.co/mindlab-research/Macaron-V1-Venti/raw/main/chat_template.jinja", "type": "primary" }, { "label": "Macaron developer comparison table", "url": "https://github.com/MindLab-Research/Macaron-V1", "type": "primary" }, { "label": "Updated Mint model lineup", "url": "https://mint-doc.macaron.im/model-api/models/", "type": "primary" }, { "label": "Pinned Venti base tensor index and complete shard headers", "url": "https://huggingface.co/mindlab-research/Macaron-V1-Venti/blob/be05322da0f462b08e29166ab92eb98dd4c93d6c/model.safetensors.index.json", "type": "weights" }, { "label": "Pinned Venti L0-L3 adapter tensor headers", "url": "https://huggingface.co/mindlab-research/Macaron-V1-Venti/tree/be05322da0f462b08e29166ab92eb98dd4c93d6c/loras", "type": "weights" }, { "label": "GLM MoE DSA reference implementation: router buffers and MTP boundary", "url": "https://github.com/huggingface/transformers/blob/6c5f8f4186c15f4bfbd67941a1736b2b347e0155/src/transformers/models/glm_moe_dsa/modeling_glm_moe_dsa.py", "type": "primary" }, { "label": "Transformers GLM-5.2 implementation documentation", "url": "https://huggingface.co/docs/transformers/model_doc/glm_moe_dsa", "type": "primary" }, { "label": "Developer GLM-5.2 training model configuration", "url": "https://github.com/THUDM/slime/blob/18868a79b0b083a093798a94ed0f8cb4f71b9c14/scripts/models/glm5.2-744B-A40B.sh", "type": "primary" }, { "label": "Developer checkpoint conversion: router-state mapping", "url": "https://github.com/THUDM/slime/blob/18868a79b0b083a093798a94ed0f8cb4f71b9c14/slime/backends/megatron_utils/hf_to_megatron/deepseek.py", "type": "primary" }, { "label": "Megatron expert-bias state update implementation", "url": "https://github.com/NVIDIA/Megatron-LM/blob/e3f528380c197e8811fb68064d9ef3acf59fd3c6/megatron/core/transformer/moe/moe_utils.py", "type": "primary" }, { "label": "Z.ai GLM-5.2 nominal 744B-A40B specification", "url": "https://raw.githubusercontent.com/zai-org/GLM-5/main/README.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mindlab/macaron-v1-venti.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete checkpoint: 784.082091008B learned parameters, including MTP and four LoRA adapters; excludes 19,456 router-state values. Mind Lab retains a 748B nominal release label and ~774.8B residency using a nominal 744B base. Active 40B is a nominal inference for the frozen GLM-5.2 base only, from Z.ai's 744B-A40B label; excludes adapters. Full Venti active count remains undisclosed. Context maximum is 1,048,576 tokens; reference serving profile is 262K. MIT release retains inherited base-model and dependency requirements.", "parameter_components_b": { "base_without_mtp": 743.377000704, "mtp": 9.95292032, "lora_l0": 7.688042496, "lora_l1": 7.688042496, "lora_l2": 7.688042496, "lora_l3": 7.688042496 }, "field_status": {}, "audited_at": "2026-10-27T15:26:45.121556Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/299.json", "active_parameters_scope": "base_model" }, { "id": "poolside/laguna-s-2.1", "slug": "poolside--laguna-s-2.1", "name": "Laguna S 2.1", "developer": "Poolside", "developer_id": "poolside", "family": "laguna", "description": "Poolside text model for agentic coding, with open weights, optional reasoning and tool calling.", "release_date": "2026-07-21", "date_precision": "day", "open_weights": true, "license": "OpenMDW-1.1", "parameters_b": 118, "active_parameters_b": 8, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://poolside.ai/blog/introducing-laguna-s-2-1", "source_type": "primary", "sources": [ { "label": "Introducing Laguna S 2.1 — July 21, 2026", "url": "https://poolside.ai/blog/introducing-laguna-s-2-1", "type": "primary" }, { "label": "Laguna S 2.1 BF16 model card, pinned revision", "url": "https://huggingface.co/poolside/Laguna-S-2.1/blob/0f573140834b11cfac0c2af97a101a7a69a13e22/README.md", "type": "primary" }, { "label": "Laguna S 2.1 BF16 configuration, pinned revision", "url": "https://huggingface.co/poolside/Laguna-S-2.1/blob/0f573140834b11cfac0c2af97a101a7a69a13e22/config.json", "type": "primary" }, { "label": "Laguna S 2.1 selected model implementation", "url": "https://huggingface.co/poolside/Laguna-S-2.1/blob/0f573140834b11cfac0c2af97a101a7a69a13e22/modeling_laguna.py", "type": "primary" }, { "label": "Laguna S 2.1 configuration defaults", "url": "https://huggingface.co/poolside/Laguna-S-2.1/blob/0f573140834b11cfac0c2af97a101a7a69a13e22/configuration_laguna.py", "type": "primary" }, { "label": "Laguna S 2.1 complete checkpoint index and all 46 pinned shard headers", "url": "https://huggingface.co/poolside/Laguna-S-2.1/blob/0f573140834b11cfac0c2af97a101a7a69a13e22/model.safetensors.index.json", "type": "primary" }, { "label": "Laguna S 2.1 model-materials license", "url": "https://huggingface.co/poolside/Laguna-S-2.1/blob/0f573140834b11cfac0c2af97a101a7a69a13e22/LICENSE.md", "type": "primary" }, { "label": "Laguna S 2.1 chat template", "url": "https://huggingface.co/poolside/Laguna-S-2.1/blob/0f573140834b11cfac0c2af97a101a7a69a13e22/chat_template.jinja", "type": "primary" }, { "label": "Official Laguna S 2.1 NVFP4 sibling card, checked for count conflicts", "url": "https://huggingface.co/poolside/Laguna-S-2.1-NVFP4/blob/826aacdf6d8b2699d4e367def6f17c83b06044c2/README.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/poolside/laguna-s-2.1.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Parameters are Poolside's nominal 118B total and approximately 8B active per token. At BF16 revision 0f573140834b11cfac0c2af97a101a7a69a13e22, all 46 indexed shard headers reconcile to 117,561,977,600 stored values; excluding 12,032 frozen router-correction values leaves 117,561,965,568 learned-weight elements. Both untied vocabulary matrices and all enabled learned modules are included; the separate DFlash draft is excluded. Quantized sibling prose reports 8.5B active, whereas the canonical BF16 card and release report approximately 8B; the latter is retained without claiming exact active-count precision. The released 1,048,576-token configuration uses YaRN; the free 256K hosted endpoint is a separate serving cap. Published code implements sigmoid MoE routing and softplus attention output gates, despite the card describing softplus router gating.", "field_status": {}, "audited_at": "2026-10-27T03:45:49.367374Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/471.json" }, { "id": "alibaba/qwen3.8-max-preview", "slug": "alibaba--qwen3.8-max-preview", "name": "Qwen3.8 Max Preview", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Hosted Qwen3.8 preview for reasoning, coding and professional tasks, with a one-million-token context and text, image and video input; later retired in favor of Qwen3.8-Max.", "release_date": "2026-07-19", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": 2400, "active_parameters_b": 95, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.8-max-preview.toml", "source_type": "catalog", "sources": [ { "label": "Alibaba exact-preview specification and July 19 access", "url": "https://www.alibabacloud.com/help/en/lingma/qwen3-8-max-preview-limited-time-offer", "type": "primary" }, { "label": "Qoder exact-preview launch promotion", "url": "https://docs.qoder.com/events/qwen-max-preview", "type": "primary" }, { "label": "Historical official Qwen Code preview integration", "url": "https://github.com/QwenLM/qwen-code/pull/7199", "type": "primary" }, { "label": "Pinned preview context and modality preset", "url": "https://github.com/QwenLM/qwen-code/blob/b0b654bc17e86c1e95c70853cedbc61b78577799/packages/core/src/providers/presets/alibaba-token-plan.ts", "type": "primary" }, { "label": "Qwen collaborator on preview function calling", "url": "https://github.com/QwenLM/qwen-code/issues/8003", "type": "primary" }, { "label": "Official preview retirement and routing notice", "url": "https://docs.qwencloud.com/token-plan/personal/token-plan-personal-overview", "type": "primary" }, { "label": "Tongyi original named preview announcement", "url": "https://tongyilab.substack.com/p/july-24-2026-the-tongyi-weekly", "type": "primary" }, { "label": "Official Max release nominal total and active parameters", "url": "https://qwen.ai/blog?id=qwen3.8", "type": "primary" }, { "label": "Alibaba preview-to-official-release notice", "url": "https://www.alibabacloud.com/en/notice/alibaba_cloud_model_studio_model_upgrade_notice_81a?_p_lc=1", "type": "primary" }, { "label": "Official Max MoE specification", "url": "https://www.qwencloud.com/models/qwen3.8-max", "type": "primary" }, { "label": "Official Qwen3.8 model card and Max lineage", "url": "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=qwen3.8-max-preview&limit=100", "label": "Official Qwen inventory search for the exact hosted model", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=Qwen3.8&limit=1000", "label": "Official Qwen Qwen3.8 weight inventory", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.8-max-preview.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "2.4T is the nominal total published for the exact preview. 95B active and MoE are nominal lineage inferences from the developer's named preview and official Qwen3.8-Max release; no structural change between them is documented. Exact preview active parameters and visual-component sizes are not separately disclosed. Historical capabilities describe the preview; its ID now routes to Max. No public weights for the exact Preview revision were found; its weight license remains undisclosed. Classified as closed weights: no public release of this hosted variant was found in the developer documentation and weight inventory.", "field_status": { "license": "unverified" }, "audited_at": "2026-10-28T08:01:10.207442+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/072.json" }, { "id": "moonshotai/kimi-k3", "slug": "moonshotai--kimi-k3", "name": "Kimi K3", "developer": "Moonshot AI", "developer_id": "moonshotai", "family": "kimi-k3", "description": "Open-weight multimodal MoE model with hybrid KDA/MLA attention, a 1,048,576-token context window, always-on reasoning, and tool calling.", "release_date": "2026-07-16", "date_precision": "day", "open_weights": true, "license": "Kimi K3 License", "parameters_b": 2780, "active_parameters_b": 104.2, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/moonshotai/Kimi-K3", "source_type": "primary", "sources": [ { "label": "Kimi K3 model card", "url": "https://huggingface.co/moonshotai/Kimi-K3", "type": "primary" }, { "label": "Kimi K3 technical report, Table 1", "url": "https://arxiv.org/html/2607.24653v1", "type": "primary" }, { "label": "Kimi K3 release date", "url": "https://www.kimi.com/code/docs/en/kimi-code/whats-new.html", "type": "primary" }, { "label": "Kimi K3 License", "url": "https://huggingface.co/moonshotai/Kimi-K3/blob/f831ab66814297da540d832a5235f8e904f29d06/LICENSE", "type": "primary" }, { "label": "Kimi K3 checkpoint configuration", "url": "https://huggingface.co/moonshotai/Kimi-K3/blob/f831ab66814297da540d832a5235f8e904f29d06/config.json", "type": "weights" }, { "label": "Kimi K3 API guide", "url": "https://platform.kimi.ai/docs/guide/kimi-k3-quickstart", "type": "primary" }, { "label": "Kimi Code model routing", "url": "https://www.kimi.com/code/docs/en/kimi-code/models.html", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/moonshotai/kimi-k3.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Counts follow technical-report Table 1 nominal precision. That table lists one MTP layer, while released config sets num_nextn_predict_layers=0; no exact checkpoint-wide count is claimed. Packed MXFP4 storage and scale tensors are not learned-parameter counts. K3 always reasons; disabling thinking in Kimi Code routes to K2.8 Preview.", "field_status": {}, "audited_at": "2026-10-27T00:03:30.457Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/364.json" }, { "id": "thinkingmachines/inkling", "slug": "thinkingmachines--inkling", "name": "Inkling", "developer": "Thinking Machines Lab", "developer_id": "thinkingmachines", "family": "inkling", "description": "Open-weight multimodal MoE model with text output, image and audio input, controllable reasoning effort, and tool calling.", "release_date": "2026-07-15", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 975, "active_parameters_b": 41, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "audio" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://thinkingmachines.ai/model-card/inkling/", "source_type": "primary", "sources": [ { "label": "Inkling Model Card", "url": "https://thinkingmachines.ai/model-card/inkling/", "type": "primary" }, { "label": "Inkling: Our Open-Weights Model", "url": "https://thinkingmachines.ai/news/introducing-inkling/", "type": "primary" }, { "label": "Inkling official checkpoint card, pinned", "url": "https://huggingface.co/thinkingmachines/Inkling/blob/828496eeae4c243ff1a22f7f28ff83694f2f7bc9/README.md", "type": "primary" }, { "label": "Inkling checkpoint configuration, pinned", "url": "https://huggingface.co/thinkingmachines/Inkling/blob/828496eeae4c243ff1a22f7f28ff83694f2f7bc9/config.json", "type": "primary" }, { "label": "Inkling chat template, pinned", "url": "https://huggingface.co/thinkingmachines/Inkling/blob/828496eeae4c243ff1a22f7f28ff83694f2f7bc9/chat_template.jinja", "type": "primary" }, { "label": "Using Inkling", "url": "https://tinker-docs.thinkingmachines.ai/cookbook/inkling/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/thinkingmachines/inkling.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/thinkingmachines/Inkling", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "975B total and 41B active are developer-reported nominal figures, not exact checkpoint counts. The automatic Hugging Face stored-element total differs and has not been reconciled with the published nominal figure or MTP accounting. The 1,048,576-token checkpoint context differs from Tinker hosted 64K/256K limits.", "field_status": {}, "audited_at": "2026-10-27T05:01:10.752941Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/509.json" }, { "id": "meta/muse-spark-1.1", "slug": "meta--muse-spark-1.1", "name": "Muse Spark 1.1", "developer": "Meta", "developer_id": "meta", "family": "muse", "description": "Meta's proprietary multimodal reasoning model for coding and agentic tasks, with tool and function calling, computer use, and multi-agent orchestration.", "release_date": "2026-07-09", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 816, "active_parameters_b": 42, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "pdf", "video", "audio" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://research.meta.ai/blog/introducing-muse-spark-meta-model-api", "source_type": "primary", "sources": [ { "label": "Introducing Muse Spark 1.1", "url": "https://research.meta.ai/blog/introducing-muse-spark-meta-model-api", "type": "primary" }, { "label": "Models - Meta Model API", "url": "https://dev.meta.ai/docs/models", "type": "primary" }, { "label": "Muse Spark 1.1 Evaluation Report", "url": "https://research.meta.ai/static/muse-spark-1-1-evaluation-report", "type": "primary" }, { "label": "Meta Model API Terms of Service", "url": "https://dev.meta.ai/legal/terms-of-service", "type": "primary" }, { "label": "Meta Model API Model Designation and Geographic Use Policy", "url": "https://dev.meta.ai/legal/geographic-use-policy", "type": "primary" }, { "label": "Muse Spark Open Model Card", "url": "https://huggingface.co/meta-models/Muse-Spark-Open-816B-A42B", "type": "primary" }, { "label": "Introducing Muse Code and Muse Spark 1.2", "url": "https://research.meta.ai/blog/introducing-muse-code-and-muse-spark-1-2", "type": "primary" }, { "label": "Muse Spark Is Now an Open-Weights Model", "url": "https://pytorch.org/blog/muse-spark-open/", "type": "primary" }, { "label": "TorchTitan Update: New SOTA Models, Performance at Scale, TitanRL", "url": "https://pytorch.org/blog/torchtitan-update-new-sota-models-performance-at-scale-titanrl/", "type": "primary" }, { "label": "Cheeky Pint #35: Muse Wonder: Zuckerberg and Wang's $100B AI — transcript", "url": "https://cheekypint.transistor.fm/35/transcript", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/meta/muse-spark-1.1.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Nominal 816B total, 42B active and MoE are inferred from Spark Open's published specifications and Meta's explicit lineage. Open is the 1.2 release with added safety post-training; Zuckerberg confirms Spark 1, 1.1, 1.2 and 1.3 share unchanged pre-training and continued RL. These are not direct 1.1 count disclosures. Muse Spark 1.1 is a proprietary hosted model launched July 9, 2026.", "field_status": {}, "audited_at": "2026-10-27T14:29:15.138415+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/282.json" }, { "id": "openai/gpt-5.6-luna", "slug": "openai--gpt-5.6-luna", "name": "GPT-5.6 Luna", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5.6", "description": "GPT-5.6 reasoning model for high-volume workloads, accepting text and images and generating text.", "release_date": "2026-07-09", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1050000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/gpt-5-6/", "source_type": "primary", "sources": [ { "label": "GPT-5.6 launch", "url": "https://openai.com/index/gpt-5-6/", "type": "primary" }, { "label": "GPT-5.6 Luna API model reference", "url": "https://developers.openai.com/api/docs/models/gpt-5.6-luna.md", "type": "primary" }, { "label": "GPT-5.6 System Card", "url": "https://deploymentsafety.openai.com/gpt-5-6/gpt-5-6.pdf", "type": "primary" }, { "label": "Public Summary of Training Content for GPT-5.6 Luna", "url": "https://cdn.openai.com/pdf/gpt-5-6-luna-eu-ai-act-public-summary-of-training-content.pdf", "type": "primary" }, { "label": "GPT-6 Astra System Card, including GPT-6 Sol and Luna appendix", "url": "https://deploymentsafety.openai.com/gpt-6-astra/gpt-6-astra.pdf", "type": "primary" }, { "label": "GPT-6.1 Sol System Card", "url": "https://deploymentsafety.openai.com/gpt-6-1-sol/gpt-6-1-sol.pdf", "type": "primary" }, { "label": "GPT-6.1 Terra System Card", "url": "https://deploymentsafety.openai.com/gpt-6-1-terra", "type": "primary" }, { "label": "How GPT-5.6 fuses frontier intelligence with frontier efficiency", "url": "https://openai.com/index/gpt-5-6-frontier-intelligence-efficiency/", "type": "primary" }, { "label": "The builder’s guide to GPT-5.6", "url": "https://openai.com/index/builders-guide-to-gpt-5-6/", "type": "primary" }, { "label": "Prompt caching", "url": "https://developers.openai.com/api/docs/guides/prompt-caching.md", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "OpenAI API changelog", "url": "https://developers.openai.com/api/docs/changelog.md", "type": "primary" }, { "label": "OpenAI API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "Previewing GPT-5.6 Sol", "url": "https://openai.com/index/previewing-gpt-5-6-sol/", "type": "primary" }, { "label": "GPT-5.6 August Updates", "url": "https://deploymentsafety.openai.com/gpt-5-6-august-update/gpt-5-6-august-update.pdf", "type": "primary" }, { "label": "BDH-CQ: In-Context Learning with Recurrent Latent Reasoning", "url": "https://arxiv.org/html/2608.09888v1", "type": "primary" }, { "label": "GPT-5.6 Luna - Amazon Bedrock model card", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-56-luna.html", "type": "primary" }, { "label": "Estimating the Parameter Counts of the Claude 5 and GPT-5.6 Families", "url": "https://eigenigma.io/en/articles/estimating-parameter-counts-of-claude-5-and-gpt-5-6/", "type": "secondary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5.6-luna.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "July 9, 2026 is the public/API launch, after the June 26 restricted partner preview. The August ChatGPT update is a distinct product version. The API reference lists only the undated gpt-5.6-luna identifier as its snapshot, so no immutable date-suffixed snapshot is inferred. The documented 1,050,000-token context comprises a 922,000-token maximum input and 128,000-token maximum output. PDF denotes document input transformed into text and page images. Training-data quantities and audiovisual content in the EU summary do not establish model parameter counts or API audio/video support. Total and active parameters remain undisclosed; the training summary identifies no prior-model base from which a nominal size can be inherited. Later comparisons report performance and cost, and IKP estimates describe effective knowledge capacity rather than weight counts. Network architecture remains unverified. No shutdown date for Luna appears in the reviewed deprecations list; endpoint operation was not tested.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "unverified" }, "audited_at": "2026-10-27T17:06:48.655734Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/438.json" }, { "id": "openai/gpt-5.6-sol", "slug": "openai--gpt-5.6-sol", "name": "GPT-5.6 Sol", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5.6", "description": "GPT-5.6 model for reasoning, coding, research, and tool-based work.", "release_date": "2026-07-09", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1050000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/gpt-5-6/", "source_type": "primary", "sources": [ { "label": "GPT-5.6: Frontier intelligence that scales with your ambition", "url": "https://openai.com/index/gpt-5-6/", "type": "primary" }, { "label": "GPT-5.6 Sol — API model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.6-sol.md", "type": "primary" }, { "label": "GPT-5.6 System Card", "url": "https://deploymentsafety.openai.com/gpt-5-6/gpt-5-6.pdf", "type": "primary" }, { "label": "How GPT-5.6 fuses frontier intelligence with frontier efficiency", "url": "https://openai.com/index/gpt-5-6-frontier-intelligence-efficiency/", "type": "primary" }, { "label": "File inputs — OpenAI API", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "Previewing GPT-5.6 Sol: a next-generation model", "url": "https://openai.com/index/previewing-gpt-5-6-sol/", "type": "primary" }, { "label": "OpenAI API changelog", "url": "https://developers.openai.com/api/docs/changelog.md", "type": "primary" }, { "label": "GPT-6 Astra System Card", "url": "https://deploymentsafety.openai.com/gpt-6-astra/gpt-6-astra.pdf", "type": "primary" }, { "label": "Addendum to GPT-6 Astra System Card: GPT-6.1 Sol", "url": "https://deploymentsafety.openai.com/gpt-6-1-sol/gpt-6-1-sol.pdf", "type": "primary" }, { "label": "Kimi K3: Open Frontier Intelligence — original Moonshot technical report", "url": "https://arxiv.org/html/2607.24653v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5.6-sol.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "July 9, 2026 is the public API release; June 26 was a restricted partner preview. The gpt-5.6 alias routes to gpt-5.6-sol; documentation lists no dated Sol snapshot. Context is 1,050,000 tokens, with explicit limits of 922,000 input and 128,000 output. PDF input is API text/page-image extraction. The August 6 ChatGPT update and August 13 Ultrafast service-tier announcement do not identify a new production Sol API checkpoint. Total and active parameters remain undisclosed; knowledge-capacity estimates and separate draft-model experiments are not Sol parameter counts. Exact architecture remains unverified.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "unverified" }, "audited_at": "2026-10-27T17:06:35.997440Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/439.json" }, { "id": "openai/gpt-5.6-terra", "slug": "openai--gpt-5.6-terra", "name": "GPT-5.6 Terra", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5.6", "description": "GPT-5.6 Terra is an OpenAI model with text and image input, text output, configurable reasoning, and tool calling.", "release_date": "2026-07-09", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1050000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/gpt-5-6/", "source_type": "primary", "sources": [ { "label": "GPT-5.6 Terra API model", "url": "https://developers.openai.com/api/docs/models/gpt-5.6-terra.md", "type": "primary" }, { "label": "GPT-5.6 launch, July 9, 2026", "url": "https://openai.com/index/gpt-5-6/", "type": "primary" }, { "label": "OpenAI API changelog", "url": "https://developers.openai.com/api/docs/changelog.md", "type": "primary" }, { "label": "GPT-5.6 System Card", "url": "https://deploymentsafety.openai.com/gpt-5-6/gpt-5-6.pdf", "type": "primary" }, { "label": "Addendum to GPT-6 Astra System Card: GPT-6.1 Terra", "url": "https://deploymentsafety.openai.com/gpt-6-1-terra/gpt-6-1-terra.pdf", "type": "primary" }, { "label": "GPT-5.6 inference efficiency, July 29, 2026", "url": "https://openai.com/index/gpt-5-6-frontier-intelligence-efficiency/", "type": "primary" }, { "label": "Prompt caching guide", "url": "https://developers.openai.com/api/docs/guides/prompt-caching.md", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "GPT-5.6 Preview System Card, June 26, 2026", "url": "https://deploymentsafety.openai.com/gpt-5-6-preview", "type": "primary" }, { "label": "OpenAI EU AI Act resources", "url": "https://help.openai.com/en/articles/12141645-eu-ai-act-openai-resources-and-customer-guidance", "type": "primary" }, { "label": "OpenAI API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "Estimating the Parameter Counts of the Claude 5 and GPT-5.6 Families", "url": "https://eigenigma.io/en/articles/estimating-parameter-counts-of-claude-5-and-gpt-5-6/", "type": "secondary" }, { "label": "Kimi K3: Open Frontier Intelligence", "url": "https://arxiv.org/html/2607.24653v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5.6-terra.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Public product/API release was July 9, 2026, after a limited trusted-partner preview on June 26. The current API page lists only the undated gpt-5.6-terra identifier, with 1,050,000 context, 922,000 input and 128,000 output tokens. Reasoning defaults to medium and supports none through max. PDF is an API file-input format converted to text and page images. Total and active parameter counts remain undisclosed; knowledge-capacity estimates and separate draft-model experiments are not Terra parameter specifications. Production architecture is unverified. No exact API shutdown notice was found; endpoint operation was not tested.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "unverified" }, "audited_at": "2026-10-27T17:09:18.336198Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/440.json" }, { "id": "xai/grok-4.5", "slug": "xai--grok-4.5", "name": "Grok 4.5", "developer": "xAI", "developer_id": "xai", "family": "grok", "description": "Grok reasoning model developed by xAI with Cursor for coding and knowledge work, with text and image input and tool calling.", "release_date": "2026-07-08", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 1500, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "moe", "context_window": 500000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://cursor.com/blog/grok-4-5", "source_type": "primary", "sources": [ { "label": "Joint release and MoE architecture", "url": "https://cursor.com/blog/grok-4-5", "type": "primary" }, { "label": "Native API specifications", "url": "https://docs.x.ai/developers/models/grok-4.5", "type": "primary" }, { "label": "API release dates", "url": "https://docs.x.ai/developers/release-notes", "type": "primary" }, { "label": "Grok 4.5 model card", "url": "https://media.x.ai/v1/website/4p5-5184fdf9.pdf", "type": "primary" }, { "label": "Founder confirms 1.5T Grok 4.5 after release", "url": "https://x.com/elonmusk/status/2078289996323148076", "type": "primary" }, { "label": "Original X record for parameter disclosure", "url": "https://cdn.syndication.twimg.com/tweet-result?id=2078289996323148076&lang=en&token=1", "type": "primary" }, { "label": "Model access terms", "url": "https://x.ai/legal/terms-of-service-enterprise/previous-2026-05-12", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/xai/grok-4.5.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Jointly trained by xAI (SpaceXAI) and Cursor. The 1,500B total is the founder’s nominal 1.5T model size, explicitly identified as Grok 4.5 after release; active parameters and expert configuration are undisclosed. Cursor identifies the architecture as MoE. Public API/Cursor release was July 8, 2026; June 28 was private beta and July 17 added EU API availability. Native API context is 500,000 tokens; Cursor documents a separate 256k product limit.", "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T18:30:16.802050+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/532.json" }, { "id": "openai/gpt-realtime-2.1", "slug": "openai--gpt-realtime-2.1", "name": "GPT-Realtime-2.1", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-realtime", "description": "Realtime speech-to-speech model with configurable reasoning, text and image input, and function calling.", "release_date": "2026-07-06", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 128000, "modalities": { "input": [ "text", "audio", "image" ], "output": [ "text", "audio" ] }, "reasoning": true, "tool_call": true, "source_url": "https://developers.openai.com/api/docs/changelog.md", "source_type": "primary", "sources": [ { "label": "GPT-Realtime-2.1 API reference", "url": "https://developers.openai.com/api/docs/models/gpt-realtime-2.1.md", "type": "primary" }, { "label": "OpenAI API changelog", "url": "https://developers.openai.com/api/docs/changelog.md", "type": "primary" }, { "label": "New Realtime models on the API", "url": "https://community.openai.com/t/new-realtime-models-on-the-api-gpt-realtime-2-1-and-gpt-realtime-2-1-mini/1385896", "type": "primary" }, { "label": "Getting started with the Realtime API", "url": "https://developers.openai.com/api/docs/guides/realtime.md", "type": "primary" }, { "label": "Realtime conversations", "url": "https://developers.openai.com/api/docs/guides/realtime-conversations.md", "type": "primary" }, { "label": "All OpenAI models", "url": "https://developers.openai.com/api/docs/models/all.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "GPT-Realtime-2 API reference — named predecessor", "url": "https://developers.openai.com/api/docs/models/gpt-realtime-2.md", "type": "primary" }, { "label": "Advancing voice intelligence with new models in the API — May 7, 2026", "url": "https://openai.com/index/advancing-voice-intelligence-with-new-models-in-the-api/", "type": "primary" }, { "label": "GPT-Live System Card — July 8, 2026, updated August 4", "url": "https://deploymentsafety.openai.com/gpt-live/gpt-live.pdf", "type": "primary" }, { "label": "GPT-Live-1.1 System Card — October 15, 2026", "url": "https://deploymentsafety.openai.com/gpt-live-1-1/gpt-live-1-1.pdf", "type": "primary" }, { "label": "NemotronLabs VoiceChat: An Open Full-duplex Speech-to-Speech Model with Tool Calling Capabilities — October 1, 2026 revision", "url": "https://arxiv.org/html/2609.21967v2", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-realtime-2.1.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released July 6, 2026 as gpt-realtime-2.1; the only documented snapshot is the undated identifier. Realtime API only, with 128000-token context and 32000 maximum output. Text/audio/image input, text/audio output, configurable reasoning and function calling are documented. Total and active sizes remain undisclosed in the checked sources; the explicitly named predecessor GPT-Realtime-2 also has no published count to inherit. Neural architecture remains unverified. GPT-Live and its delegated backend are separate models.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "unverified" }, "audited_at": "2026-10-27T17:15:47.370607Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/450.json" }, { "id": "openai/gpt-realtime-2.1-mini", "slug": "openai--gpt-realtime-2.1-mini", "name": "GPT-realtime-2.1-mini", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-realtime", "description": null, "release_date": "2026-07-06", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 128000, "modalities": { "input": [ "text", "audio", "image" ], "output": [ "text", "audio" ] }, "reasoning": true, "tool_call": null, "source_url": "https://developers.openai.com/api/docs/changelog", "source_type": "primary", "sources": [ { "label": "Dated publisher API releases", "url": "https://developers.openai.com/api/docs/changelog", "type": "documentation" }, { "label": "Publisher model documentation and snapshot mapping", "url": "https://developers.openai.com/api/docs/models/gpt-realtime-2.1-mini", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:31:17.277Z", "last_seen": "2026-10-29T10:31:17.277Z", "upstream_present": false, "availability": "public", "notes": "Original publisher conversational audio model. " }, { "id": "tencent/hy3", "slug": "tencent--hy3", "name": "Hy3", "developer": "Tencent", "developer_id": "tencent", "family": "Hy", "description": "Tencent Hy text-generation MoE model with optional reasoning and tool calling, released as the successor to Hy3 preview.", "release_date": "2026-07-06", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 295, "active_parameters_b": 21, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.tencent.com/tencent-hunyuan-officially-releases-hy3-advancing-agent-capabilities-and-deeper-product-integration/", "source_type": "primary", "sources": [ { "label": "Tencent Hy3 release announcement", "url": "https://www.tencent.com/tencent-hunyuan-officially-releases-hy3-advancing-agent-capabilities-and-deeper-product-integration/", "type": "primary" }, { "label": "Hy3 official model card (pinned)", "url": "https://huggingface.co/tencent/Hy3/blob/a960ebc3da325ba167f069f76c41eb62c9280d22/README.md", "type": "primary" }, { "label": "Hy3 checkpoint configuration (pinned)", "url": "https://huggingface.co/tencent/Hy3/blob/a960ebc3da325ba167f069f76c41eb62c9280d22/config.json", "type": "primary" }, { "label": "Hy3 Apache-2.0 license (pinned)", "url": "https://huggingface.co/tencent/Hy3/blob/a960ebc3da325ba167f069f76c41eb62c9280d22/LICENSE", "type": "primary" }, { "label": "Hy3 chat template (pinned)", "url": "https://huggingface.co/tencent/Hy3/blob/a960ebc3da325ba167f069f76c41eb62c9280d22/chat_template.jinja", "type": "primary" }, { "label": "Tencent TokenHub model list", "url": "https://intl.cloud.tencent.com/zh/document/product/1300/78934", "type": "primary" }, { "label": "Hugging Face", "url": "https://huggingface.co/tencent/Hy3", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/tencent/hy3.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Tencent reports nominal 295B model/backbone parameters and 21B active parameters; the official card separately lists a 3.8B MTP layer. These are author-reported rounded counts, not an exact complete-checkpoint sum; MTP active parameters are not disclosed. The checkpoint config specifies 262,144 context positions, while TokenHub advertises 256k with separate input/output limits. Hy3 was publicly released on 2026-07-06; the earlier repository creation and later config revision are not release dates.", "parameters_scope": "language_backbone", "active_parameters_scope": "language_backbone", "parameter_components_b": { "language_backbone": 295, "mtp": 3.8 }, "active_parameters_by_phase_b": { "language_backbone": 21 }, "field_status": {}, "audited_at": "2026-10-27T04:50:47.379392Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/503.json" }, { "id": "poolside/laguna-xs-2.1", "slug": "poolside--laguna-xs-2.1", "name": "Laguna XS 2.1", "developer": "Poolside", "developer_id": "poolside", "family": "laguna", "description": "Open-weight text model for coding agents, with 33B total parameters, 3B active per token, optional reasoning and tool calling.", "release_date": "2026-07-02", "date_precision": "day", "open_weights": true, "license": "OpenMDW-1.1", "parameters_b": 33, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://poolside.ai/blog/introducing-laguna-xs-2-1", "source_type": "primary", "sources": [ { "label": "Introducing Laguna XS 2.1", "url": "https://poolside.ai/blog/introducing-laguna-xs-2-1", "type": "primary" }, { "label": "Laguna XS 2.1 pinned model card", "url": "https://huggingface.co/poolside/Laguna-XS-2.1/blob/c5f36269bbdbd3f27fddc9a9f9dbae0cf2cf57db/README.md", "type": "primary" }, { "label": "Laguna XS 2.1 pinned configuration", "url": "https://huggingface.co/poolside/Laguna-XS-2.1/blob/c5f36269bbdbd3f27fddc9a9f9dbae0cf2cf57db/config.json", "type": "weights" }, { "label": "All pinned BF16 safetensors shard headers", "url": "https://huggingface.co/poolside/Laguna-XS-2.1/tree/c5f36269bbdbd3f27fddc9a9f9dbae0cf2cf57db", "type": "weights" }, { "label": "Pinned Laguna model implementation", "url": "https://huggingface.co/poolside/Laguna-XS-2.1/blob/c5f36269bbdbd3f27fddc9a9f9dbae0cf2cf57db/modeling_laguna.py", "type": "weights" }, { "label": "Pinned Laguna configuration implementation", "url": "https://huggingface.co/poolside/Laguna-XS-2.1/blob/c5f36269bbdbd3f27fddc9a9f9dbae0cf2cf57db/configuration_laguna.py", "type": "weights" }, { "label": "Pinned Laguna XS 2.1 chat template", "url": "https://huggingface.co/poolside/Laguna-XS-2.1/blob/c5f36269bbdbd3f27fddc9a9f9dbae0cf2cf57db/chat_template.jinja", "type": "weights" }, { "label": "Laguna XS 2.1 OpenMDW license", "url": "https://huggingface.co/poolside/Laguna-XS-2.1/blob/c5f36269bbdbd3f27fddc9a9f9dbae0cf2cf57db/LICENSE.md", "type": "weights" }, { "label": "Laguna XS 2.1 repository history", "url": "https://huggingface.co/api/models/poolside/Laguna-XS-2.1/commits/main", "type": "primary" }, { "label": "Later Laguna S 2.1 announcement and named comparisons", "url": "https://poolside.ai/blog/introducing-laguna-s-2-1", "type": "primary" }, { "label": "Laguna M.1/XS.2 technical report linked by XS 2.1", "url": "https://poolside.ai/assets/laguna/laguna-m1-xs2-technical-report.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/poolside/laguna-xs-2.1.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "33B total and 3B active are Poolside’s rounded figures. All 14 pinned BF16 shards reconcile with the model code:33,442,617,088 stored elements include 9,984 non-gradient router correction values; no separate DFlash draft is included. The supported 262,144-token context uses YaRN on global-attention layers with an 8,192-token original-position setting. Model weights use OpenMDW-1.1; Apache notices in implementation files are separate. Reasoning is configurable; the pinned template defaults it off.", "field_status": {}, "audited_at": "2026-10-27T03:47:29.789304Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/472.json" }, { "id": "celeris/celeris-1", "slug": "celeris--celeris-1", "name": "Celeris-1", "developer": "Celeris", "developer_id": "celeris", "family": "celeris", "description": null, "release_date": "2026-07", "date_precision": "month", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "diffusion", "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://celeris.ai/diffusion-decoding-in-production", "source_type": "primary", "sources": [ { "label": "Publisher release evidence", "url": "https://celeris.ai/diffusion-decoding-in-production", "type": "primary" }, { "label": "Celeris model documentation", "url": "https://docs.celeris.ai/models", "type": "primary" }, { "label": "Celeris reasoning and tools", "url": "https://docs.celeris.ai/making-requests", "type": "primary" }, { "label": "Company and API identity", "url": "https://celeris.ai/privacy", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:07:44.790Z", "last_seen": "2026-10-30T07:07:44.790Z", "upstream_present": false, "availability": "public", "notes": "Hosted hybrid diffusion language model. Total and active parameter counts are not disclosed in the reviewed primary documentation; no downloadable checkpoint is documented. July22 publisher article explicitly confirms API availability; exact first-access day remains unresolved, so month precision is retained." }, { "id": "anthropic/claude-sonnet-5", "slug": "anthropic--claude-sonnet-5", "name": "Claude Sonnet 5", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude-sonnet", "description": "Claude Sonnet model with adaptive thinking and a 1M-token context window.", "release_date": "2026-06-30", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1000000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.anthropic.com/news/claude-sonnet-5", "source_type": "primary", "sources": [ { "label": "Introducing Claude Sonnet 5", "url": "https://www.anthropic.com/news/claude-sonnet-5", "type": "primary" }, { "label": "Claude Sonnet 5 - Claude Platform Docs", "url": "https://platform.claude.com/docs/en/models/sonnet-5/overview", "type": "primary" }, { "label": "Claude Sonnet 5 System Card", "url": "https://www.anthropic.com/claude-sonnet-5-system-card", "type": "primary" }, { "label": "GLM-5.3 and the spread of advanced cyber capabilities", "url": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities", "type": "primary" }, { "label": "Commercial Terms of Service", "url": "https://www.anthropic.com/legal/commercial-terms", "type": "primary" }, { "label": "PDF support", "url": "https://platform.claude.com/docs/en/build-with-claude/pdf-support", "type": "primary" }, { "label": "Tool use with Claude", "url": "https://platform.claude.com/docs/en/agents-and-tools/tool-use/overview", "type": "primary" }, { "label": "Anthropic Transparency Hub - Claude Sonnet 5 summary", "url": "https://www.anthropic.com/transparency", "type": "primary" }, { "label": "Claude Sonnet 5.5 System Card", "url": "https://www.anthropic.com/claude-sonnet-5-5-system-card", "type": "primary" }, { "label": "Claude Haiku 5.5 System Card", "url": "https://www.anthropic.com/claude-haiku-5-5-system-card", "type": "primary" }, { "label": "Claude Opus 5.5 System Card", "url": "https://www.anthropic.com/claude-opus-5-5-system-card", "type": "primary" }, { "label": "Claude Opus 5 System Card", "url": "https://www.anthropic.com/claude-opus-5-system-card", "type": "primary" }, { "label": "Claude Fable 5.1 and Claude Mythos 5.1 System Card", "url": "https://www.anthropic.com/claude-fable-5-1-mythos-5-1-system-card", "type": "primary" }, { "label": "Automated Researchers Can Mitigate Well-Characterized Alignment Failures", "url": "https://alignment.anthropic.com/2026/automated-alignment-researchers/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/anthropic/claude-sonnet-5.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Nominal total and active parameter counts remain undisclosed; third-party knowledge-capacity estimates are not weight counts. Architecture remains undisclosed. Proprietary describes nonpublic weights accessed under commercial service terms. PDF input is document processing through text and vision; output is text.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T10:49:45.887222+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/115.json" }, { "id": "google/gemini-3.1-flash-lite-image", "slug": "google--gemini-3.1-flash-lite-image", "name": "Gemini 3.1 flash lite image", "developer": "Google", "developer_id": "google", "family": "gemini", "description": null, "release_date": "2026-06-30", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 65536, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text", "image" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.google.dev/gemini-api/docs/changelog", "source_type": "primary", "sources": [ { "label": "Dated publisher API releases", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "documentation" }, { "label": "Publisher interleaved text and image capability", "url": "https://ai.google.dev/gemini-api/docs/image-generation", "type": "documentation" }, { "label": "Publisher model specifications", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-lite-image", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:31:17.277Z", "last_seen": "2026-10-29T10:31:17.277Z", "upstream_present": false, "availability": "public", "notes": "Generates interleaved text and images. Context and capabilities describe current published model documentation; preview limits may have differed." }, { "id": "meituan/longcat-2.0", "slug": "meituan--longcat-2.0", "name": "LongCat-2.0", "developer": "Meituan", "developer_id": "meituan", "family": "longcat", "description": "Meituan LongCat-2.0 is an open-weight MoE language model with nominal 1.6T total parameters and about 48B activated per token, LongCat Sparse Attention, N-gram embeddings, reasoning and tool calling, and a documented 1,048,576-token API context window.", "release_date": "2026-06-30", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 1600, "active_parameters_b": 48, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://tech.meituan.com/2026/06/30/LongCat2.0.html", "source_type": "primary", "sources": [ { "label": "Meituan LongCat-2.0 release announcement", "url": "https://tech.meituan.com/2026/06/30/LongCat2.0.html", "type": "primary" }, { "label": "Official LongCat-2.0 model card and weight license", "url": "https://huggingface.co/meituan-longcat/LongCat-2.0/blob/834bf5ffe3047aa9f6cc7a64a9bc068b146b8274/README.md", "type": "primary" }, { "label": "Audited BF16 checkpoint configuration", "url": "https://huggingface.co/meituan-longcat/LongCat-2.0/blob/834bf5ffe3047aa9f6cc7a64a9bc068b146b8274/config.json", "type": "weights" }, { "label": "Audited full checkpoint index", "url": "https://huggingface.co/meituan-longcat/LongCat-2.0/blob/834bf5ffe3047aa9f6cc7a64a9bc068b146b8274/model.safetensors.index.json", "type": "weights" }, { "label": "LongCat API context limits", "url": "https://longcat.chat/platform/docs/api/model-params", "type": "primary" }, { "label": "Official MTP embedding sharing implementation", "url": "https://github.com/meituan-longcat/SGLang-FluentLLM/blob/55d719f3223f697cbeb4a5f31b7071e1892980d0/python/sglang/srt/speculative/base_spec_worker.py", "type": "primary" }, { "label": "Meituan open-weight release announcement", "url": "https://www.meituan.com/news/NN260706216009547", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/meituan/longcat-2.0.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Nominal 1.6T parameters; average ~48B active per token (dynamic 33B–56B). Audited BF16/F32 checkpoint 834bf5ffe3047aa9f6cc7a64a9bc068b146b8274 stores 1,775.560491136B elements including separately saved MTP/embedding namespaces; this is not asserted as an exact unique learned count. Documented API context 1,048,576 differs from checkpoint max_position_embeddings 262,144. Final model released 2026-06-30; weights announced 2026-07-06.", "field_status": {}, "audited_at": "2026-10-26T21:16:42.280176+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/252.json" }, { "id": "deepreinforce/ornith-1.0-31b", "slug": "deepreinforce--ornith-1.0-31b", "name": "Ornith 1.0 31B", "developer": "DeepReinforce", "developer_id": "deepreinforce", "family": "ornith", "description": "Announced 31B dense Ornith 1.0 variant for agentic coding; public checkpoint availability remains unverified.", "release_date": "2026-06-25", "date_precision": "day", "open_weights": false, "license": "MIT", "parameters_b": 31, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://raw.githubusercontent.com/ornith-ai/Ornith-1/16d32b11211ac83373d37dc67234f77e77f909a1/README.md", "source_type": "primary", "sources": [ { "label": "Ornith 1.0 developer announcement", "url": "https://ornith.ai/ornith_1_0.html", "type": "primary" }, { "label": "Original developer README including 31B", "url": "https://raw.githubusercontent.com/ornith-ai/Ornith-1/16d32b11211ac83373d37dc67234f77e77f909a1/README.md", "type": "primary" }, { "label": "Current official checkpoint documentation", "url": "https://github.com/ornith-ai/Ornith-1/blob/main/README.md", "type": "primary" }, { "label": "Official Ornith 1.0 weights collection", "url": "https://huggingface.co/collections/ornith-ai/ornith-10", "type": "weights" }, { "label": "Official public model listing", "url": "https://huggingface.co/api/models?author=ornith-ai&limit=100", "type": "primary" }, { "label": "Later first-party comparison", "url": "https://ornith.ai/ornith_1_5.html", "type": "primary" }, { "label": "Announcement README history", "url": "https://api.github.com/repos/ornith-ai/Ornith-1/commits?path=README.md&per_page=100", "type": "primary" }, { "label": "Ornith official distribution and serving instructions", "url": "https://raw.githubusercontent.com/ornith-ai/Ornith-1/main/README.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepreinforce/ornith-1.0-31b.toml", "type": "catalog" }, { "label": "Announcement", "url": "https://deep-reinforce.com/ornith_1_0.html", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "31B is the developer-announced nominal total; active parameters are not separately disclosed. June 25, 2026 is the announced release date, with MIT licensing. No public 31B checkpoint was found: current official distribution lists omit it and its metadata endpoint is inaccessible. Context, image input, dedicated thinking and native tool calling remain unverified for this variant. Classified closed for current public-download availability; the announced MIT license is preserved.", "field_status": { "active_parameters_b": "undisclosed", "context_window": "unverified", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-28T07:54:23.586828Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/161.json" }, { "id": "deepreinforce/ornith-1.0-35b", "slug": "deepreinforce--ornith-1.0-35b", "name": "Ornith 1.0 35B", "developer": "DeepReinforce", "developer_id": "deepreinforce", "family": "ornith", "description": "Qwen3.5-based mixture-of-experts model post-trained with self-scaffolding reinforcement learning for agentic coding, with text, image and video inputs.", "release_date": "2026-06-25", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 35.107181936, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://raw.githubusercontent.com/ornith-ai/Ornith-1/16d32b11211ac83373d37dc67234f77e77f909a1/README.md", "source_type": "primary", "sources": [ { "label": "Ornith-1.0-35B model card at audited revision", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-35B/raw/5df2ed3f675c7beaa490328cc70bb573b65fb660/README.md", "type": "primary" }, { "label": "Ornith-1.0: Self-Scaffolding LLMs for Agentic Coding", "url": "https://ornith.ai/ornith_1_0.html", "type": "primary" }, { "label": "Ornith-1.0-35B checkpoint configuration", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-35B/raw/5df2ed3f675c7beaa490328cc70bb573b65fb660/config.json", "type": "weights" }, { "label": "Exact checkpoint tensor index and indexed BF16 shard headers", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-35B/raw/5df2ed3f675c7beaa490328cc70bb573b65fb660/model.safetensors.index.json", "type": "weights" }, { "label": "Exact checkpoint multimodal processor configuration", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-35B/raw/5df2ed3f675c7beaa490328cc70bb573b65fb660/processor_config.json", "type": "weights" }, { "label": "Exact checkpoint chat template", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-35B/raw/5df2ed3f675c7beaa490328cc70bb573b65fb660/chat_template.jinja", "type": "weights" }, { "label": "Later first-party comparison identifying the exact Ornith 1.0 35B-A3B predecessor", "url": "https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B/raw/10fbf86fed7ecee4a061f8b499a618f46001cac1/README.md", "type": "primary" }, { "label": "Transformers implementation for the checkpoint architecture", "url": "https://raw.githubusercontent.com/huggingface/transformers/v5.8.1/src/transformers/models/qwen3_5_moe/modeling_qwen3_5_moe.py", "type": "primary" }, { "label": "Preliminary exact-model active-count claim; obsolete architecture rejected", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-35B/raw/09d9ab0f05519cab3787dd4995ddd49c691e9abd/README.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepreinforce/ornith-1.0-35b.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B", "type": "weights" }, { "label": "Announcement", "url": "https://deep-reinforce.com/ornith_1_0.html", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Total 35,107,181,936 is the complete BF16 checkpoint learned-parameter count, including 446,571,248 vision parameters; nominal model size is 35B. Active 3B is the publisher nominal A3B class, not an exact count or a statement that all multimodal components activate on every text token. Released architecture is 40 layers, 256 routed experts/top 8; early draft 48-layer/128-expert dimensions are obsolete. The index total_parameters=664944 is erroneous; all 16 shard headers and the BF16 byte total support the complete count.", "field_status": {}, "audited_at": "2026-10-26T18:48:39.012848+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/162.json" }, { "id": "deepreinforce/ornith-1.0-397b", "slug": "deepreinforce--ornith-1.0-397b", "name": "Ornith 1.0 397B", "developer": "DeepReinforce", "developer_id": "deepreinforce", "family": "ornith", "description": "Mixture-of-experts coding and reasoning model trained with reinforcement learning to generate task scaffolds and software-task solutions.", "release_date": "2026-06-25", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 396.802360816, "active_parameters_b": 17, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://raw.githubusercontent.com/ornith-ai/Ornith-1/16d32b11211ac83373d37dc67234f77e77f909a1/README.md", "source_type": "primary", "sources": [ { "label": "Ornith-1.0-397B model card", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-397B", "type": "primary" }, { "label": "Exact 397B checkpoint configuration", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-397B/raw/5e3e761811e804c295c1d3c0ce68b21da6154209/config.json", "type": "primary" }, { "label": "Exact BF16 checkpoint index and indexed tensor headers", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-397B/raw/5e3e761811e804c295c1d3c0ce68b21da6154209/model.safetensors.index.json", "type": "primary" }, { "label": "Hugging Face metadata for official checkpoint", "url": "https://huggingface.co/api/models/ornith-ai/Ornith-1.0-397B", "type": "primary" }, { "label": "Exact 397B native chat template", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-397B/raw/5e3e761811e804c295c1d3c0ce68b21da6154209/chat_template.jinja", "type": "primary" }, { "label": "Exact 397B image and video processors", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-397B/raw/5e3e761811e804c295c1d3c0ce68b21da6154209/processor_config.json", "type": "primary" }, { "label": "Ornith-1.0 announcement", "url": "https://ornith.ai/ornith_1_0.html", "type": "primary" }, { "label": "397B checkpoint commit history", "url": "https://huggingface.co/api/models/ornith-ai/Ornith-1.0-397B/commits/main", "type": "primary" }, { "label": "Dated original family announcement commit", "url": "https://api.github.com/repos/ornith-ai/Ornith-1/commits/16d32b11211ac83373d37dc67234f77e77f909a1", "type": "primary" }, { "label": "Later first-party comparison table", "url": "https://huggingface.co/ornith-ai/Ornith-1.5-397B", "type": "primary" }, { "label": "Qwen3.5 MoE implementation", "url": "https://raw.githubusercontent.com/huggingface/transformers/main/src/transformers/models/qwen3_5_moe/modeling_qwen3_5_moe.py", "type": "primary" }, { "label": "Original Qwen3.5-397B-A17B model card", "url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "type": "primary" }, { "label": "Original Qwen3.5-397B-A17B configuration", "url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B/raw/main/config.json", "type": "primary" }, { "label": "Original Qwen3.5-397B-A17B checkpoint index", "url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B/raw/main/model.safetensors.index.json", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepreinforce/ornith-1.0-397b.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B", "type": "weights" }, { "label": "Announcement", "url": "https://deep-reinforce.com/ornith_1_0.html", "type": "announcement" }, { "label": "Hugging Face (FP8)", "url": "https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B-FP8", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete released total: 396.802360816B, including 0.45601048B vision parameters. The nominal 17B active language-model count is inferred from documented Qwen3.5 post-training lineage and matching main-network architecture; it excludes image encoding and optional speculative heads. No MTP weights are distributed. Native text/image/video interfaces are present, but practical vision reliability remains unresolved. MIT is the stated license; the linked standalone license file is absent. The card’s single-GPU wording and historical A3B heading conflict with the released model.", "field_status": {}, "audited_at": "2026-10-27T12:36:17.590764+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/163.json", "parameter_components_b": { "language_model": 396.346350336, "vision_encoder": 0.45601048 }, "active_parameters_scope": "language_model" }, { "id": "deepreinforce/ornith-1.0-9b", "slug": "deepreinforce--ornith-1.0-9b", "name": "Ornith 1.0 9B", "developer": "DeepReinforce", "developer_id": "deepreinforce", "family": "ornith", "description": "Open dense Qwen3.5-based reasoning model trained with self-scaffolding reinforcement learning for agentic coding.", "release_date": "2026-06-25", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 9.409813744, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://raw.githubusercontent.com/ornith-ai/Ornith-1/16d32b11211ac83373d37dc67234f77e77f909a1/README.md", "source_type": "primary", "sources": [ { "label": "Ornith-1.0-9B model card", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-9B/raw/83dc1f5e24ef8527af019a6b3bf66ac0f1c2c999/README.md", "type": "primary" }, { "label": "Exact Ornith-1.0-9B configuration", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-9B/raw/83dc1f5e24ef8527af019a6b3bf66ac0f1c2c999/config.json", "type": "primary" }, { "label": "Exact checkpoint tensor index", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-9B/resolve/83dc1f5e24ef8527af019a6b3bf66ac0f1c2c999/model.safetensors.index.json", "type": "weights" }, { "label": "Checkpoint shard 1 safetensors header", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-9B/resolve/83dc1f5e24ef8527af019a6b3bf66ac0f1c2c999/model-00001-of-00004.safetensors", "type": "weights" }, { "label": "Checkpoint shard 2 safetensors header", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-9B/resolve/83dc1f5e24ef8527af019a6b3bf66ac0f1c2c999/model-00002-of-00004.safetensors", "type": "weights" }, { "label": "Checkpoint shard 3 safetensors header", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-9B/resolve/83dc1f5e24ef8527af019a6b3bf66ac0f1c2c999/model-00003-of-00004.safetensors", "type": "weights" }, { "label": "Checkpoint shard 4 safetensors header", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-9B/resolve/83dc1f5e24ef8527af019a6b3bf66ac0f1c2c999/model-00004-of-00004.safetensors", "type": "weights" }, { "label": "Exact multimodal processor configuration", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-9B/raw/83dc1f5e24ef8527af019a6b3bf66ac0f1c2c999/processor_config.json", "type": "primary" }, { "label": "Exact model chat template", "url": "https://huggingface.co/ornith-ai/Ornith-1.0-9B/raw/83dc1f5e24ef8527af019a6b3bf66ac0f1c2c999/chat_template.jinja", "type": "primary" }, { "label": "Original developer release README", "url": "https://raw.githubusercontent.com/ornith-ai/Ornith-1/16d32b11211ac83373d37dc67234f77e77f909a1/README.md", "type": "primary" }, { "label": "Developer release README commit metadata", "url": "https://api.github.com/repos/ornith-ai/Ornith-1/commits/16d32b11211ac83373d37dc67234f77e77f909a1", "type": "primary" }, { "label": "Exact model repository history", "url": "https://huggingface.co/api/models/ornith-ai/Ornith-1.0-9B/commits/main", "type": "primary" }, { "label": "Later Ornith-1.5 announcement and comparison tables", "url": "https://ornith.ai/ornith_1_5.html", "type": "primary" }, { "label": "Transformers Qwen3.5 implementation matching checkpoint version", "url": "https://raw.githubusercontent.com/huggingface/transformers/v5.8.1/src/transformers/models/qwen3_5/modeling_qwen3_5.py", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepreinforce/ornith-1.0-9b.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B", "type": "weights" }, { "label": "Announcement", "url": "https://deep-reinforce.com/ornith_1_0.html", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The complete checkpoint contains 9.409813744B learned parameters: 8.953803264B text model and untied head, plus 0.456010480B vision encoder/projector. Its index total_parameters field is erroneous; tensor shapes and total_size support the retained count. Active parameters are not separately disclosed. The dense Qwen3.5 hybrid has 24 linear-attention and 8 full-attention layers. Native inputs include text, images and video; hosted endpoints may expose fewer. June 25, 2026 is the documented weight release. Distribution migrated from deepreinforce-ai to ornith-ai.", "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T12:36:11.593862+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/164.json" }, { "id": "liquid/lfm2.5-230m", "slug": "liquid--lfm2.5-230m", "name": "LFM2.5-230M", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm", "description": null, "release_date": "2026-06-25", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 0.23, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "hybrid", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://www.liquid.ai/blog/lfm2-5-230m", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/lfm2-5-230m", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/LiquidAI/LFM2.5-230M/blob/40cb2ad3b3044d5a41eee083a6103c8b523afa45/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/LiquidAI/LFM2.5-230M/blob/40cb2ad3b3044d5a41eee083a6103c8b523afa45/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:23.459Z", "last_seen": "2026-10-29T10:17:23.459Z", "upstream_present": false, "availability": "public", "notes": "Post-trained checkpoint explicitly included in the developer release. Parameter and context values follow the exact model card; MoE variants also use the LFM convolution/attention hybrid backbone." }, { "id": "liquid/lfm2.5-230m-base", "slug": "liquid--lfm2.5-230m-base", "name": "LFM2.5-230M-Base", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm", "description": null, "release_date": "2026-06-25", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 0.23, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "hybrid", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.liquid.ai/blog/lfm2-5-230m", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/lfm2-5-230m", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/LiquidAI/LFM2.5-230M-Base/blob/9d2be5519834990d30996f878b6771cccbd24f2c/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/LiquidAI/LFM2.5-230M-Base/blob/9d2be5519834990d30996f878b6771cccbd24f2c/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:23.459Z", "last_seen": "2026-10-29T10:17:23.459Z", "upstream_present": false, "availability": "public", "notes": "Pretrained checkpoint explicitly included in the developer release. Parameter and context values follow the exact model card; MoE variants also use the LFM convolution/attention hybrid backbone." }, { "id": "bytedance-seed/seed-2.1-pro", "slug": "bytedance-seed--seed-2.1-pro", "name": "Seed 2.1 Pro", "developer": "ByteDance", "developer_id": "bytedance-seed", "family": "seed", "description": "ByteDance Seed 2.1 model for multimodal reasoning, coding, and tool-assisted tasks.", "release_date": "2026-06-23", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1024000, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "source_type": "primary", "sources": [ { "label": "Seed2.1 Officially Released: Advancing AI Productivity", "url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "type": "primary" }, { "label": "Seed2.1 Model Card: Agentic Intelligence for Productivity", "url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "type": "primary" }, { "label": "Volcano Ark model list (updated 2026-10-25)", "url": "https://console.volcengine.com/ark/region:cn-beijing/docs/model-list", "type": "primary" }, { "label": "Doubao Model Service Agreement", "url": "https://docs.volcengine.com/docs/ark/doubao-model-service-agreement?lang=zh", "type": "primary" }, { "label": "Seed2.1 current model overview and comparison tables", "url": "https://seed.bytedance.com/en/seed2_1", "type": "primary" }, { "label": "Volcano Ark model list - official Markdown", "url": "https://ark.volcengine.com/region:cn-beijing/docs/model-list.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/bytedance-seed/seed-2.1-pro.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Context is nominal 1024K for current Pro revision doubao-seed-2-1-pro-260915; the earlier 260628 revision is 256K. The canonical record is undated. June 23 is the initial model launch, not either revision date. Nominal total and active counts remain undisclosed; no publicly sized base relationship is documented. Production architecture remains undisclosed; audio input is not established for Pro.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T11:03:13.621608+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/136.json" }, { "id": "bytedance-seed/seed-2.1-turbo", "slug": "bytedance-seed--seed-2.1-turbo", "name": "Seed 2.1 Turbo", "developer": "ByteDance", "developer_id": "bytedance-seed", "family": "seed", "description": "ByteDance Seed 2.1 model for multimodal reasoning, coding, and tool calling.", "release_date": "2026-06-23", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 256000, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://seed.bytedance.com/en/seed2_1", "source_type": "primary", "sources": [ { "label": "Seed2.1 official model page", "url": "https://seed.bytedance.com/en/seed2_1", "type": "primary" }, { "label": "Seed2.1 Officially Released: Advancing AI Productivity", "url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "type": "primary" }, { "label": "Seed2.1 Model Card: Agentic Intelligence for Productivity", "url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "type": "primary" }, { "label": "ModelArk model list", "url": "https://docs.byteplus.com/en/docs/modelark/model-list", "type": "primary" }, { "label": "Volcano Engine Ark model list", "url": "https://www.volcengine.com/docs/82379/1330310", "type": "primary" }, { "label": "ModelArk quick start", "url": "https://docs.byteplus.com/en/docs/modelark/quick-start?redirect=1", "type": "primary" }, { "label": "Video understanding", "url": "https://docs.byteplus.com/en/docs/modelark/video-understanding?redirect=1", "type": "primary" }, { "label": "Service Specific Terms — ModelArk and Model Services", "url": "https://docs.byteplus.com/en/docs/legal/docs-service-specific-terms", "type": "primary" }, { "label": "Seed2.1 original launch with exact variants", "url": "https://seed.bytedance.com/zh/blog/seed2-1-officially-released-advancing-ai-productivity", "type": "primary" }, { "label": "ByteDance Seed current model directory", "url": "https://seed.bytedance.com/en/models", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/bytedance-seed/seed-2.1-turbo.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Context is documented as nominal 256K; 256000 is a display normalization, not an independently documented exact token ceiling. Current first-party Turbo specifications refer to doubao-seed-2-1-turbo-260628 / dola-seed-2-1-turbo-260628. The 2026-06-23 date is the initial Seed2.1 launch; 260628 identifies the later API revision. Nominal total and active counts remain undisclosed. ByteDance identifies Pro and Turbo as different-sized models; no publicly sized base relationship is established. Neural architecture remains undisclosed. Audio understanding is explicitly unsupported in the current BytePlus Turbo table.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T11:04:05.797360+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/137.json" }, { "id": "sakana/fugu", "slug": "sakana--fugu", "name": "Fugu", "developer": "Sakana AI", "developer_id": "sakana", "family": "fugu", "description": "Learned router that selects a worker model for each input, exposed through a text-and-image API.", "release_date": "2026-06-22", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1000000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://sakana.ai/fugu-release/", "source_type": "primary", "sources": [ { "label": "Sakana Fugu: One Model to Command Them All", "url": "https://sakana.ai/fugu-release/", "type": "announcement" }, { "label": "Sakana Fugu Technical Report, v2, sections 3.1 and4", "url": "https://arxiv.org/html/2606.21228v2", "type": "paper" }, { "label": "Toward backbone-independent orchestration: validation of Gemma 4-based Sakana Fugu", "url": "https://sakana.ai/fugu-gemma4/", "type": "docs" }, { "label": "Pinned official Fugu integration model catalog", "url": "https://raw.githubusercontent.com/SakanaAI/fugu/8023e4dd279339ada5d0e6ac2ffe4416da62e030/configs/files/fugu.json", "type": "docs" }, { "label": "Sakana API models and supported endpoints", "url": "https://console.sakana.ai/models", "type": "docs" }, { "label": "Sakana AI API Platform Terms of Service", "url": "https://console.sakana.ai/terms-of-service", "type": "docs" }, { "label": "Sakana Fugu beta announcement", "url": "https://sakana.ai/fugu-beta/", "type": "announcement" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/sakana/fugu.toml", "type": "catalog" }, { "label": "Official model catalog", "url": "https://raw.githubusercontent.com/SakanaAI/fugu/refs/heads/main/configs/files/fugu.json", "type": "docs" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "June 22, 2026 is the documented general release of bare Fugu; the April 24 beta used the Fugu Mini name, whose revision continuity is unconfirmed. Fugu is a learned router that selects one worker per input. Its earlier backbone is identified only as Qwen-based; Trinity's 0.6B coordinator and the separate Gemma 4 E2B validation do not establish Fugu's deployed size. Total and active parameter counts and exact neural topology remain undisclosed. The 1M context, text/image input, reasoning and tool support describe the hosted service interface.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T17:46:19.723627Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/482.json" }, { "id": "moonshotai/kimi-k2.7-code-highspeed", "slug": "moonshotai--kimi-k2.7-code-highspeed", "name": "Kimi K2.7 Code HighSpeed", "developer": "Moonshot AI", "developer_id": "moonshotai", "family": "kimi-k2", "description": "High-throughput serving variant of Kimi K2.7 Code, using the same multimodal reasoning model for coding and tool use.", "release_date": "2026-06-15", "date_precision": "day", "open_weights": true, "license": "Modified MIT", "parameters_b": 1026.879376368, "active_parameters_b": 32, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://platform.kimi.ai/docs/guide/kimi-k2-7-code-quickstart", "source_type": "primary", "sources": [ { "label": "Official API guide and exact HighSpeed mapping", "url": "https://platform.kimi.ai/docs/guide/kimi-k2-7-code-quickstart", "type": "primary" }, { "label": "Hugging Face", "url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "type": "weights" }, { "label": "Official checkpoint license", "url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code/blob/main/LICENSE", "type": "primary" }, { "label": "Underlying checkpoint configuration at pinned revision", "url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code/raw/74797c9c62378b951a1f6fcf5c4631024e9b8bef/config.json", "type": "primary" }, { "label": "June 15 HighSpeed beta announcement", "url": "https://x.com/Kimi_Moonshot/status/2066467110960959833", "type": "primary" }, { "label": "Launch announcement accessible first-party embed", "url": "https://publish.twitter.com/oembed?url=https%3A%2F%2Fx.com%2FKimi_Moonshot%2Fstatus%2F2066467110960959833", "type": "primary" }, { "label": "Official release chronology", "url": "https://www.kimi.com/code/docs/en/kimi-code/whats-new.html", "type": "primary" }, { "label": "Official HighSpeed capabilities", "url": "https://www.kimi.ai/help/kimi-code/models", "type": "primary" }, { "label": "Underlying Code checkpoint complete tensor index at pinned revision", "url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code/raw/74797c9c62378b951a1f6fcf5c4631024e9b8bef/model.safetensors.index.json", "type": "primary" }, { "label": "Underlying Code language implementation at pinned revision", "url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code/raw/74797c9c62378b951a1f6fcf5c4631024e9b8bef/modeling_deepseek.py", "type": "primary" }, { "label": "Underlying Code vision and projector implementation at pinned revision", "url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code/raw/74797c9c62378b951a1f6fcf5c4631024e9b8bef/modeling_kimi_k25.py", "type": "primary" }, { "label": "Underlying Code pinned safetensors headers and census review", "url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code/resolve/74797c9c62378b951a1f6fcf5c4631024e9b8bef/model-00001-of-000064.safetensors", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/moonshotai/kimi-k2.7-code-highspeed.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Moonshot explicitly identifies HighSpeed as the same K2.7 Code model served faster. Exact released underlying checkpoint total is 1026.879376368B: language 1026.408232448B, vision encoder 0.416866032B, projector 0.054277888B. The complete pinned census reconstructs logical INT4 weights, excludes scales/shape metadata and nonpersistent buffers, and includes router correction-bias parameters. This is not a separate census of hosted HighSpeed infrastructure. The card total 1T and retained active count 32B are nominal. Open weights and Modified MIT apply to the underlying checkpoint, not the hosted acceleration stack or a self-hosted speed guarantee. June 15, 2026 was public beta rollout; July 9 was Kimi Code general availability.", "parameter_components_b": { "language_backbone": 1026.408232448, "vision_encoder": 0.416866032, "vision_projector": 0.054277888 }, "field_status": {}, "audited_at": "2026-10-27T00:03:20.730855Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/362.json" }, { "id": "zhipuai/glm-5.2", "slug": "zhipuai--glm-5.2", "name": "GLM-5.2", "developer": "Z.ai", "developer_id": "zhipuai", "family": "glm", "description": "Text MoE model with IndexShare sparse attention, a 1M-token context and configurable reasoning effort.", "release_date": "2026-06-13", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 744, "active_parameters_b": 40, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://x.com/Zai_org/status/2065704919299235870", "source_type": "primary", "sources": [ { "label": "Original GLM-5.2 Coding Plan availability announcement", "url": "https://x.com/Zai_org/status/2065704919299235870", "type": "announcement" }, { "label": "Z.ai release notes for the full GLM-5.2 launch", "url": "https://docs.z.ai/release-notes/new-released.md", "type": "announcement" }, { "label": "GLM-5.2 exact checkpoint card", "url": "https://huggingface.co/zai-org/GLM-5.2/blob/cf457fa734ab149ffef225f80893eb38c6ff5cdc/README.md", "type": "weights" }, { "label": "Official GLM family download table at pinned revision", "url": "https://github.com/zai-org/GLM-5/blob/83c03d317378f86f0a012f7c4cd48b6243c7e69b/README.md", "type": "docs" }, { "label": "GLM-5.2 exact checkpoint configuration", "url": "https://huggingface.co/zai-org/GLM-5.2/blob/cf457fa734ab149ffef225f80893eb38c6ff5cdc/config.json", "type": "weights" }, { "label": "GLM-5.2 tokenizer configuration", "url": "https://huggingface.co/zai-org/GLM-5.2/blob/cf457fa734ab149ffef225f80893eb38c6ff5cdc/tokenizer_config.json", "type": "weights" }, { "label": "GLM-5.2 checkpoint MIT license", "url": "https://huggingface.co/zai-org/GLM-5.2/blob/cf457fa734ab149ffef225f80893eb38c6ff5cdc/LICENSE", "type": "weights" }, { "label": "GLM-5.2 exact chat template", "url": "https://huggingface.co/zai-org/GLM-5.2/blob/cf457fa734ab149ffef225f80893eb38c6ff5cdc/chat_template.jinja", "type": "weights" }, { "label": "GLM-5.2 native API guide", "url": "https://docs.z.ai/guides/llm/glm-5.2", "type": "docs" }, { "label": "GLM-5.2 original technical launch article", "url": "https://z.ai/blog/glm-5.2", "type": "announcement" }, { "label": "Earlier GLM-5 technical report checked for counting scope", "url": "https://arxiv.org/html/2602.15763v1", "type": "paper" }, { "label": "Exact GLM-5.2 public checkpoint metadata", "url": "https://huggingface.co/api/models/zai-org/GLM-5.2/revision/cf457fa734ab149ffef225f80893eb38c6ff5cdc", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/zhipuai/glm-5.2.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/GLM-5.2", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "First available to all GLM Coding Plan tiers on June 13, 2026; the broader launch followed June 16. Z.ai reports nominal 744B total and 40B active for this exact model. Its model-specific download row does not resolve embedding/output/MTP accounting; the earlier GLM-5 counting convention is not assumed. The checkpoint uses MoE with MLA, DSA and IndexShare, plus one MTP module. Configured and tokenizer context capacity is 1048576 tokens, advertised as 1M; API maximum output is separate. MIT applies to the public weights.", "field_status": {}, "audited_at": "2026-10-27T06:22:20.974795+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/560.json" }, { "id": "moonshotai/kimi-k2.7-code", "slug": "moonshotai--kimi-k2.7-code", "name": "Kimi K2.7 Code", "developer": "Moonshot AI", "developer_id": "moonshotai", "family": "kimi-k2", "description": "Coding-focused multimodal reasoning model built on Kimi K2.6, with open weights and multi-step tool calling.", "release_date": "2026-06-12", "date_precision": "day", "open_weights": true, "license": "Modified MIT", "parameters_b": 1026.879376368, "active_parameters_b": 32, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.kimi.com/code/docs/en/kimi-code/whats-new.html", "source_type": "primary", "sources": [ { "label": "Official release changelog", "url": "https://www.kimi.com/code/docs/en/kimi-code/whats-new.html", "type": "primary" }, { "label": "Hugging Face", "url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code", "type": "weights" }, { "label": "Checkpoint license", "url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code/raw/main/LICENSE", "type": "primary" }, { "label": "Checkpoint configuration at pinned revision", "url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code/raw/74797c9c62378b951a1f6fcf5c4631024e9b8bef/config.json", "type": "primary" }, { "label": "Official API quickstart", "url": "https://platform.kimi.ai/docs/guide/kimi-k2-7-code-quickstart", "type": "primary" }, { "label": "Official model overview", "url": "https://www.kimi.ai/resources/kimi-k2-7-code", "type": "primary" }, { "label": "Kimi-K2.7-Code complete tensor index at pinned revision", "url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code/raw/74797c9c62378b951a1f6fcf5c4631024e9b8bef/model.safetensors.index.json", "type": "primary" }, { "label": "Kimi-K2.7-Code language-model implementation at pinned revision", "url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code/raw/74797c9c62378b951a1f6fcf5c4631024e9b8bef/modeling_deepseek.py", "type": "primary" }, { "label": "Kimi-K2.7-Code vision and projector implementation at pinned revision", "url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code/raw/74797c9c62378b951a1f6fcf5c4631024e9b8bef/modeling_kimi_k25.py", "type": "primary" }, { "label": "Kimi-K2.7-Code headers of all 64 indexed shards at pinned revision", "url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code/resolve/74797c9c62378b951a1f6fcf5c4631024e9b8bef/model-00001-of-000064.safetensors", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/moonshotai/kimi-k2.7-code.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Exact complete checkpoint: 1026.408232448B language + 0.416866032B vision encoder + 0.054277888B projector = 1026.879376368B. All 64 pinned headers reconcile with independent configuration/code arithmetic; logical INT4 weights are reconstructed, scales and shape metadata excluded, and router correction-bias parameters included. Published 1T total and 32B active are nominal; active remains the developer-reported figure without an exact vision-inclusive claim. Official public-release announcement is June 12, 2026; June 11 upload timestamps alone do not establish earlier public availability. Modified MIT weights; thinking is always enabled.", "parameter_components_b": { "language_backbone": 1026.408232448, "vision_encoder": 0.416866032, "vision_projector": 0.054277888 }, "field_status": {}, "audited_at": "2026-10-27T00:00:18.681595Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/361.json" }, { "id": "google/diffusiongemma-26b-a4b-it", "slug": "google--diffusiongemma-26b-a4b-it", "name": "DiffusionGemma 26B A4B IT", "developer": "Google", "developer_id": "google", "family": "gemma", "description": "Experimental Gemma 4 based MoE model using discrete text diffusion, with multimodal input, reasoning, and function calling.", "release_date": "2026-06-10", "date_precision": "day", "open_weights": true, "license": "Apache 2.0", "parameters_b": 25.823778864, "active_parameters_b": 3.8, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://blog.google/innovation-and-ai/technology/developers-tools/diffusion-gemma-faster-text-generation/", "source_type": "primary", "sources": [ { "label": "Google launch announcement", "url": "https://blog.google/innovation-and-ai/technology/developers-tools/diffusion-gemma-faster-text-generation/", "type": "primary" }, { "label": "Google model card", "url": "https://huggingface.co/google/diffusiongemma-26B-A4B-it/blob/f7f5b7f5fa82ffc52addd066915886d497f5517b/README.md", "type": "weights" }, { "label": "Official checkpoint configuration", "url": "https://huggingface.co/google/diffusiongemma-26B-A4B-it/blob/f7f5b7f5fa82ffc52addd066915886d497f5517b/config.json", "type": "weights" }, { "label": "Official parameter-count metadata", "url": "https://huggingface.co/google/diffusiongemma-26B-A4B-it/blob/f7f5b7f5fa82ffc52addd066915886d497f5517b/model.safetensors.index.json", "type": "weights" }, { "label": "Official checkpoint shards", "url": "https://huggingface.co/google/diffusiongemma-26B-A4B-it/tree/f7f5b7f5fa82ffc52addd066915886d497f5517b", "type": "weights" }, { "label": "DiffusionGemma implementation", "url": "https://github.com/huggingface/transformers/blob/6c5f8f4186c15f4bfbd67941a1736b2b347e0155/src/transformers/models/diffusion_gemma/modeling_diffusion_gemma.py", "type": "primary" }, { "label": "Gemma 4 vision implementation", "url": "https://github.com/huggingface/transformers/blob/6c5f8f4186c15f4bfbd67941a1736b2b347e0155/src/transformers/models/gemma4/modeling_gemma4.py", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/diffusiongemma-26b-a4b-it.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/google/diffusiongemma-26B-A4B-it", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Total includes the vision tower and projector. Google reports 3.8B active for text generation; image and video encoding runs separately.", "parameter_components_b": { "language_model": 25.250986752, "vision_encoder": 0.56954808, "vision_projector": 0.003244032 }, "active_parameters_scope": "language_model", "field_status": {}, "audited_at": "2026-10-26T19:29:23.686866+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/189.json" }, { "id": "anthropic/claude-fable-5", "slug": "anthropic--claude-fable-5", "name": "Claude Fable 5", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude-fable", "description": "Claude model for coding, complex knowledge work, and long-running agent workflows, with additional safeguards for high-risk domains.", "release_date": "2026-06-09", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1000000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.anthropic.com/news/claude-fable-5-mythos-5", "source_type": "primary", "sources": [ { "label": "Claude Fable 5 and Claude Mythos 5", "url": "https://www.anthropic.com/news/claude-fable-5-mythos-5", "type": "primary" }, { "label": "Claude Fable 5 - Claude Platform Docs", "url": "https://platform.claude.com/docs/en/models/fable-5/overview", "type": "primary" }, { "label": "PDF support", "url": "https://platform.claude.com/docs/en/build-with-claude/pdf-support", "type": "primary" }, { "label": "System Card: Claude Fable 5 & Claude Mythos 5", "url": "https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf", "type": "primary" }, { "label": "System Card: Claude Fable 5.1 & Claude Mythos 5.1", "url": "https://www-cdn.anthropic.com/0339e6a7c5c7b87f5c07798616dc32c215d14235/Claude%20Fable%205.1%20%26%20Claude%20Mythos%205.1%20System%20Card.pdf", "type": "primary" }, { "label": "System Card: Claude Opus 5", "url": "https://www-cdn.anthropic.com/b514064af1408018e64b1ad24e7d5e75850b4ffd/Claude%20Opus%205%20System%20Card.pdf", "type": "primary" }, { "label": "Prompting Claude Fable 5", "url": "https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5", "type": "primary" }, { "label": "Commercial Terms of Service", "url": "https://www.anthropic.com/legal/commercial-terms", "type": "primary" }, { "label": "GLM-5.3 and the spread of advanced cyber capabilities", "url": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities", "type": "research" }, { "label": "System Card: Claude Opus 5.5", "url": "https://www.anthropic.com/claude-opus-5-5-system-card", "type": "primary" }, { "label": "System Card: Claude Sonnet 5.5", "url": "https://www.anthropic.com/claude-sonnet-5-5-system-card", "type": "primary" }, { "label": "System Card: Claude Haiku 5.5", "url": "https://www.anthropic.com/claude-haiku-5-5-system-card", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/anthropic/claude-fable-5.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Nominal total and active parameter counts remain undisclosed in the accessed Anthropic specifications and original/later system cards. Fable 5 and Mythos 5 share underlying weights, but no published size is supplied. Exact architecture remains undisclosed; safeguards may route some requests to another model.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T10:23:59.371411Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/096.json" }, { "id": "anthropic/claude-mythos-5", "slug": "anthropic--claude-mythos-5", "name": "Claude Mythos 5", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude-mythos", "description": "Restricted Claude model for advanced cybersecurity and biology research workflows", "release_date": "2026-06-09", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1000000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://platform.claude.com/docs/en/models/mythos-5/overview", "source_type": "primary", "sources": [ { "label": "Claude Mythos 5 official specifications", "url": "https://platform.claude.com/docs/en/models/mythos-5/overview", "type": "primary" }, { "label": "Claude Fable 5 and Mythos 5 launch", "url": "https://www.anthropic.com/news/claude-fable-5-mythos-5", "type": "primary" }, { "label": "Claude Fable 5 and Mythos 5 system card", "url": "https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf", "type": "primary" }, { "label": "Anthropic commercial terms", "url": "https://www.anthropic.com/legal/commercial-terms", "type": "primary" }, { "label": "PDF support", "url": "https://platform.claude.com/docs/en/build-with-claude/pdf-support", "type": "primary" }, { "label": "Tool search compatibility", "url": "https://platform.claude.com/docs/en/agents-and-tools/tool-use/tool-search-tool", "type": "primary" }, { "label": "Anthropic statement on Claude weight availability", "url": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities", "type": "research" }, { "label": "Claude Fable 5.1 and Mythos 5.1 system card comparisons", "url": "https://www.anthropic.com/claude-fable-5-1-mythos-5-1-system-card", "type": "primary" }, { "label": "Claude Sonnet 5.5 system card", "url": "https://www.anthropic.com/claude-sonnet-5-5-system-card", "type": "primary" }, { "label": "Anthropic model catalog reference", "url": "https://raw.githubusercontent.com/anthropics/skills/refs/heads/main/skills/claude-api/shared/models.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/anthropic/claude-mythos-5.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T10:31:18Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/102.json", "notes": "Mythos 5 and Fable 5 explicitly share underlying model weights. Anthropic does not disclose a nominal total or active parameter count in the reviewed model documentation and system cards." }, { "id": "cohere/north-mini-code-1-0", "slug": "cohere--north-mini-code-1-0", "name": "North Mini Code", "developer": "Cohere", "developer_id": "cohere", "family": "north", "description": "Cohere coding model for practical software engineering and agentic edits", "release_date": "2026-06-09", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 30.484303872, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 256000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://docs.cohere.com/docs/north-mini-code-1.0", "source_type": "primary", "sources": [ { "label": "North Mini Code model documentation", "url": "https://docs.cohere.com/docs/north-mini-code-1.0", "type": "primary" }, { "label": "Announcing Cohere’s North Mini Code — June 9, 2026", "url": "https://docs.cohere.com/changelog/north-mini-code-1-0", "type": "primary" }, { "label": "North-Mini-Code-1.0 model card", "url": "https://huggingface.co/CohereLabs/North-Mini-Code-1.0", "type": "primary" }, { "label": "Introducing North Mini Code: Cohere’s First Model For Developers", "url": "https://huggingface.co/blog/CohereLabs/introducing-north-mini-code", "type": "primary" }, { "label": "North-Mini-Code-1.0 pinned checkpoint configuration", "url": "https://huggingface.co/CohereLabs/North-Mini-Code-1.0/raw/d11e61a842617a22dc328552fa5bb86231ee4f37/config.json", "type": "primary" }, { "label": "North-Mini-Code-1.0 BF16 tensor index and corresponding shard headers", "url": "https://huggingface.co/CohereLabs/North-Mini-Code-1.0/raw/d11e61a842617a22dc328552fa5bb86231ee4f37/model.safetensors.index.json", "type": "primary" }, { "label": "North Mini Code current standalone chat template", "url": "https://huggingface.co/CohereLabs/North-Mini-Code-1.0/raw/d11e61a842617a22dc328552fa5bb86231ee4f37/chat_template.jinja", "type": "primary" }, { "label": "Transformers Cohere2MoE implementation", "url": "https://raw.githubusercontent.com/huggingface/transformers/main/src/transformers/models/cohere2_moe/modeling_cohere2_moe.py", "type": "primary" }, { "label": "Transformers Cohere2MoE configuration implementation", "url": "https://raw.githubusercontent.com/huggingface/transformers/main/src/transformers/models/cohere2_moe/configuration_cohere2_moe.py", "type": "primary" }, { "label": "Introducing North Mini Code: Cohere’s first model for developers", "url": "https://cohere.com/blog/north-mini-code", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/cohere/north-mini-code-1-0.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete BF16 checkpoint contains 30,484,303,872 learned parameters (30.484303872B), marketed as 30B. Cohere separately reports a nominal 3B active; this is not calculated from expert fractions. Architecture is a decoder-only sparse MoE Transformer with one initial dense layer and 128 experts, selecting 8 per token. Advertised context is 256,000 tokens with 64,000 maximum output; the 500,000 position configuration and 320,000 vLLM example are not substituted for the documented context. June 9 is the explicit official release date; June 5 repository creation alone does not establish public release.", "field_status": {}, "audited_at": "2026-10-26T18:26:57.957828Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/155.json" }, { "id": "google/gemini-3.5-live-translate-preview", "slug": "google--gemini-3.5-live-translate-preview", "name": "Gemini 3.5 Live Translate Preview", "developer": "Google", "developer_id": "google", "family": "gemini-pro", "description": "Speech translation preview for more than 70 languages, with translated audio and optional transcripts through the Gemini Live API.", "release_date": "2026-06-09", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "audio" ], "output": [ "audio", "text" ] }, "reasoning": false, "tool_call": false, "source_url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.5-live-translate-preview", "source_type": "primary", "sources": [ { "label": "Google exact model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.5-live-translate-preview", "type": "primary" }, { "label": "Google Live Translation guide", "url": "https://ai.google.dev/gemini-api/docs/live-api/live-translate", "type": "primary" }, { "label": "Google June 9 release announcement", "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-live-3-5-translate/", "type": "primary" }, { "label": "Google DeepMind Gemini 3.5 Audio model card", "url": "https://deepmind.google/models/model-cards/gemini-3-5-audio/", "type": "primary" }, { "label": "Explicitly referenced Gemini 3 Pro architecture", "url": "https://deepmind.google/models/model-cards/gemini-3-pro/", "type": "primary" }, { "label": "Gemini API service terms", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemini-3.5-live-translate-preview.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Sparse MoE Transformer based on Gemini 3 Pro; neither Live Translate nor its named base has a published total or active parameter count in the reviewed sources. Text output denotes transcripts accompanying speech translation. The public endpoint accepts audio only despite the broader model card's audio-and-text input statement. Earliest public preview: June 9, 2026; Cloud's August date is later platform availability.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T13:03:40.184527Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/209.json" }, { "id": "xiaomi/mimo-v2.5-pro-ultraspeed", "slug": "xiaomi--mimo-v2.5-pro-ultraspeed", "name": "MiMo-V2.5-Pro-UltraSpeed", "developer": "Xiaomi", "developer_id": "xiaomi", "family": "mimo", "description": "Text reasoning model using an FP4-quantized MiMo-V2.5-Pro MoE backbone and a DFlash speculative-decoding drafter.", "release_date": "2026-06-08", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 1020, "active_parameters_b": 42, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://mimo.xiaomi.com/blog/mimo-tilert-1000tps", "source_type": "primary", "sources": [ { "label": "MiMo-V2.5-Pro-UltraSpeed launch", "url": "https://mimo.xiaomi.com/blog/mimo-tilert-1000tps", "type": "primary" }, { "label": "MiMo-V2.5-Pro-FP4-DFlash checkpoint card", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro-FP4-DFlash/blob/b754e6c86008bdb5cc901308dda5a38173ec7276/README.md", "type": "weights" }, { "label": "UltraSpeed underlying backbone configuration", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro-FP4-DFlash/blob/b754e6c86008bdb5cc901308dda5a38173ec7276/config.json", "type": "weights" }, { "label": "DFlash drafter configuration", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro-FP4-DFlash/blob/b754e6c86008bdb5cc901308dda5a38173ec7276/dflash/config.json", "type": "weights" }, { "label": "UltraSpeed checkpoint chat template", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro-FP4-DFlash/blob/b754e6c86008bdb5cc901308dda5a38173ec7276/tokenizer_config.json", "type": "weights" }, { "label": "MiMo-V2.5-Pro-UltraSpeed product page", "url": "https://mimo.mi.com/models/en-US/mimo-v2.5-pro-ultraspeed", "type": "primary" }, { "label": "UltraSpeed trial extension and ending notice", "url": "https://mimo.mi.com/docs/en-US/news/latest/beta-extended", "type": "primary" }, { "label": "XiaomiMiMo UltraSpeed announcement via X syndication", "url": "https://cdn.syndication.twimg.com/tweet-result?id=2063993790587904362&lang=en&token=x", "type": "primary" }, { "label": "Official checkpoint repository metadata", "url": "https://huggingface.co/api/models/XiaomiMiMo/MiMo-V2.5-Pro-FP4-DFlash", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/xiaomi/mimo-v2.5-pro-ultraspeed.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro-FP4-DFlash", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Nominal 1.02T total and 42B active refer to the target backbone; the separate five-layer DFlash drafter and MTP weights are not reconciled into an exact combined parameter count. UltraSpeed uses separately published expert FP4-QAT weights plus the drafter, so it is retained as a distinct derivative despite experience-mode branding. Original announcement and linked open checkpoint: June 8, 2026; approved hosted trial window began June 9 Beijing time and was later extended, then ended. The backbone supports 1,048,576 positions; the drafter configuration separately lists 262144. Current availability of the former hosted endpoint is not asserted.", "parameters_scope": "language_backbone", "parameter_components_b": { "language_backbone": 1020 }, "active_parameters_scope": "backbone_forward", "active_parameters_by_phase_b": { "backbone_forward": 42 }, "field_status": {}, "audited_at": "2026-10-27T05:54:32.574667Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/542.json" }, { "id": "nvidia/nemotron-3-ultra-550b-a55b", "slug": "nvidia--nemotron-3-ultra-550b-a55b", "name": "Nemotron 3 Ultra 550B A55B", "developer": "NVIDIA", "developer_id": "nvidia", "family": "nemotron", "description": "Text model with hybrid Mamba-2, attention and latent MoE layers, configurable reasoning, tool calling and multi-token prediction.", "release_date": "2026-06-04", "date_precision": "day", "open_weights": true, "license": "OpenMDW-1.1", "parameters_b": 560.524578816, "active_parameters_b": 55, "parameters_source": "primary_source", "architecture": "hybrid-ssm-transformer-moe", "context_window": 1048576, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16/blob/77df655d5e9f8362164ed14dd8b48f8bce657498/README.md", "source_type": "primary", "sources": [ { "label": "NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 model card", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16/blob/77df655d5e9f8362164ed14dd8b48f8bce657498/README.md", "type": "primary" }, { "label": "Pinned Ultra BF16 configuration", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16/blob/77df655d5e9f8362164ed14dd8b48f8bce657498/config.json", "type": "weights" }, { "label": "Pinned Ultra BF16 index and 225 shard headers", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16/blob/77df655d5e9f8362164ed14dd8b48f8bce657498/model.safetensors.index.json", "type": "weights" }, { "label": "Nemotron 3 Ultra technical report, sections 2.1, 2.5 and 3", "url": "https://arxiv.org/html/2606.15007v1", "type": "primary" }, { "label": "Transformers Nemotron-H parameter and buffer definitions", "url": "https://github.com/huggingface/transformers/blob/8979c57d98a0c43ea8278d8a8e93a48e9a3b6ccb/src/transformers/models/nemotron_h/modeling_nemotron_h.py", "type": "primary" }, { "label": "vLLM Nemotron-H MTP implementation", "url": "https://github.com/vllm-project/vllm/blob/ed1a7307950ab5e6593d25123579ed7ca998b350/vllm/model_executor/models/nemotron_h_mtp.py", "type": "primary" }, { "label": "NVIDIA Nemotron 3 Ultra release", "url": "https://research.nvidia.com/labs/nemotron/Nemotron-3-Ultra/", "type": "primary" }, { "label": "NVIDIA Nemotron LLM information, updated August 2026", "url": "https://www.nvidia.com/en-us/ai-data-science/foundation-models/nemotron/llm-info/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/nvidia/nemotron-3-ultra-550b-a55b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete checkpoint: 560,524,578,816 learned parameters, comprising 549,308,968,960 in the main model and 11,215,609,856 in the shared MTP module. The 550B name is nominal; NVIDIA reports about 55B active for the main model. MTP drafting reuses the main embedding/output weights and adds separate optional execution. Maximum context is 1,048,576 tokens with the documented override; the default configuration is 262,144.", "parameter_components_b": { "main_model": 549.30896896, "mtp": 11.215609856 }, "active_parameters_scope": "main_model", "field_status": {}, "audited_at": "2026-10-27T00:45:52.782418Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/384.json" }, { "id": "google/gemma-4-12b-it", "slug": "google--gemma-4-12b-it", "name": "Gemma 4 12B IT", "developer": "Google", "developer_id": "google", "family": "gemma", "description": "Google’s instruction-tuned Gemma 4 model with a dense, encoder-free Transformer, text, image, audio and video inputs, text output, configurable thinking and function calling.", "release_date": "2026-06-03", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 11.9597, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 262144, "modalities": { "input": [ "text", "image", "audio", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/google/gemma-4-12B-it", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/google/gemma-4-12B-it", "type": "weights" }, { "label": "Google release history", "url": "https://ai.google.dev/gemma/docs/releases", "type": "primary" }, { "label": "Official checkpoint configuration", "url": "https://huggingface.co/google/gemma-4-12B-it/resolve/707f0a3b8a3c7ad586ed01e27eafbad8a27dd0f7/config.json", "type": "weights" }, { "label": "Official checkpoint tensor header", "url": "https://huggingface.co/google/gemma-4-12B-it/resolve/707f0a3b8a3c7ad586ed01e27eafbad8a27dd0f7/model.safetensors", "type": "weights" }, { "label": "Gemma 4 Technical Report", "url": "https://arxiv.org/html/2607.02770v1", "type": "primary" }, { "label": "Gemma 4 Unified implementation", "url": "https://github.com/huggingface/transformers/blob/main/src/transformers/models/gemma4_unified/modeling_gemma4_unified.py", "type": "primary" }, { "label": "Checkpoint converter", "url": "https://github.com/huggingface/transformers/blob/main/src/transformers/models/gemma4_unified/convert_gemma4_unified_weights.py", "type": "primary" }, { "label": "Google JAX Transformer block", "url": "https://github.com/google-deepmind/gemma/blob/9e1f74131ba9d45e5fe6ae7bb3941d9f2abb39c3/gemma/gm/nn/gemma4/_modules.py", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemma-4-12b-it.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Approximately 11.9597B parameters in the main checkpoint, including multimodal projections and tied embedding/output weights once; Google reports a nominal 11.95B. The separate MTP assistant is excluded. Runtime conventions differ on 48 layer scalars, without changing the rounded total. No separate active-parameter count was found in the checked primary sources. Video is processed as image frames.", "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T13:24:27.370635+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/222.json" }, { "id": "microsoft/mai-code-1-flash", "slug": "microsoft--mai-code-1-flash", "name": "MAI-Code-1-Flash", "developer": "Microsoft", "developer_id": "microsoft", "family": "mai", "description": "Microsoft's text-only coding model with adaptive reasoning, trained to use GitHub Copilot's tools.", "release_date": "2026-06-02", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 137, "active_parameters_b": 5, "parameters_source": "primary_source", "architecture": "moe", "context_window": 256000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "source_type": "primary", "sources": [ { "label": "Model card", "url": "https://microsoft.ai/pdf/MAI-Code-1-Flash-Model-Card.PDF", "type": "model_card" }, { "label": "Announcement", "url": "https://microsoft.ai/news/introducingmai-code-1-flash/", "type": "announcement" }, { "label": "GitHub initial public rollout", "url": "https://github.blog/changelog/2026-06-02-mai-code-1-flash-is-now-available-for-github-copilot/", "type": "primary" }, { "label": "GitHub retirement notice", "url": "https://github.blog/changelog/2026-09-10-mai-code-1-flash-deprecated/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/microsoft/mai-code-1-flash.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Microsoft reports 137B total and 5B active parameters. Context is nominally 256K; exact token capacity is not specified. GitHub retired this version on September 10, 2026. The original product-page URL now describes MAI-Code-1.1-Flash, whose capabilities must be kept separate. No public standalone API identifier was verified.", "field_status": {}, "audited_at": "2026-10-26T22:03:31Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/293.json" }, { "id": "alibaba/qwen3.7-plus", "slug": "alibaba--qwen3.7-plus", "name": "Qwen3.7 Plus", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Qwen3.7 multimodal model with a 1M-token context, text, image and video inputs, thinking mode, and function calling for coding and agent workflows.", "release_date": "2026-06-01", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": 397, "active_parameters_b": 17, "parameters_source": "primary_source", "architecture": null, "context_window": 1000000, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.7-plus.toml", "source_type": "catalog", "sources": [ { "label": "Alibaba Cloud exact model documentation", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-7-plus", "type": "primary" }, { "label": "Alibaba Cloud regional release history", "url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "type": "primary" }, { "label": "Qwen3.7-Plus announcement mirror", "url": "https://www.alibabacloud.com/blog/qwen3-7-plus-multimodal-agent-intelligence_603206", "type": "primary" }, { "label": "QwenCloud thinking support", "url": "https://docs.qwencloud.com/developer-guides/text-generation/thinking", "type": "primary" }, { "label": "QwenCloud model release changelog", "url": "https://docs.qwencloud.com/changelog/models", "type": "primary" }, { "label": "Qwen official parameter comparison table", "url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next/raw/de4b8e4d43b917e7706784d8bb445c9af86a3540/README.md", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=qwen3.7-plus&limit=100", "label": "Official Qwen inventory search for the exact hosted model", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=Qwen3.7&limit=1000", "label": "Official Qwen Qwen3.7 weight inventory", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.7-plus.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The alias maps to qwen3.7-plus-2026-05-26; June 1 is the documented public release. The later official Flash-Next Language comparison table explicitly reports 397B parameters and 17B active for Qwen3.7-Plus. These are nominal language-model counts; complete hosted components are not quantified. Internal architecture and weight license remain unverified. No public Plus weight release is available. Classified as closed weights: no public release of this hosted variant was found in the developer documentation and weight inventory.", "field_status": { "license": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-28T08:01:10.206539+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/065.json", "active_parameters_scope": "language_model", "parameters_scope": "language_model", "parameter_components_b": { "language_model": 397 } }, { "id": "minimax/MiniMax-M3", "slug": "minimax--minimax-m3", "name": "MiniMax-M3", "developer": "MiniMax", "developer_id": "minimax", "family": "minimax", "description": "Native multimodal MoE model using MiniMax Sparse Attention for long-context coding, reasoning, and tool use.", "release_date": "2026-06-01", "date_precision": "day", "open_weights": true, "license": "MiniMax Community License", "parameters_b": 428, "active_parameters_b": 23, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.minimax.io/blog/minimax-m3", "source_type": "primary", "sources": [ { "label": "MiniMax M3 release announcement", "url": "https://www.minimax.io/blog/minimax-m3", "type": "primary" }, { "label": "MiniMax M3 official model card", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/d95e30cb912fe3021bb9c2ba66a8ce60b6c937a9/README.md", "type": "weights" }, { "label": "MiniMax M3 checkpoint configuration", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/d95e30cb912fe3021bb9c2ba66a8ce60b6c937a9/config.json", "type": "weights" }, { "label": "MiniMax M3 license", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/d95e30cb912fe3021bb9c2ba66a8ce60b6c937a9/LICENSE", "type": "weights" }, { "label": "MiniMax M3 checkpoint tensor index", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/d95e30cb912fe3021bb9c2ba66a8ce60b6c937a9/model.safetensors.index.json", "type": "weights" }, { "label": "MiniMax API model invocation", "url": "https://platform.minimax.io/docs/guides/text-generation", "type": "primary" }, { "label": "MiniMax OpenAI-compatible API reference", "url": "https://platform.minimax.io/docs/api-reference/text-openai-api", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/minimax/MiniMax-M3.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M3", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "428B total and 23B active are MiniMax nominal figures; active-count component accounting is not specified. All 59 BF16 shard headers total 427.040140160B stored elements, including vision/projectors and 7,296 routing-bias buffers; no MTP tensors are released in this checkpoint despite MTP config fields. Checkpoint capacity is 1,048,576 positions; the current API documents 1,000,000 tokens. Public release was June 1, 2026; the initial weight commit was June 12. Weights use the custom MiniMax Community License.", "field_status": {}, "audited_at": "2026-10-26T22:39:02.061459Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/307.json" }, { "id": "bytedance-seed/seed-evolving", "slug": "bytedance-seed--seed-evolving", "name": "Seed Evolving", "developer": "ByteDance", "developer_id": "bytedance-seed", "family": "seed", "description": "Continuously updated Seed model for reasoning, coding, task planning, tool calling, and image and video understanding.", "release_date": "2026-06", "date_precision": "month", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1024000, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ark.volcengine.com/region:cn-beijing/docs/latest-model.md", "source_type": "primary", "sources": [ { "label": "Seed-Evolving official documentation", "url": "https://ark.volcengine.com/region:cn-beijing/docs/latest-model.md", "type": "primary" }, { "label": "Official model release announcements", "url": "https://ark.volcengine.com/region:cn-beijing/docs/model-release-announcement.md", "type": "primary" }, { "label": "Doubao model service agreement", "url": "https://ark.volcengine.com/region:cn-beijing/docs/doubao-model-service-agreement.md", "type": "primary" }, { "label": "Seed2.1 model card", "url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "type": "primary" }, { "label": "Seed2.0 model card", "url": "https://arxiv.org/pdf/2607.00248", "type": "primary" }, { "label": "Official model list", "url": "https://ark.volcengine.com/region:cn-beijing/docs/model-list.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/bytedance-seed/seed-evolving.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Rolling doubao-seed-evolving endpoint: specifications reflect the official card accessed October 27, 2026, whose stated model revision is September 9, 2026. Initial release is documented only to June 2026; the July entry is a context upgrade. Total and active parameter counts remain undisclosed; no publicly sized base supports an inference. Architecture remains undisclosed.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T11:08:45.529942Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/139.json" }, { "id": "nex-agi/nex-n2-pro", "slug": "nex-agi--nex-n2-pro", "name": "Nex-N2-Pro", "developer": "Nex AGI", "developer_id": "nex-agi", "family": "nex-n2", "description": "Open agentic MoE model (397B total, 17B active) for coding, tool use, and research workflows", "release_date": "2026-06", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 397, "active_parameters_b": 17, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/nex-agi/nex-n2-pro.toml", "source_type": "catalog", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/nex-agi/Nex-N2-Pro", "type": "weights" }, { "label": "Exact Nex-N2-Pro configuration", "url": "https://huggingface.co/nex-agi/Nex-N2-Pro/blob/e7e4a4a29eb46a41a730d0b283ec5af0f16185d9/config.json", "type": "primary" }, { "label": "Identified Qwen foundation specification", "url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "type": "weights" }, { "label": "Nex-N2-Pro checkpoint metadata", "url": "https://huggingface.co/api/models/nex-agi/Nex-N2-Pro", "type": "weights" }, { "label": "Nex-N2-Pro tensor index", "url": "https://huggingface.co/nex-agi/Nex-N2-Pro/blob/e7e4a4a29eb46a41a730d0b283ec5af0f16185d9/model.safetensors.index.json", "type": "weights" }, { "label": "Nex-N2-Pro image and video processor", "url": "https://huggingface.co/nex-agi/Nex-N2-Pro/blob/e7e4a4a29eb46a41a730d0b283ec5af0f16185d9/processor_config.json", "type": "primary" }, { "label": "Nex-N2-Pro chat template", "url": "https://huggingface.co/nex-agi/Nex-N2-Pro/blob/e7e4a4a29eb46a41a730d0b283ec5af0f16185d9/chat_template.jinja", "type": "primary" }, { "label": "Founding-institute release announcement", "url": "https://www.sii.edu.cn/_s3/2026/0607/c27a1089/page.psp", "type": "primary" }, { "label": "Nex-N2-Pro repository history", "url": "https://huggingface.co/api/models/nex-agi/Nex-N2-Pro/commits/main", "type": "weights" }, { "label": "Initial Pro release card", "url": "https://huggingface.co/nex-agi/Nex-N2-Pro/blob/a85b319a367e85c544a79e882638964be8238941/README.md", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/nex-agi/nex-n2-pro.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Post-trained from Qwen3.5-397B-A17B. Counts are nominal 397B total / 17B active for the language backbone, corroborated by the matching Nex architecture and approximately 396.802B BF16 checkpoint elements including vision; no exact learned-parameter tensor count is claimed. Vision processing is input-dependent. Exact processor/template support video as well as text/images; hosted routes may differ. June 2026 release is verified, but the earliest public day is unresolved: catalog June 2, repository creation/initial release text June 3, founding-institute announcement June 4. Repository timestamps are not treated as public-release proof.", "field_status": {}, "audited_at": "2026-10-27T00:13:01.618430Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/369.json" }, { "id": "stepfun/step-3.7-flash", "slug": "stepfun--step-3.7-flash", "name": "Step 3.7 Flash", "developer": "StepFun", "developer_id": "stepfun", "family": "step", "description": "Multimodal reasoning model with sparse MoE routing, image and video input, tool calling, and three multi-token prediction branches.", "release_date": "2026-05-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 201.365304064, "active_parameters_b": 11, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/stepfun-ai/Step-3.7-Flash/blob/5f6244077ac62e04eec3f320501ff8c2b293373a/README.md", "source_type": "primary", "sources": [ { "label": "Exact Step 3.7 Flash model card", "url": "https://huggingface.co/stepfun-ai/Step-3.7-Flash/blob/5f6244077ac62e04eec3f320501ff8c2b293373a/README.md", "type": "primary" }, { "label": "Dated Step 3.7 Flash launch and availability", "url": "https://static.stepfun.com/blog/step-3.7-flash/", "type": "primary" }, { "label": "Step 3.7 Flash API guide", "url": "https://platform.stepfun.ai/docs/en/guides/models/step-3.7-flash", "type": "primary" }, { "label": "Exact released BF16 checkpoint configuration", "url": "https://huggingface.co/stepfun-ai/Step-3.7-Flash/blob/5f6244077ac62e04eec3f320501ff8c2b293373a/config.json", "type": "primary" }, { "label": "Exact model configuration defaults", "url": "https://huggingface.co/stepfun-ai/Step-3.7-Flash/blob/5f6244077ac62e04eec3f320501ff8c2b293373a/configuration_step3p7.py", "type": "primary" }, { "label": "Exact Step 3.7 backbone and projection implementation", "url": "https://huggingface.co/stepfun-ai/Step-3.7-Flash/blob/5f6244077ac62e04eec3f320501ff8c2b293373a/modeling_step3p7.py", "type": "primary" }, { "label": "Exact visual encoder implementation", "url": "https://huggingface.co/stepfun-ai/Step-3.7-Flash/blob/5f6244077ac62e04eec3f320501ff8c2b293373a/vision_encoder.py", "type": "primary" }, { "label": "Complete checkpoint index and bounded headers from all 26 shards", "url": "https://huggingface.co/stepfun-ai/Step-3.7-Flash/blob/5f6244077ac62e04eec3f320501ff8c2b293373a/model.safetensors.index.json", "type": "primary" }, { "label": "Pinned vLLM exact Step 3.7 wrapper and loading map", "url": "https://github.com/vllm-project/vllm/blob/6fe3d447ac6429ef13aff117744d0298e28879ed/vllm/model_executor/models/step3p7.py", "type": "primary" }, { "label": "Pinned configured decoder and separate output head", "url": "https://github.com/vllm-project/vllm/blob/6fe3d447ac6429ef13aff117744d0298e28879ed/vllm/model_executor/models/step3p5.py", "type": "primary" }, { "label": "Pinned MTP head ownership and exact checkpoint loading rules", "url": "https://github.com/vllm-project/vllm/blob/6fe3d447ac6429ef13aff117744d0298e28879ed/vllm/model_executor/models/step3p5_mtp.py", "type": "primary" }, { "label": "Official release Apache 2.0 license", "url": "https://github.com/stepfun-ai/Step-3.7-Flash/blob/main/LICENSE", "type": "primary" }, { "label": "Exact checkpoint commit history", "url": "https://huggingface.co/api/models/stepfun-ai/Step-3.7-Flash/commits/main", "type": "primary" }, { "label": "Exact hosted Step 3.7 endpoint metadata", "url": "https://openrouter.ai/api/v1/models/stepfun/step-3.7-flash/endpoints", "type": "primary" }, { "label": "NVIDIA exact-model integration documentation", "url": "https://docs.nvidia.com/nemo/automodel/recipes-e2e-examples/step-3-7", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/stepfun/step-3.7-flash.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/stepfun-ai/Step-3.7-Flash", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Parameters count the complete BF16 checkpoint, including three MTP branches and vision: 201.365304064B learned weights, excluding 12096 frozen router-bias values. StepFun advertises nominal 198B total and approximately 11B active; the active-count convention is unspecified, and NVIDIA reports 13B. The checkpoint allows 262144 tokens, while the StepFun hosted route advertises 256000. May 29 is the explicit developer launch; earlier upload/listing dates do not prove earlier public serving.", "parameter_components_b": { "language_backbone": 196.956118272, "mtp": 2.428171008, "vision_encoder_and_downsamplers": 1.95584896, "vision_projector": 0.025165824 }, "field_status": {}, "audited_at": "2026-10-27T04:39:49.484295Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/496.json" }, { "id": "anthropic/claude-opus-4-8", "slug": "anthropic--claude-opus-4-8", "name": "Claude Opus 4.8", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude-opus", "description": "Claude Opus model with adaptive thinking for coding, tool use, and long-running agentic work.", "release_date": "2026-05-28", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1000000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.anthropic.com/news/claude-opus-4-8", "source_type": "primary", "sources": [ { "label": "Anthropic Opus 4.8 announcement", "url": "https://www.anthropic.com/news/claude-opus-4-8", "type": "primary" }, { "label": "Anthropic exact-model specifications", "url": "https://platform.claude.com/docs/en/models/opus-4-8/overview", "type": "primary" }, { "label": "Anthropic Opus 4.8 system card", "url": "https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea.pdf", "type": "primary" }, { "label": "Anthropic commercial terms", "url": "https://www.anthropic.com/legal/commercial-terms", "type": "primary" }, { "label": "Anthropic Opus 4.8 prompting guidance", "url": "https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-opus-4-8", "type": "primary" }, { "label": "Anthropic PDF support", "url": "https://platform.claude.com/docs/en/build-with-claude/pdf-support", "type": "primary" }, { "label": "GLM-5.3 and the spread of advanced cyber capabilities", "url": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities", "type": "primary" }, { "label": "Claude Opus 5 System Card", "url": "https://www.anthropic.com/claude-opus-5-system-card", "type": "primary" }, { "label": "Claude Opus 5.5 System Card", "url": "https://www.anthropic.com/claude-opus-5-5-system-card", "type": "primary" }, { "label": "Claude Fable 5 and Claude Mythos 5 System Card", "url": "https://www.anthropic.com/claude-fable-5-mythos-5-system-card", "type": "primary" }, { "label": "Claude Fable 5.1 and Claude Mythos 5.1 System Card", "url": "https://www.anthropic.com/claude-fable-5-1-mythos-5-1-system-card", "type": "primary" }, { "label": "Automated Researchers Can Mitigate Well-Characterized Alignment Failures", "url": "https://alignment.anthropic.com/2026/automated-alignment-researchers/", "type": "primary" }, { "label": "Models overview - Claude Platform Docs", "url": "https://platform.claude.com/docs/en/models/overview", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/anthropic/claude-opus-4-8.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Anthropic has not published total or active parameter counts in the reviewed specifications and system cards. No size-bearing base lineage is established; third-party size estimates are not specifications. Architecture remains undisclosed. PDF inputs are processed as text and page images.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T10:40:31.448535Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/109.json" }, { "id": "liquid/lfm2.5-8b-a1b", "slug": "liquid--lfm2.5-8b-a1b", "name": "LFM2.5-8B-A1B", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm", "description": null, "release_date": "2026-05-28", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 8.3, "active_parameters_b": 1.5, "parameters_source": "primary_source", "architecture": "moe", "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/LiquidAI/LFM2.5-8B-A1B/blob/5dd22602c2e9f6a097b1de4c4efe0658b605015c/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/LiquidAI/LFM2.5-8B-A1B/blob/5dd22602c2e9f6a097b1de4c4efe0658b605015c/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:23.459Z", "last_seen": "2026-10-29T10:17:23.459Z", "upstream_present": false, "availability": "public", "notes": "Post-trained checkpoint explicitly included in the developer release. Parameter and context values follow the exact model card; MoE variants also use the LFM convolution/attention hybrid backbone." }, { "id": "liquid/lfm2.5-8b-a1b-base", "slug": "liquid--lfm2.5-8b-a1b-base", "name": "LFM2.5-8B-A1B-Base", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm", "description": null, "release_date": "2026-05-28", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 8.3, "active_parameters_b": 1.5, "parameters_source": "primary_source", "architecture": "moe", "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/LiquidAI/LFM2.5-8B-A1B-Base/blob/6cfb83b04b0f0f12a64cb38ab553275d3ffe95d8/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/LiquidAI/LFM2.5-8B-A1B-Base/blob/6cfb83b04b0f0f12a64cb38ab553275d3ffe95d8/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:23.459Z", "last_seen": "2026-10-29T10:17:23.459Z", "upstream_present": false, "availability": "public", "notes": "Pretrained checkpoint explicitly included in the developer release. Parameter and context values follow the exact model card; MoE variants also use the LFM convolution/attention hybrid backbone." }, { "id": "tencent/hy-mt2-1.8b", "slug": "tencent--hy-mt2-1.8b", "name": "Hy-MT2-1.8B", "developer": "Tencent", "developer_id": "tencent", "family": "hy-mt2", "description": "Tencent Hy-MT2 dense multilingual translation model supporting 33 languages and translation instructions for terminology, style, formatting and context.", "release_date": "2026-05-21", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 1.8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/tencent/Hy-MT2-1.8B/blob/9a341cd1b679d3efd23b46e847b01745a71ed792/README.md", "source_type": "primary", "sources": [ { "label": "Hy-MT2-1.8B model card, pinned revision", "url": "https://huggingface.co/tencent/Hy-MT2-1.8B/blob/9a341cd1b679d3efd23b46e847b01745a71ed792/README.md", "type": "primary" }, { "label": "Hy-MT2 official deployment README", "url": "https://github.com/Tencent-Hunyuan/Hy-MT2/blob/ff1903ecaa724e10951a23c16817a2413c752b35/README.md", "type": "primary" }, { "label": "Hy-MT2 technical report, revised 25 May 2026", "url": "https://arxiv.org/html/2605.22064v2", "type": "primary" }, { "label": "Hy-MT2-1.8B checkpoint configuration", "url": "https://huggingface.co/tencent/Hy-MT2-1.8B/blob/9a341cd1b679d3efd23b46e847b01745a71ed792/config.json", "type": "primary" }, { "label": "Hy-MT2-1.8B model license", "url": "https://huggingface.co/tencent/Hy-MT2-1.8B/blob/9a341cd1b679d3efd23b46e847b01745a71ed792/LICENSE.txt", "type": "primary" }, { "label": "Official model repository file inventory", "url": "https://huggingface.co/api/models/tencent/Hy-MT2-1.8B", "type": "primary" }, { "label": "Complete bounded safetensors tensor header", "url": "https://huggingface.co/tencent/Hy-MT2-1.8B/resolve/9a341cd1b679d3efd23b46e847b01745a71ed792/model.safetensors", "type": "primary" }, { "label": "Configured HunYuanDenseV1 model implementation, Transformers v5.6.0", "url": "https://github.com/huggingface/transformers/blob/v5.6.0/src/transformers/models/hunyuan_v1_dense/modeling_hunyuan_v1_dense.py", "type": "primary" }, { "label": "Transformers v5.6.0 weight-tying loader behavior", "url": "https://github.com/huggingface/transformers/blob/v5.6.0/src/transformers/modeling_utils.py", "type": "primary" }, { "label": "Exact checkpoint chat template", "url": "https://huggingface.co/tencent/Hy-MT2-1.8B/blob/9a341cd1b679d3efd23b46e847b01745a71ed792/chat_template.jinja", "type": "primary" }, { "label": "Official training guide for the exact dense variant", "url": "https://huggingface.co/tencent/Hy-MT2-1.8B/blob/9a341cd1b679d3efd23b46e847b01745a71ed792/train/README.md", "type": "primary" }, { "label": "Hy-MT2 technical report version history, rechecked October 27", "url": "https://arxiv.org/abs/2605.22064", "type": "primary" }, { "label": "SubVision WMT26 original QLoRA study, October 2026", "url": "https://aclanthology.org/2026.wmt-main.203.pdf", "type": "primary" }, { "label": "paramanandaAI WMT26 original translation adaptation study, October 2026", "url": "https://aclanthology.org/2026.wmt-main.205.pdf", "type": "primary" }, { "label": "WMT26 staged multilingual adaptation of Hy-MT2-1.8B, October 2026", "url": "https://aclanthology.org/2026.wmt-main.184.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/tencent/hy-mt2-1.8b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Tencent reports nominal 1.8B parameters. The 2.038515712B serialized-element count includes separately stored embedding/output weights; their equality and exact unique total remain unresolved. Active parameters are not separately established. Official deployment context is 8192 tokens despite the 262144 positional configuration. Reasoning-mode and tool-calling support remain unverified.", "field_status": { "active_parameters_b": "undisclosed", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T18:01:29.005353Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/500.json" }, { "id": "tencent/hy-mt2-30b-a3b", "slug": "tencent--hy-mt2-30b-a3b", "name": "Hy-MT2-30B-A3B", "developer": "Tencent", "developer_id": "tencent", "family": "hy-mt2", "description": "Tencent Hy-MT2 multilingual text translation model with a mixture-of-experts architecture and support for terminology, style, context and structured-translation instructions.", "release_date": "2026-05-21", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 30.064719872, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://huggingface.co/tencent/Hy-MT2-30B-A3B/blob/d3ead4dba61c09aac60a261a96ad1df3e705febb/README.md", "source_type": "primary", "sources": [ { "label": "Tencent Hy-MT2-30B-A3B model card", "url": "https://huggingface.co/tencent/Hy-MT2-30B-A3B/blob/d3ead4dba61c09aac60a261a96ad1df3e705febb/README.md", "type": "primary" }, { "label": "Hy-MT2 official repository README", "url": "https://github.com/Tencent-Hunyuan/Hy-MT2/blob/ff1903ecaa724e10951a23c16817a2413c752b35/README.md", "type": "primary" }, { "label": "Hy-MT2-30B-A3B license", "url": "https://huggingface.co/tencent/Hy-MT2-30B-A3B/blob/d3ead4dba61c09aac60a261a96ad1df3e705febb/LICENSE.txt", "type": "primary" }, { "label": "Hy-MT2-30B-A3B checkpoint configuration", "url": "https://huggingface.co/tencent/Hy-MT2-30B-A3B/blob/d3ead4dba61c09aac60a261a96ad1df3e705febb/config.json", "type": "primary" }, { "label": "Complete indexed checkpoint and all 51 safetensors headers", "url": "https://huggingface.co/tencent/Hy-MT2-30B-A3B/blob/d3ead4dba61c09aac60a261a96ad1df3e705febb/model.safetensors.index.json", "type": "primary" }, { "label": "Tencent checkpoint conversion implementation", "url": "https://huggingface.co/tencent/Hy-MT2-30B-A3B/blob/d3ead4dba61c09aac60a261a96ad1df3e705febb/train/tools/convert_ckpt_to_outer.py", "type": "primary" }, { "label": "HYV3 model implementation", "url": "https://github.com/huggingface/transformers/blob/c0d024fbc873ab0325e1b40fc1bd69069609de65/src/transformers/models/hy_v3/modeling_hy_v3.py", "type": "primary" }, { "label": "HYV3 configuration defaults and layer construction", "url": "https://github.com/huggingface/transformers/blob/c0d024fbc873ab0325e1b40fc1bd69069609de65/src/transformers/models/hy_v3/configuration_hy_v3.py", "type": "primary" }, { "label": "Exact released checkpoint chat template", "url": "https://huggingface.co/tencent/Hy-MT2-30B-A3B/blob/d3ead4dba61c09aac60a261a96ad1df3e705febb/chat_template.jinja", "type": "primary" }, { "label": "Hy-MT2 technical report, revision 2 (May 25, 2026)", "url": "https://arxiv.org/html/2605.22064v2", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/tencent/hy-mt2-30b-a3b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Total parameters (30.064719872B) are a complete checkpoint count after excluding 6,016 non-parameter router-bias buffer elements; all 51 shard headers and the configured module graph were reconciled. Active 3B is Tencent's nominal A3B designation, with no exact activation-count convention disclosed. Documented deployment context is 8,192 tokens; config permits 262,144 positions. Exact template supports tool-call protocol and optional reasoning modes, but trained reasoning capability is unresolved.", "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-27T04:49:47.246456Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/501.json" }, { "id": "tencent/hy-mt2-7b", "slug": "tencent--hy-mt2-7b", "name": "Hy-MT2-7B", "developer": "Tencent", "developer_id": "tencent", "family": "hy-mt2", "description": "Tencent multilingual translation model using a dense decoder-only Transformer, with instructions for terminology, style, contextual translation and structured text.", "release_date": "2026-05-21", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/tencent/Hy-MT2-7B", "source_type": "primary", "sources": [ { "label": "Tencent Hy-MT2-7B model card", "url": "https://huggingface.co/tencent/Hy-MT2-7B/blob/9b0eb4e8f001def3e5ff6469a0ac96fdb39ec223/README.md", "type": "primary" }, { "label": "Tencent Hy-MT2 release and inference documentation", "url": "https://github.com/Tencent-Hunyuan/Hy-MT2/blob/ff1903ecaa724e10951a23c16817a2413c752b35/README.md", "type": "primary" }, { "label": "Hy-MT2-7B checkpoint configuration", "url": "https://huggingface.co/tencent/Hy-MT2-7B/blob/9b0eb4e8f001def3e5ff6469a0ac96fdb39ec223/config.json", "type": "primary" }, { "label": "Hy-MT2-7B Apache-2.0 license", "url": "https://huggingface.co/tencent/Hy-MT2-7B/blob/9b0eb4e8f001def3e5ff6469a0ac96fdb39ec223/LICENSE.txt", "type": "primary" }, { "label": "Hy-MT2 technical report", "url": "https://arxiv.org/html/2605.22064v1", "type": "paper" }, { "label": "Hy-MT2 dense training guide", "url": "https://github.com/Tencent-Hunyuan/Hy-MT2/blob/ff1903ecaa724e10951a23c16817a2413c752b35/train/README.md", "type": "primary" }, { "label": "Hy-MT2-7B safetensors index", "url": "https://huggingface.co/tencent/Hy-MT2-7B/blob/9b0eb4e8f001def3e5ff6469a0ac96fdb39ec223/model.safetensors.index.json", "type": "primary" }, { "label": "HunYuanDenseV1 implementation and weight tying", "url": "https://github.com/huggingface/transformers/blob/v5.6.0/src/transformers/models/hunyuan_v1_dense/modeling_hunyuan_v1_dense.py", "type": "primary" }, { "label": "Hy-MT2-7B chat template", "url": "https://huggingface.co/tencent/Hy-MT2-7B/blob/9b0eb4e8f001def3e5ff6469a0ac96fdb39ec223/chat_template.jinja", "type": "primary" }, { "label": "Later Hy-MT2 technical report revision", "url": "https://arxiv.org/html/2605.22064v2", "type": "paper" }, { "label": "Current Tencent Hy-MT2-7B nominal-size specification", "url": "https://huggingface.co/tencent/Hy-MT2-7B/blob/main/README.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/tencent/hy-mt2-7b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "7B is Tencent's nominal size. The 8.029540352B stored-element sum includes separately serialized input/output embeddings despite configured tying; 7.50456832B is the configured unique count if tied, with payload equality unverified. Active parameters remain undisclosed. Supported inference context is 8192; the 262144 configuration ceiling is not a validated context claim. Native reasoning-mode and tool-calling support remain unverified.", "field_status": { "active_parameters_b": "undisclosed", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T18:01:10.264592Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/502.json" }, { "id": "alibaba/qwen3.7-max", "slug": "alibaba--qwen3.7-max", "name": "Qwen3.7 Max", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Text-only Qwen3.7 model for coding and agent tasks, with a one-million-token context window, thinking mode, and function calling.", "release_date": "2026-05-20", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1000000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "source_type": "primary", "sources": [ { "label": "Alibaba Cloud model release history", "url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "type": "primary" }, { "label": "Qwen3.7-Max official model specification", "url": "https://help.aliyun.com/en/model-studio/qwen3-7-max", "type": "primary" }, { "label": "Qwen3.7 launch announcement", "url": "https://qwen.ai/blog?id=qwen3.7&locale=en", "type": "primary" }, { "label": "Alibaba Cloud Qwen3.7-Max README", "url": "https://github.com/AlibabaCloud-Official/Qwen3.7-max-readme/blob/main/README.md", "type": "primary" }, { "label": "Qwen3.8 launch and Qwen3.7 benchmark comparison", "url": "https://qwen.ai/blog?id=qwen3.8", "type": "primary" }, { "label": "Alibaba Cloud MaxCompute supported-model comparison", "url": "https://www.alibabacloud.com/help/en/maxcompute/token-fee", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.7-max.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Qwen3.7-Max is proprietary; total/active parameter counts and internal architecture remain undisclosed. No named-base relationship establishes a nominal size, and Qwen3.8-Max specifications apply to the successor. The qwen3.7-max alias is equivalent to the text-only May 20 snapshot; the separate June 8 snapshot adds vision. Earliest explicit regional release is May 20, 2026 in Global scope in Virginia, Hong Kong and Frankfurt; China and Singapore list May 21.", "field_status": { "license": "not_applicable", "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T09:53:19Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/064.json" }, { "id": "cohere/command-a-plus-05-2026", "slug": "cohere--command-a-plus-05-2026", "name": "Command A+", "developer": "Cohere", "developer_id": "cohere", "family": "command-a", "description": "Cohere's open-weight sparse MoE model for multilingual agents, reasoning, and enterprise tasks, with text and image inputs.", "release_date": "2026-05-20", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 218, "active_parameters_b": 25, "parameters_source": "primary_source", "architecture": "moe", "context_window": 128000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://docs.cohere.com/docs/command-a-plus", "source_type": "primary", "sources": [ { "label": "Cohere model documentation", "url": "https://docs.cohere.com/docs/command-a-plus", "type": "primary" }, { "label": "Cohere launch announcement", "url": "https://cohere.com/blog/command-a-plus", "type": "primary" }, { "label": "Cohere model card", "url": "https://huggingface.co/CohereLabs/command-a-plus-05-2026-bf16/raw/main/README.md", "type": "primary" }, { "label": "Cohere release note", "url": "https://docs.cohere.com/v2/changelog/command-a-plus-05-2026", "type": "primary" }, { "label": "Later Cohere report, Table 1 and section 2.1", "url": "https://arxiv.org/html/2609.13916v1", "type": "primary" }, { "label": "Cohere checkpoint configuration", "url": "https://huggingface.co/CohereLabs/command-a-plus-05-2026-bf16/raw/main/config.json", "type": "primary" }, { "label": "Cohere2 MoE shared-expert implementation", "url": "https://raw.githubusercontent.com/huggingface/transformers/main/src/transformers/models/cohere2_moe/modeling_cohere2_moe.py", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/cohere/command-a-plus-05-2026.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "218B total and 25B active are developer-reported nominal counts, not exact learned-tensor enumerations. The model combines a sparse MoE language backbone with vision support. Active counts are not broken down by modality or execution phase. The documented 128,000-token context is retained despite a larger configuration position capacity.", "field_status": {}, "audited_at": "2026-10-26T18:16:33Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/147.json" }, { "id": "google/gemini-3.5-flash", "slug": "google--gemini-3.5-flash", "name": "Gemini 3.5 Flash", "developer": "Google", "developer_id": "google", "family": "gemini-flash", "description": "Gemini Flash model with multimodal input, text output, thinking and function calling.", "release_date": "2026-05-19", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-5/", "source_type": "primary", "sources": [ { "label": "Gemini 3.5 Flash model card", "url": "https://deepmind.google/models/model-cards/gemini-3-5-flash/", "type": "primary" }, { "label": "Gemini 3 Flash model card", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-Flash-Model-Card.pdf", "type": "primary" }, { "label": "Gemini 3 Pro model card, updated May 2026", "url": "https://deepmind.google/models/model-cards/gemini-3-pro/", "type": "primary" }, { "label": "Gemini 3.5 Flash API model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.5-flash", "type": "primary" }, { "label": "Gemini 3.5: frontier intelligence with action", "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-5/", "type": "primary" }, { "label": "Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemini-3.5-flash.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Sparse MoE Transformer based on Gemini 3 Flash, whose card names Gemini 3 Pro as its base. Total and active parameter counts are undisclosed for this model and its named bases. Specifications describe the May 19, 2026 release; Google now routes API requests using this deprecated identifier to Gemini 3.6 Flash.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T13:02:54.669113Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/207.json" }, { "id": "perceptron/perceptron-mk1", "slug": "perceptron--perceptron-mk1", "name": "Perceptron Mk1", "developer": "Perceptron", "developer_id": "perceptron", "family": "perceptron", "description": "Vision-language model for image and video understanding, spatial grounding, and embodied reasoning, with optional thinking.", "release_date": "2026-05-12", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 32768, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": false, "source_url": "https://www.perceptron.inc/blog/introducing-perceptron-mk1", "source_type": "primary", "sources": [ { "label": "Introducing Perceptron Mk1", "url": "https://www.perceptron.inc/blog/introducing-perceptron-mk1", "type": "primary" }, { "label": "Perceptron Mk1 model guide", "url": "https://docs.perceptron.inc/perceptron-mk1/models/perceptron-mk1.md", "type": "primary" }, { "label": "Mk1 Chat Completions API reference", "url": "https://docs.perceptron.inc/perceptron-mk1/api-reference/endpoint/chat-completions.md", "type": "primary" }, { "label": "Perceptron Mk1 changelog", "url": "https://docs.perceptron.inc/perceptron-mk1/changelog.md", "type": "primary" }, { "label": "Terms of Use", "url": "https://www.perceptron.inc/terms-of-use", "type": "primary" }, { "label": "Mk1 MCP integration", "url": "https://docs.perceptron.inc/perceptron-mk1/guides/mcp.md", "type": "primary" }, { "label": "Introducing Perceptron Mk1.5", "url": "https://www.perceptron.inc/blog/introducing-perceptron-mk1-5", "type": "primary" }, { "label": "Teaching VLMs to think visually", "url": "https://www.perceptron.inc/blog/teaching-vlms-to-think-visually", "type": "primary" }, { "label": "Composing Weight and Data Sparsity in MoE", "url": "https://www.perceptron.inc/blog/composing-weight-and-data-sparsity-in-moe", "type": "primary" }, { "label": "Isaac 0.5: Percepts Scale Control", "url": "https://pub-d90b81cad7254a1aa6b148ac18153c0c.r2.dev/isaac-0.5.pdf", "type": "primary" }, { "label": "Perceptron Mk1 API model page", "url": "https://openrouter.ai/perceptron/perceptron-mk1", "type": "provider" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/perceptron/perceptron-mk1.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Mk1 supports optional reasoning but not function calling; MCP lets other agents invoke its vision capabilities. Total and active parameter counts, architecture and a named base model remain undisclosed. Isaac and Mk1.5 counts do not establish Mk1 specifications. The developer specifies 32K context; its OpenRouter route specifies 32,768 tokens.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T17:34:43.561057Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/464.json" }, { "id": "google/gemini-3.1-flash-lite", "slug": "google--gemini-3.1-flash-lite", "name": "Gemini 3.1 Flash-Lite", "developer": "Google", "developer_id": "google", "family": "gemini-flash-lite", "description": "Low-latency Gemini model for high-volume multimodal and agent workloads", "release_date": "2026-05-07", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-lite", "source_type": "primary", "sources": [ { "label": "Gemini 3.1 Flash-Lite — Gemini API", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-lite", "type": "primary" }, { "label": "Gemini API release notes — May 7 and March 3, 2026", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "primary" }, { "label": "Gemini 3.1 Flash-Lite — Google Cloud model documentation", "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/3-1-flash-lite", "type": "primary" }, { "label": "Gemini 3.1 Flash-Lite Model Card — updated May 2026", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-1-Flash-Lite-Model-Card.pdf", "type": "primary" }, { "label": "Gemini 3 Pro Model Card — updated May 2026", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-Pro-Model-Card.pdf", "type": "primary" }, { "label": "Gemini 3.5 Flash-Lite Model Card — July 21, 2026", "url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "type": "primary" }, { "label": "Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Gemini 3.1 Flash-Lite evaluation methodology — March 2026", "url": "https://deepmind.google/models/evals-methodology/gemini-3-1-flash-lite", "type": "primary" }, { "label": "Vertex AI release notes — March 3, 2026", "url": "https://docs.cloud.google.com/vertex-ai/docs/release-notes", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemini-3.1-flash-lite.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Stable release: May 7, 2026. March 3 was a preview; historical Vertex notes used the unsuffixed name, but unchanged preview-to-stable checkpoint identity is not established. Sparse MoE Transformer based on Gemini 3 Pro; total and active parameter counts are undisclosed for both this model and its named base.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T12:57:39.173876+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/202.json" }, { "id": "openai/gpt-realtime-2", "slug": "openai--gpt-realtime-2", "name": "GPT-realtime-2", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-realtime", "description": null, "release_date": "2026-05-07", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 128000, "modalities": { "input": [ "text", "audio", "image" ], "output": [ "text", "audio" ] }, "reasoning": true, "tool_call": null, "source_url": "https://developers.openai.com/api/docs/changelog", "source_type": "primary", "sources": [ { "label": "Dated publisher API releases", "url": "https://developers.openai.com/api/docs/changelog", "type": "documentation" }, { "label": "Publisher model documentation and snapshot mapping", "url": "https://developers.openai.com/api/docs/models/gpt-realtime-2", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:31:17.277Z", "last_seen": "2026-10-29T10:31:17.277Z", "upstream_present": false, "availability": "public", "notes": "Original publisher conversational audio model. " }, { "id": "baidu/cobuddy", "slug": "baidu--cobuddy", "name": "CoBuddy", "developer": "Baidu", "developer_id": "baidu", "family": null, "description": "Baidu code-generation model offered through a text chat API for coding and agent workflows.", "release_date": "2026-05-06", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openrouter.ai/baidu/cobuddy", "source_type": "secondary", "sources": [ { "label": "Baidu Comate article identifying Cobuddy", "url": "https://cloud.baidu.com/article/8387535", "type": "primary" }, { "label": "Novita CoBuddy model documentation", "url": "https://novita.ai/models/model-detail/baidu-cobuddy?from=pricing", "type": "primary" }, { "label": "Novita exact model and context documentation", "url": "https://blogs.novita.ai/cobuddy-on-novita-ai-coding-llm-api/", "type": "primary" }, { "label": "OpenRouter documented CoBuddy release date", "url": "https://openrouter.ai/baidu/cobuddy", "type": "primary" }, { "label": "Baidu confirms CoBuddy launch and OpenRouter availability", "url": "https://x.com/Baidu_ERNIE/status/2052975339387990391", "type": "primary" }, { "label": "Exact Cobuddy Hugging Face weight search", "url": "https://huggingface.co/api/models?search=cobuddy&limit=100", "type": "primary" }, { "label": "Baidu official public model inventory", "url": "https://huggingface.co/api/models?author=baidu&limit=100", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/baidu/cobuddy.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Baidu identifies Cobuddy as a code-generation model; Novita documents its served capabilities. Nominal total and active counts remain undisclosed; the 21B/3B estimate has no developer-confirmed base lineage. Architecture, license and family remain unresolved. May 6, 2026 is the earliest provider-documented public release, corroborated by Baidu confirming availability on May 9; the original developer announcement date remains unresolved. Classified closed because no public downloadable Cobuddy checkpoint was found.", "field_status": { "family": "unverified", "license": "unverified", "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-28T07:53:40.412749Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/123.json" }, { "id": "openai/gpt-5.5-instant", "slug": "openai--gpt-5.5-instant", "name": "GPT-5.5 Instant", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5.5", "description": "OpenAI conversational model with image understanding, internal reasoning and tool use; launched in the API through chat-latest.", "release_date": "2026-05-05", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": null, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/gpt-5-5-instant/", "source_type": "primary", "sources": [ { "label": "GPT-5.5 Instant: smarter, clearer, and more personalized", "url": "https://openai.com/index/gpt-5-5-instant/", "type": "primary" }, { "label": "GPT-5.5 Instant System Card", "url": "https://deploymentsafety.openai.com/gpt-5-5-instant/gpt-5-5-instant.pdf", "type": "primary" }, { "label": "Chat Latest API model documentation", "url": "https://developers.openai.com/api/docs/models/chat-latest.md", "type": "primary" }, { "label": "OpenAI API changelog", "url": "https://developers.openai.com/api/docs/changelog.md", "type": "primary" }, { "label": "Foundry Models sold by Azure: GPT-chat-latest", "url": "https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure", "type": "primary" }, { "label": "How we built a realtime system for responsive voice AI in six months", "url": "https://openai.com/index/continuous-voice-interaction-with-gpt-live/", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201.md", "type": "primary" }, { "label": "Improving health intelligence in ChatGPT", "url": "https://openai.com/index/improving-health-intelligence-in-chatgpt/", "type": "primary" }, { "label": "GPT-5.6 System Card", "url": "https://deploymentsafety.openai.com/gpt-5-6/gpt-5-6.pdf", "type": "primary" }, { "label": "GPT-6 Astra System Card", "url": "https://deploymentsafety.openai.com/gpt-6-astra/gpt-6-astra.pdf", "type": "primary" }, { "label": "Public Summary of Training Content for GPT-5.5", "url": "https://cdn.openai.com/pdf/gpt-5-5-eu-ai-act-public-summary-of-training-content.pdf", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "OpenAI API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "Model Release Notes", "url": "https://help.openai.com/en/articles/9624314-model-release-notes", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "OpenAI model inventory", "url": "https://developers.openai.com/api/docs/models/all.md", "type": "primary" }, { "label": "Addendum: GPT-6.1 Sol System Card", "url": "https://deploymentsafety.openai.com/gpt-6-1-sol/gpt-6-1-sol.pdf", "type": "primary" }, { "label": "Addendum to GPT-6 Astra System Card: GPT-6.1 Terra", "url": "https://deploymentsafety.openai.com/gpt-6-1-terra/gpt-6-1-terra.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5.5-instant.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released May 5, 2026 through ChatGPT and the mutable chat-latest API alias; the original card cover is May 4. The catalog ID is not verified as a standalone API endpoint. Subsequent alias updates include May 28, June 24, August 6 and October 7. Current chat-latest documentation states 400K, while Azure May/June versions state 128K; neither establishes an exact OpenAI-hosted capacity for the original model, so context is unverified. Reasoning was deployed at low effort; xhigh was used for capability evaluation, not documented as a selectable public control. PDF is an API format. Transformer is a qualified inference from example-bound Instant delegation caching and the pre-GPT-5.6 KV guide, not a complete topology disclosure. Total and active parameters remain undisclosed; no published nominal count or explicit counted-base lineage was found for Instant. GPT-5.5 Thinking/Pro and successor models are distinct evidence scopes. No exact API shutdown notice was found, and current alias identity or endpoint operation was not tested.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "context_window": "unverified" }, "audited_at": "2026-10-27T17:00:33.425303Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/435.json" }, { "id": "openbmb/minicpm5-1b", "slug": "openbmb--minicpm5-1b", "name": "MiniCPM5-1B", "developer": "OpenBMB", "developer_id": "openbmb", "family": "minicpm5", "description": "Dense text model with optional reasoning, function calling, and a 131,072-token context.", "release_date": "2026-05", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 1.080632832, "active_parameters_b": 1.080632832, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/OpenBMB/MiniCPM/releases/tag/5.0", "source_type": "primary", "sources": [ { "label": "MiniCPM5-1B model card", "url": "https://huggingface.co/openbmb/MiniCPM5-1B/blob/87179e5c1f455ef22e6223592d2d61351b525bfc/README.md", "type": "primary" }, { "label": "MiniCPM5-1B pinned configuration", "url": "https://huggingface.co/openbmb/MiniCPM5-1B/blob/87179e5c1f455ef22e6223592d2d61351b525bfc/config.json", "type": "primary" }, { "label": "MiniCPM5-1B checkpoint index", "url": "https://huggingface.co/openbmb/MiniCPM5-1B/blob/87179e5c1f455ef22e6223592d2d61351b525bfc/model.safetensors.index.json", "type": "primary" }, { "label": "MiniCPM5-1B complete bounded safetensors header", "url": "https://huggingface.co/openbmb/MiniCPM5-1B/resolve/87179e5c1f455ef22e6223592d2d61351b525bfc/model-00000-of-00001.safetensors", "type": "primary" }, { "label": "Transformers v5.6.2 Llama implementation", "url": "https://github.com/huggingface/transformers/blob/v5.6.2/src/transformers/models/llama/modeling_llama.py", "type": "primary" }, { "label": "Transformers v5.6.2 Llama configuration defaults", "url": "https://github.com/huggingface/transformers/blob/v5.6.2/src/transformers/models/llama/configuration_llama.py", "type": "primary" }, { "label": "MiniCPM5-1B chat template", "url": "https://huggingface.co/openbmb/MiniCPM5-1B/blob/87179e5c1f455ef22e6223592d2d61351b525bfc/chat_template.jinja", "type": "primary" }, { "label": "MiniCPM5-1B knowledge-base guide", "url": "https://openbmb.cn/docs/en/guides/minicpm-text/minicpm5-1b", "type": "primary" }, { "label": "MiniCPM5-1B release-tag changelog", "url": "https://github.com/OpenBMB/MiniCPM/blob/5.0/README.md", "type": "primary" }, { "label": "MiniCPM5-1B version 5.0 release", "url": "https://github.com/OpenBMB/MiniCPM/releases/tag/5.0", "type": "primary" }, { "label": "MiniCPM5-1B checkpoint history", "url": "https://huggingface.co/openbmb/MiniCPM5-1B/commits/main", "type": "primary" }, { "label": "Paiton small-model qualification: exact-checkpoint parameter comparison", "url": "https://huggingface.co/EliovpAI/MiniCPM5-2B-W4A16-Paiton-RDNA4/raw/4bcf80e1bcf1bc8f5c2d657086121958cfe145cb/SELECTION.md", "type": "primary" }, { "label": "Paiton MiniCPM5-2B original runtime benchmark methodology", "url": "https://huggingface.co/EliovpAI/MiniCPM5-2B-W4A16-Paiton-RDNA4/raw/4bcf80e1bcf1bc8f5c2d657086121958cfe145cb/BENCHMARKS.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openbmb/minicpm5-1b.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/openbmb/MiniCPM5-1B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "OpenBMB publishes 1,080,632,832 total learned parameters, including separate input/output vocabulary matrices; its 679,552,512 non-embedding count excludes both. Active 1.080632832B uses EliovpAI/Paiton's published convention for this pinned model, an unmeasured comparison alternative, rather than an OpenBMB per-token disclosure. Native context is 131,072 tokens. Release month is May 2026: the changelog says May 19, while the formal release was May 26; earliest public day remains unresolved. HF creation/upload dates do not prove public availability.", "field_status": {}, "audited_at": "2026-10-27T17:29:27.253741Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/461.json" }, { "id": "xai/grok-build-0.1", "slug": "xai--grok-build-0.1", "name": "Grok Build 0.1", "developer": "xAI", "developer_id": "xai", "family": "grok-build", "description": "Coding model trained for agentic software engineering, with text and image input, reasoning, and tool calling.", "release_date": "2026-05", "date_precision": "month", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": null, "context_window": 256000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://x.ai/news/grok-build-0-1", "source_type": "primary", "sources": [ { "label": "Grok Build 0.1 native model documentation", "url": "https://docs.x.ai/developers/models/grok-build-0.1", "type": "primary" }, { "label": "Grok Build 0.1 on API, May 29, 2026", "url": "https://x.ai/news/grok-build-0-1", "type": "primary" }, { "label": "Release notes: Grok Build beta and Build 0.1 early access", "url": "https://docs.x.ai/developers/release-notes", "type": "primary" }, { "label": "May 15, 2026 Model Retirement", "url": "https://docs.x.ai/developers/migration/may-15-retirement", "type": "primary" }, { "label": "Chat with Files", "url": "https://docs.x.ai/developers/model-capabilities/files/chat-with-files", "type": "primary" }, { "label": "Enterprise terms, May 12, 2026", "url": "https://x.ai/legal/terms-of-service-enterprise/previous-2026-05-12", "type": "primary" }, { "label": "Grok Build is Now Open Source, July 15, 2026", "url": "https://x.ai/news/grok-build-open-source", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/xai/grok-build-0.1.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released in May 2026; the exact first public day is unresolved across the harness beta, compatibility routing, early access and public API beta milestones. Total and active parameter counts remain undisclosed in the reviewed originals; no counted base is identified. Unattributed 309B/15B figures are unverified. Neural architecture is undisclosed. The open-source Grok Build harness and its newer default models are separate from Build 0.1; Code Fast aliases are compatibility redirects. Native inputs are text and images; PDF attachments use document-search tools.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T18:34:03.075064+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/535.json" }, { "id": "ibm/granite-4.1-30b", "slug": "ibm--granite-4.1-30b", "name": "granite-4.1-30b", "developer": "IBM", "developer_id": "ibm", "family": "granite", "description": null, "release_date": "2026-04-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 30, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://research.ibm.com/blog/granite-4-1-ai-foundation-models", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://research.ibm.com/blog/granite-4-1-ai-foundation-models", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/ibm-granite/granite-4.1-30b/blob/4fae6278f7132abf5e971f9de49ebbad09c54cce/README.md", "type": "model_card" }, { "label": "Original pretraining and architecture report", "url": "https://huggingface.co/blog/ibm-granite/granite-4-1", "type": "technical_report" }, { "label": "Official model configuration", "url": "https://huggingface.co/ibm-granite/granite-4.1-30b/blob/4fae6278f7132abf5e971f9de49ebbad09c54cce/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:23.459Z", "last_seen": "2026-10-29T10:17:23.459Z", "upstream_present": false, "availability": "public", "notes": "Instruction-tuned checkpoint from IBM’s independently pretrained Granite 4.1 lineage. Nominal parameter count. Context records the 131,072-token published configuration; the technical article also describes longer training phases for the 8B and 30B sizes." }, { "id": "ibm/granite-4.1-30b-base", "slug": "ibm--granite-4.1-30b-base", "name": "granite-4.1-30b-base", "developer": "IBM", "developer_id": "ibm", "family": "granite", "description": null, "release_date": "2026-04-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 30, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://research.ibm.com/blog/granite-4-1-ai-foundation-models", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://research.ibm.com/blog/granite-4-1-ai-foundation-models", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/ibm-granite/granite-4.1-30b-base/blob/9bb22a92d17f4a51788c4750d32764ed068d94f2/README.md", "type": "model_card" }, { "label": "Original pretraining and architecture report", "url": "https://huggingface.co/blog/ibm-granite/granite-4-1", "type": "technical_report" }, { "label": "Official model configuration", "url": "https://huggingface.co/ibm-granite/granite-4.1-30b-base/blob/9bb22a92d17f4a51788c4750d32764ed068d94f2/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:23.459Z", "last_seen": "2026-10-29T10:17:23.459Z", "upstream_present": false, "availability": "public", "notes": "Pretrained checkpoint from IBM’s independently pretrained Granite 4.1 lineage. Nominal parameter count. Context records the 131,072-token published configuration; the technical article also describes longer training phases for the 8B and 30B sizes." }, { "id": "ibm/granite-4.1-3b", "slug": "ibm--granite-4.1-3b", "name": "granite-4.1-3b", "developer": "IBM", "developer_id": "ibm", "family": "granite", "description": null, "release_date": "2026-04-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 3, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://research.ibm.com/blog/granite-4-1-ai-foundation-models", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://research.ibm.com/blog/granite-4-1-ai-foundation-models", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/ibm-granite/granite-4.1-3b/blob/c0650403e44e78ec0262dab1c90914c65b196c4e/README.md", "type": "model_card" }, { "label": "Original pretraining and architecture report", "url": "https://huggingface.co/blog/ibm-granite/granite-4-1", "type": "technical_report" }, { "label": "Official model configuration", "url": "https://huggingface.co/ibm-granite/granite-4.1-3b/blob/c0650403e44e78ec0262dab1c90914c65b196c4e/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:23.459Z", "last_seen": "2026-10-29T10:17:23.459Z", "upstream_present": false, "availability": "public", "notes": "Instruction-tuned checkpoint from IBM’s independently pretrained Granite 4.1 lineage. Nominal parameter count. Context records the 131,072-token published configuration; the technical article also describes longer training phases for the 8B and 30B sizes." }, { "id": "ibm/granite-4.1-3b-base", "slug": "ibm--granite-4.1-3b-base", "name": "granite-4.1-3b-base", "developer": "IBM", "developer_id": "ibm", "family": "granite", "description": null, "release_date": "2026-04-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 3, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://research.ibm.com/blog/granite-4-1-ai-foundation-models", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://research.ibm.com/blog/granite-4-1-ai-foundation-models", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/ibm-granite/granite-4.1-3b-base/blob/dacb9cb9157bec98e99b09f285c92a4d58405c96/README.md", "type": "model_card" }, { "label": "Original pretraining and architecture report", "url": "https://huggingface.co/blog/ibm-granite/granite-4-1", "type": "technical_report" }, { "label": "Official model configuration", "url": "https://huggingface.co/ibm-granite/granite-4.1-3b-base/blob/dacb9cb9157bec98e99b09f285c92a4d58405c96/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:23.459Z", "last_seen": "2026-10-29T10:17:23.459Z", "upstream_present": false, "availability": "public", "notes": "Pretrained checkpoint from IBM’s independently pretrained Granite 4.1 lineage. Nominal parameter count. Context records the 131,072-token published configuration; the technical article also describes longer training phases for the 8B and 30B sizes." }, { "id": "ibm/granite-4.1-8b", "slug": "ibm--granite-4.1-8b", "name": "granite-4.1-8b", "developer": "IBM", "developer_id": "ibm", "family": "granite", "description": null, "release_date": "2026-04-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://research.ibm.com/blog/granite-4-1-ai-foundation-models", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://research.ibm.com/blog/granite-4-1-ai-foundation-models", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/ibm-granite/granite-4.1-8b/blob/1504002f650e656a0a3789d99574df12e3e94ed0/README.md", "type": "model_card" }, { "label": "Original pretraining and architecture report", "url": "https://huggingface.co/blog/ibm-granite/granite-4-1", "type": "technical_report" }, { "label": "Official model configuration", "url": "https://huggingface.co/ibm-granite/granite-4.1-8b/blob/1504002f650e656a0a3789d99574df12e3e94ed0/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:23.459Z", "last_seen": "2026-10-29T10:17:23.459Z", "upstream_present": false, "availability": "public", "notes": "Instruction-tuned checkpoint from IBM’s independently pretrained Granite 4.1 lineage. Nominal parameter count. Context records the 131,072-token published configuration; the technical article also describes longer training phases for the 8B and 30B sizes." }, { "id": "ibm/granite-4.1-8b-base", "slug": "ibm--granite-4.1-8b-base", "name": "granite-4.1-8b-base", "developer": "IBM", "developer_id": "ibm", "family": "granite", "description": null, "release_date": "2026-04-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://research.ibm.com/blog/granite-4-1-ai-foundation-models", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://research.ibm.com/blog/granite-4-1-ai-foundation-models", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/ibm-granite/granite-4.1-8b-base/blob/cbc32f94fbf8179ca4b22b664831f6a73965bb16/README.md", "type": "model_card" }, { "label": "Original pretraining and architecture report", "url": "https://huggingface.co/blog/ibm-granite/granite-4-1", "type": "technical_report" }, { "label": "Official model configuration", "url": "https://huggingface.co/ibm-granite/granite-4.1-8b-base/blob/cbc32f94fbf8179ca4b22b664831f6a73965bb16/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:23.459Z", "last_seen": "2026-10-29T10:17:23.459Z", "upstream_present": false, "availability": "public", "notes": "Pretrained checkpoint from IBM’s independently pretrained Granite 4.1 lineage. Nominal parameter count. Context records the 131,072-token published configuration; the technical article also describes longer training phases for the 8B and 30B sizes." }, { "id": "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning", "slug": "nvidia--nemotron-3-nano-omni-30b-a3b-reasoning", "name": "Nemotron 3 Nano Omni 30B A3B Reasoning", "developer": "NVIDIA", "developer_id": "nvidia", "family": "nemotron", "description": "Mamba2-Transformer MoE reasoning model with C-RADIO vision and Parakeet audio encoders; accepts text, images, video and audio and generates text.", "release_date": "2026-04-28", "date_precision": "day", "open_weights": true, "license": "NVIDIA Open Model Agreement", "parameters_b": 33.015546816, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "hybrid-ssm-transformer-moe", "context_window": 262144, "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16/blob/e5e9932441de940c9a62185c870ea5bcd4cd24e2/README.md", "source_type": "primary", "sources": [ { "label": "NVIDIA model card", "url": "https://huggingface.co/nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16/blob/e5e9932441de940c9a62185c870ea5bcd4cd24e2/README.md", "type": "primary" }, { "label": "NVIDIA launch announcement", "url": "https://blogs.nvidia.com/blog/nemotron-3-nano-omni-multimodal-ai-agents/", "type": "primary" }, { "label": "NVIDIA technical report v2", "url": "https://arxiv.org/html/2604.24954v2", "type": "primary" }, { "label": "NVIDIA checkpoint configuration", "url": "https://huggingface.co/nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16/blob/e5e9932441de940c9a62185c870ea5bcd4cd24e2/config.json", "type": "primary" }, { "label": "Complete BF16 weight index", "url": "https://huggingface.co/nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16/blob/e5e9932441de940c9a62185c870ea5bcd4cd24e2/model.safetensors.index.json", "type": "weights" }, { "label": "NVIDIA decoder activity scope", "url": "https://docs.nvidia.com/nemotron/nightly/nemotron/omni3/architecture.html", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/nvidia/nemotron-3-nano-omni-30b-a3b-reasoning.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete BF16 checkpoint: 33.015546816B learned parameters, including vision/audio encoders and projectors. The advertised ~3B active figure is nominal and language-backbone-only; full multimodal activity varies by input and phase. Decoder uses 23 Mamba, 6 attention and 23 MoE blocks with 6 of 128 routed experts plus a shared expert. The report and nested language config support 262144 tokens; top-level max_sequence_length is 131072, so deployment settings may impose a lower limit.", "active_parameters_scope": "language_backbone", "parameter_components_b": { "language_backbone": 31.577937344, "vision_encoder": 0.65361152, "vision_projector": 0.15991296, "audio_encoder": 0.608879616, "audio_projector": 0.015205376 }, "field_status": {}, "audited_at": "2026-10-27T00:43:18.668026Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/382.json" }, { "id": "poolside/laguna-m.1", "slug": "poolside--laguna-m.1", "name": "Laguna M.1", "developer": "Poolside", "developer_id": "poolside", "family": "laguna", "description": "Poolside's text-only mixture-of-experts model for agentic coding, with optional interleaved reasoning and tool calling.", "release_date": "2026-04-28", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 225.8, "active_parameters_b": 23.4, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://poolside.ai/blog/introducing-laguna-xs2-m1", "source_type": "primary", "sources": [ { "label": "Introducing Laguna XS.2 and Laguna M.1 — April 28, 2026", "url": "https://poolside.ai/blog/introducing-laguna-xs2-m1", "type": "primary" }, { "label": "Laguna M.1/XS.2 Technical Report — May 25, 2026", "url": "https://poolside.ai/assets/laguna/laguna-m1-xs2-technical-report.pdf", "type": "primary" }, { "label": "Laguna M.1 pinned official model card", "url": "https://huggingface.co/poolside/Laguna-M.1/blob/2bf8a4ab6d2e8e1e0d92a41318532a0fee1640d6/README.md", "type": "primary" }, { "label": "Laguna M.1 pinned configuration", "url": "https://huggingface.co/poolside/Laguna-M.1/blob/2bf8a4ab6d2e8e1e0d92a41318532a0fee1640d6/config.json", "type": "primary" }, { "label": "Laguna M.1 pinned weight index and all indexed shard headers", "url": "https://huggingface.co/poolside/Laguna-M.1/blob/2bf8a4ab6d2e8e1e0d92a41318532a0fee1640d6/model.safetensors.index.json", "type": "primary" }, { "label": "Laguna M.1 pinned complete model implementation", "url": "https://huggingface.co/poolside/Laguna-M.1/blob/2bf8a4ab6d2e8e1e0d92a41318532a0fee1640d6/modeling_laguna.py", "type": "primary" }, { "label": "Laguna M.1 pinned Apache License 2.0", "url": "https://huggingface.co/poolside/Laguna-M.1/blob/2bf8a4ab6d2e8e1e0d92a41318532a0fee1640d6/LICENSE.md", "type": "primary" }, { "label": "Long context update: Laguna XS.2 and M.1 — May 26, 2026", "url": "https://poolside.ai/blog/long-context-update-laguna-xs-2-and-m-1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/poolside/laguna-m.1.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Total and active parameter counts are Poolside’s nominal 225.8B and 23.4B; the active figure explicitly includes embeddings. Current BF16 checkpoint headers contain 225,795,725,568 elements, including 17,152 non-gradient router correction values; the storage total is not presented as an exact learned-parameter count. First publicly available by API on April 28, 2026; open-weight availability is later. The 262,144-token context was announced for serving on May 26 and uses YaRN (original position limit 4,096, factor 64), not unscaled native context.", "field_status": {}, "audited_at": "2026-10-27T03:45:56.877733Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/470.json" }, { "id": "poolside/laguna-xs.2", "slug": "poolside--laguna-xs.2", "name": "Laguna XS.2", "developer": "Poolside", "developer_id": "poolside", "family": "laguna", "description": "Poolside text model for agentic coding, with open weights, optional reasoning and tool calling.", "release_date": "2026-04-28", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 33.4, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://poolside.ai/blog/introducing-laguna-xs2-m1", "source_type": "primary", "sources": [ { "label": "Introducing Laguna XS.2 and Laguna M.1 — April 28, 2026", "url": "https://poolside.ai/blog/introducing-laguna-xs2-m1", "type": "primary" }, { "label": "Laguna M.1/XS.2 Technical Report — May 25, 2026", "url": "https://poolside.ai/assets/laguna/laguna-m1-xs2-technical-report.pdf", "type": "primary" }, { "label": "Laguna XS.2 canonical BF16 model card, pinned revision", "url": "https://huggingface.co/poolside/Laguna-XS.2/blob/a397bde04501591d0ac3d74a5d9d0a4ec298ffb8/README.md", "type": "primary" }, { "label": "Laguna XS.2 exact BF16 model configuration", "url": "https://huggingface.co/poolside/Laguna-XS.2/blob/a397bde04501591d0ac3d74a5d9d0a4ec298ffb8/config.json", "type": "primary" }, { "label": "Laguna XS.2 selected model implementation", "url": "https://huggingface.co/poolside/Laguna-XS.2/blob/a397bde04501591d0ac3d74a5d9d0a4ec298ffb8/modeling_laguna.py", "type": "primary" }, { "label": "Laguna XS.2 configuration defaults", "url": "https://huggingface.co/poolside/Laguna-XS.2/blob/a397bde04501591d0ac3d74a5d9d0a4ec298ffb8/configuration_laguna.py", "type": "primary" }, { "label": "Laguna XS.2 complete checkpoint index and all 14 pinned shard headers", "url": "https://huggingface.co/poolside/Laguna-XS.2/blob/a397bde04501591d0ac3d74a5d9d0a4ec298ffb8/model.safetensors.index.json", "type": "primary" }, { "label": "Laguna XS.2 model-materials license", "url": "https://huggingface.co/poolside/Laguna-XS.2/blob/a397bde04501591d0ac3d74a5d9d0a4ec298ffb8/LICENSE.md", "type": "primary" }, { "label": "Laguna XS.2 native chat template", "url": "https://huggingface.co/poolside/Laguna-XS.2/blob/a397bde04501591d0ac3d74a5d9d0a4ec298ffb8/chat_template.jinja", "type": "primary" }, { "label": "Long context update: Laguna XS.2 and M.1 — May 26, 2026", "url": "https://poolside.ai/blog/long-context-update-laguna-xs-2-and-m-1", "type": "primary" }, { "label": "Introducing Laguna XS 2.1 — July 2, 2026", "url": "https://poolside.ai/blog/introducing-laguna-xs-2-1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/poolside/laguna-xs.2.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Parameters are Poolside’s nominal 33.4B total and 3B active per token, including embeddings. At BF16 revision a397bde04501591d0ac3d74a5d9d0a4ec298ffb8, all 14 shard headers reconcile to 33,442,617,088 stored values; excluding 9,984 frozen routing-correction values leaves 33,442,607,104 learned-weight elements. Both untied vocabulary matrices are included; the optional separate DFlash draft is excluded. The 262,144-token limit was released in May 2026 using YaRN after training through 128K; it is not native 256K training. Reasoning is optional and the template defaults it off unless enabled. The July 2 successor announcement scheduled this original model’s API sunset after one week; open weights remain available.", "field_status": {}, "audited_at": "2026-10-27T03:49:49.085706Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/473.json" }, { "id": "deepseek/deepseek-v4-flash", "slug": "deepseek--deepseek-v4-flash", "name": "DeepSeek V4 Flash", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-flash", "description": "Original DeepSeek V4 Flash preview checkpoint with text input/output, thinking and non-thinking modes, tool calling, and a 1M-token context.", "release_date": "2026-04-24", "date_precision": "day", "open_weights": true, "license": "MIT License", "parameters_b": 290.942289362, "active_parameters_b": 13, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash/blob/60d8d70770c6776ff598c94bb586a859a38244f1/README.md", "source_type": "primary", "sources": [ { "label": "DeepSeek-V4-Flash original model card", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash/blob/60d8d70770c6776ff598c94bb586a859a38244f1/README.md", "type": "primary" }, { "label": "Official V4 preview release", "url": "https://api-docs.deepseek.com/news/news260424/", "type": "primary" }, { "label": "Pinned V4-Flash configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash/blob/60d8d70770c6776ff598c94bb586a859a38244f1/config.json", "type": "weights" }, { "label": "DeepSeek-V4 technical report", "url": "https://arxiv.org/html/2606.19348v1", "type": "primary" }, { "label": "Official inference model and parameter layout", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash/blob/60d8d70770c6776ff598c94bb586a859a38244f1/inference/model.py", "type": "primary" }, { "label": "Official packed FP4 conversion", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash/blob/60d8d70770c6776ff598c94bb586a859a38244f1/inference/convert.py", "type": "primary" }, { "label": "Checkpoint index and bounded tensor headers", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash/resolve/60d8d70770c6776ff598c94bb586a859a38244f1/model.safetensors.index.json", "type": "weights" }, { "label": "Original reasoning and tool encoding", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash/blob/60d8d70770c6776ff598c94bb586a859a38244f1/encoding/README.md", "type": "primary" }, { "label": "API alias lifecycle", "url": "https://api-docs.deepseek.com/updates/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepseek/deepseek-v4-flash.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Historical April 24, 2026 DeepSeek-V4-Flash preview checkpoint, preserved from the baseline name/date/weight link. Complete logical total is 290.942289362B: 284.332240471B main (advertised 284B) plus 6.610048891B MTP; excludes packed-format scale metadata and fixed routing indices. Nominal 13B active refers to the main model. Advertised context is 1M; config positional ceiling is 1048576. July 31 V4-Flash-0731 preserved core architecture/size but changed post-training. On September 10 the API identifier deepseek-v4-flash was redirected to V4.1 Flash, whose October 17 upgrade is V4.1-Flash-1017. The current Models.dev provider record tracks that legacy alias; its successor specs are not substituted for this checkpoint. Live endpoint behavior was not probed.", "active_parameters_scope": "main", "parameter_components_b": { "main": 284.332240471, "mtp": 6.610048891 }, "field_status": {}, "audited_at": "2026-10-26T19:11:10.974851+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/177.json" }, { "id": "deepseek/deepseek-v4-flash-0423", "slug": "deepseek--deepseek-v4-flash-0423", "name": "DeepSeek V4 Flash 0423", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-flash", "description": "April 2026 DeepSeek V4 Flash preview: a text-only MoE language model with hybrid CSA/HCA attention, thinking and non-thinking modes, tool calling, and an advertised 1M-token context.", "release_date": "2026-04-24", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 290.942289362, "active_parameters_b": 13, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.deepseek.com/en/news/v4-preview/", "source_type": "primary", "sources": [ { "label": "DeepSeek-V4 Preview announcement, April 24, 2026", "url": "https://www.deepseek.com/en/news/v4-preview/", "type": "primary" }, { "label": "Pinned DeepSeek-V4-Flash preview model card", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash/blob/60d8d70770c6776ff598c94bb586a859a38244f1/README.md", "type": "primary" }, { "label": "OpenRouter: exact V4 checkpoint names and modality mapping", "url": "https://openrouter.ai/blog/insights/deepseek-v4-vision/", "type": "primary" }, { "label": "OpenRouter dated DeepSeek V4 Flash 0423 model page", "url": "https://openrouter.ai/deepseek/deepseek-v4-flash-20260423/", "type": "primary" }, { "label": "Pinned DeepSeek-V4-Flash configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash/resolve/60d8d70770c6776ff598c94bb586a859a38244f1/config.json", "type": "primary" }, { "label": "DeepSeek-V4 technical report", "url": "https://arxiv.org/html/2606.19348v1", "type": "primary" }, { "label": "Pinned preview tensor index and all 46 indexed safetensors headers", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash/resolve/60d8d70770c6776ff598c94bb586a859a38244f1/model.safetensors.index.json", "type": "primary" }, { "label": "Pinned official DeepSeek-V4 inference implementation", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash/resolve/60d8d70770c6776ff598c94bb586a859a38244f1/inference/model.py", "type": "primary" }, { "label": "Pinned official DeepSeek-V4 checkpoint conversion code", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash/resolve/60d8d70770c6776ff598c94bb586a859a38244f1/inference/convert.py", "type": "primary" }, { "label": "Pinned DeepSeek-V4 conversation and tool encoding", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash/resolve/60d8d70770c6776ff598c94bb586a859a38244f1/encoding/README.md", "type": "primary" }, { "label": "DeepSeek-V4-Flash MIT license", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash/raw/main/LICENSE", "type": "primary" }, { "label": "DeepSeek API change log", "url": "https://api-docs.deepseek.com/updates/", "type": "primary" }, { "label": "Models.dev exact 0423 record and source mapping", "url": "https://raw.githubusercontent.com/anomalyco/models.dev/dev/models/deepseek/deepseek-v4-flash-0423.toml", "type": "catalog" }, { "label": "DeepSeek-V4-Flash official repository history", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash/commits/main", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepseek/deepseek-v4-flash-0423.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The 0423 label maps to the original April V4 Flash preview; official public release was April 24, 2026. Complete checkpoint has 290.942289362B logical parameters: 284.332240471B main plus 6.610048891B unique MTP. Advertised 284B/13B describe the rounded main-model total/activation; quantization metadata and fixed hash indices are excluded. Context is advertised 1M and configured as 1048576. July 31 GA and later V4.1/vision revisions are different checkpoints; the current DeepSeek API stable alias has changed since release. No live endpoint operation is claimed.", "active_parameters_scope": "main", "parameter_components_b": { "main": 284.332240471, "mtp": 6.610048891 }, "field_status": {}, "audited_at": "2026-10-26T19:12:43.852629+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/178.json" }, { "id": "deepseek/deepseek-v4-pro", "slug": "deepseek--deepseek-v4-pro", "name": "DeepSeek V4 Pro", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-thinking", "description": "April 2026 DeepSeek-V4 preview: open-weight text MoE with hybrid compressed attention, million-token context, reasoning modes and tool calling.", "release_date": "2026-04-24", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 1598.837347742, "active_parameters_b": 49, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://api-docs.deepseek.com/news/news260424/", "source_type": "primary", "sources": [ { "label": "DeepSeek V4 Preview Release, April 24, 2026", "url": "https://api-docs.deepseek.com/news/news260424/", "type": "primary" }, { "label": "Original DeepSeek-V4-Pro model card", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/b5968e9190ef611bbf34a7229255be88a0e937c1/README.md", "type": "primary" }, { "label": "DeepSeek-V4-Pro checkpoint configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/b5968e9190ef611bbf34a7229255be88a0e937c1/config.json", "type": "primary" }, { "label": "DeepSeek-V4-Pro reference inference model", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/b5968e9190ef611bbf34a7229255be88a0e937c1/inference/model.py", "type": "primary" }, { "label": "DeepSeek-V4-Pro checkpoint conversion", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/b5968e9190ef611bbf34a7229255be88a0e937c1/inference/convert.py", "type": "primary" }, { "label": "DeepSeek-V4-Pro weight index and all indexed shard headers", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/b5968e9190ef611bbf34a7229255be88a0e937c1/model.safetensors.index.json", "type": "weights" }, { "label": "DeepSeek-V4-Pro MIT license", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/b5968e9190ef611bbf34a7229255be88a0e937c1/LICENSE", "type": "primary" }, { "label": "DeepSeek-V4 prompt and tool-call encoding", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/b5968e9190ef611bbf34a7229255be88a0e937c1/encoding/README.md", "type": "primary" }, { "label": "DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence", "url": "https://arxiv.org/html/2606.19348v1", "type": "primary" }, { "label": "DeepSeek current API model mapping", "url": "https://api-docs.deepseek.com/quick_start/pricing/", "type": "primary" }, { "label": "DeepSeek-V4-Pro GA Release, August 13, 2026", "url": "https://api-docs.deepseek.com/news/news260813/", "type": "primary" }, { "label": "Original DeepSeek-V4-Pro repository history", "url": "https://huggingface.co/api/models/deepseek-ai/DeepSeek-V4-Pro/commits/main", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepseek/deepseek-v4-pro.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "April 24, 2026 preview checkpoint. Complete logical weights total 1598.837347742B: 1572.997201763B main plus 25.840145979B unique MTP; packed FP4 weights expanded, quantization scales/fixed routing indices excluded, shared embedding/head counted once. Official 1.6T is nominal; official 49B active describes main-model activation, not additional MTP work. Advertised context is 1M, while configured positional ceiling is 1048576. Current API alias deepseek-v4-pro maps to the later DeepSeek-V4-Pro-0813; this record preserves the original preview linked by its baseline date and checkpoint. No endpoint execution was tested.", "active_parameters_scope": "main", "parameter_components_b": { "main": 1572.997201763, "mtp": 25.840145979 }, "field_status": {}, "audited_at": "2026-10-26T19:16:11.457805Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/181.json" }, { "id": "deepseek/deepseek-v4-pro-0423", "slug": "deepseek--deepseek-v4-pro-0423", "name": "DeepSeek V4 Pro 0423", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-thinking", "description": "DeepSeek V4 Pro initial snapshot with million-token context and support for thinking and non-thinking modes", "release_date": "2026-04-24", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 1598.837347742, "active_parameters_b": 49, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/b5968e9190ef611bbf34a7229255be88a0e937c1/README.md", "source_type": "primary", "sources": [ { "label": "OpenRouter exact V4 variant mapping", "url": "https://openrouter.ai/blog/insights/deepseek-v4-vision/", "type": "primary" }, { "label": "DeepSeek V4 Preview Release, April 24, 2026", "url": "https://api-docs.deepseek.com/news/news260424/", "type": "primary" }, { "label": "Pinned original DeepSeek-V4-Pro model card", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/b5968e9190ef611bbf34a7229255be88a0e937c1/README.md", "type": "primary" }, { "label": "Pinned DeepSeek-V4-Pro configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/b5968e9190ef611bbf34a7229255be88a0e937c1/config.json", "type": "primary" }, { "label": "Official DeepSeek-V4-Pro inference implementation", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/b5968e9190ef611bbf34a7229255be88a0e937c1/inference/model.py", "type": "primary" }, { "label": "Official FP4 checkpoint conversion", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/b5968e9190ef611bbf34a7229255be88a0e937c1/inference/convert.py", "type": "primary" }, { "label": "Pinned index and all 64 bounded safetensors headers", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/resolve/b5968e9190ef611bbf34a7229255be88a0e937c1/model.safetensors.index.json", "type": "weights" }, { "label": "Original V4 text, reasoning and tool encoding", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/b5968e9190ef611bbf34a7229255be88a0e937c1/encoding/README.md", "type": "primary" }, { "label": "DeepSeek-V4-Pro MIT license", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/blob/b5968e9190ef611bbf34a7229255be88a0e937c1/LICENSE", "type": "primary" }, { "label": "DeepSeek-V4 technical report", "url": "https://arxiv.org/html/2606.19348v1", "type": "primary" }, { "label": "DeepSeek release and alias change log", "url": "https://api-docs.deepseek.com/updates/", "type": "primary" }, { "label": "Official checkpoint revision history", "url": "https://huggingface.co/api/models/deepseek-ai/DeepSeek-V4-Pro/commits/main", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepseek/deepseek-v4-pro-0423.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "0423 identifies the original DeepSeek-V4-Pro preview, publicly released April 24, 2026; it is separate from the August 13 GA model. Complete logical checkpoint total is 1598.837347742B: 1572.997201763B main plus 25.840145979B unique MTP. Count expands packed FP4 weights and excludes quantization scales/fixed routing indices while retaining learned mHC scales and parameterized gate biases. Shared embedding/head counted once. DeepSeek reports nominal 1.6T total and 49B active; active figure describes main-model inference rather than combined MTP drafting. Advertised context is 1M; configured positional ceiling is 1048576. The official deepseek-v4-pro API alias was upgraded to GA on August 13, so it no longer establishes this dated preview identity. No endpoint execution was tested.", "active_parameters_scope": "main", "parameter_components_b": { "main": 1572.997201763, "mtp": 25.840145979 }, "field_status": {}, "audited_at": "2026-10-26T19:16:28.661059+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/182.json" }, { "id": "openai/gpt-5.5", "slug": "openai--gpt-5.5", "name": "GPT-5.5", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5.5", "description": "OpenAI reasoning model for coding and general tasks, with text and image input, tool calling and computer use.", "release_date": "2026-04-23", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 1050000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-gpt-5-5/", "source_type": "primary", "sources": [ { "label": "Introducing GPT-5.5", "url": "https://openai.com/index/introducing-gpt-5-5/", "type": "primary" }, { "label": "GPT-5.5 API model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.5.md", "type": "primary" }, { "label": "GPT-5.5 System Card", "url": "https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf", "type": "primary" }, { "label": "GPT-5.6 System Card", "url": "https://deploymentsafety.openai.com/gpt-5-6/gpt-5-6.pdf", "type": "primary" }, { "label": "Self-replicating prompt injections exist", "url": "https://alignment.openai.com/misalignment-reports/self-replicating-prompt-injections-exist", "type": "primary" }, { "label": "Public Summary of Training Content for GPT-5.5", "url": "https://cdn.openai.com/pdf/gpt-5-5-eu-ai-act-public-summary-of-training-content.pdf", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201.md", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "OpenAI API changelog", "url": "https://developers.openai.com/api/docs/changelog.md", "type": "primary" }, { "label": "Model Release Notes", "url": "https://help.openai.com/en/articles/9624314-model-release-notes", "type": "primary" }, { "label": "OpenAI API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "OpenAI model inventory", "url": "https://developers.openai.com/api/docs/models/all.md", "type": "primary" }, { "label": "GPT-6 Astra System Card", "url": "https://deploymentsafety.openai.com/gpt-6-astra/gpt-6-astra.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5.5.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released April 23, 2026 in ChatGPT and Codex; API release followed April 24, with default snapshot gpt-5.5-2026-04-23. API context is 1,050,000 tokens with a 128,000-token output limit; Codex launch context was 400K. PDF denotes API extraction into text and page images. Transformer is a qualified broad inference from exact prompt-caching support and OpenAI pre-GPT-5.6 KV-cache documentation; full topology and total/active parameter counts remain undisclosed. The training summary identifies no prior-model fine-tune lineage and reports data quantities rather than parameter counts; behavioral size estimates are not specifications. October 14 retirement from ChatGPT, Work and Codex explicitly does not affect API access. No exact API shutdown notice was found; live endpoint operation was not tested.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T17:01:01.410963Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/434.json" }, { "id": "openai/gpt-5.5-pro", "slug": "openai--gpt-5.5-pro", "name": "GPT-5.5 Pro", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5.5", "description": "GPT-5.5 reasoning model using parallel test-time compute, with text and image input and tool calling.", "release_date": "2026-04-23", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 1050000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-gpt-5-5/", "source_type": "primary", "sources": [ { "label": "Introducing GPT-5.5", "url": "https://openai.com/index/introducing-gpt-5-5/", "type": "primary" }, { "label": "GPT-5.5 Pro API model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.5-pro.md", "type": "primary" }, { "label": "GPT-5.5 System Card", "url": "https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf", "type": "primary" }, { "label": "GPT-5.6 System Card", "url": "https://deploymentsafety.openai.com/gpt-5-6/gpt-5-6.pdf", "type": "primary" }, { "label": "GeneBench: Assessing AI Agents for Multi-Stage Inference Problems in Genomics and Quantitative Biology", "url": "https://cdn.openai.com/pdf/6dc7175d-d9e7-4b8d-96b8-48fe5798cd5b/oai_genebench_benchmark.pdf", "type": "primary" }, { "label": "Public Summary of Training Content for GPT-5.5", "url": "https://cdn.openai.com/pdf/gpt-5-5-eu-ai-act-public-summary-of-training-content.pdf", "type": "primary" }, { "label": "Prompt caching", "url": "https://developers.openai.com/api/docs/guides/prompt-caching.md", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201.md", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "OpenAI API changelog", "url": "https://developers.openai.com/api/docs/changelog.md", "type": "primary" }, { "label": "OpenAI API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "Model Release Notes", "url": "https://help.openai.com/en/articles/9624314-model-release-notes", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "OpenAI model inventory", "url": "https://developers.openai.com/api/docs/models/all.md", "type": "primary" }, { "label": "GPT-5.5 API model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.5.md", "type": "primary" }, { "label": "GPT-6 Astra System Card", "url": "https://deploymentsafety.openai.com/gpt-6-astra/gpt-6-astra.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5.5-pro.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released in ChatGPT April 23, 2026; API availability followed April 24. Default API snapshot is gpt-5.5-pro-2026-04-23. Context is 1,050,000 tokens with 128,000 maximum output; reasoning supports medium, high(default) and xhigh. PDF denotes API extraction into text and page images. The system card identifies Pro as the same underlying GPT-5.5 model with parallel test-time compute. Transformer is a qualified broad inference from exact Pro prompt-caching documentation and the pre-GPT-5.6 KV mechanism, corroborated by that model relationship; detailed topology and total/active counts remain undisclosed. The same-model relationship supplies no size because GPT-5.5 has no supported count; additional test-time compute and later benchmark scores do not establish parameters. The GPT-5.5 October 14 product-retirement notice explicitly excludes API access and does not separately enumerate Pro. No exact Pro API shutdown notice was found; endpoint operation was not tested.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T17:02:45.305597Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/436.json" }, { "id": "tencent/hy3-preview", "slug": "tencent--hy3-preview", "name": "Hy3 preview", "developer": "Tencent", "developer_id": "tencent", "family": "Hy", "description": "Tencent's text-only MoE language model with optional reasoning, tool calling, and coding and agent capabilities.", "release_date": "2026-04-23", "date_precision": "day", "open_weights": true, "license": "Tencent Hy Community License Agreement", "parameters_b": 298.786140416, "active_parameters_b": 21, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/tencent/Hy3-preview/blob/549c2b3a0fd5b9a6c6059a9935bf0d59ab69d75a/README.md", "source_type": "primary", "sources": [ { "label": "Hy3 preview official model card (pinned)", "url": "https://huggingface.co/tencent/Hy3-preview/blob/549c2b3a0fd5b9a6c6059a9935bf0d59ab69d75a/README.md", "type": "primary" }, { "label": "Hy3 preview checkpoint configuration (pinned)", "url": "https://huggingface.co/tencent/Hy3-preview/blob/549c2b3a0fd5b9a6c6059a9935bf0d59ab69d75a/config.json", "type": "primary" }, { "label": "Tencent Hy Community License Agreement for Hy3 preview (pinned)", "url": "https://huggingface.co/tencent/Hy3-preview/blob/549c2b3a0fd5b9a6c6059a9935bf0d59ab69d75a/LICENSE", "type": "primary" }, { "label": "Tencent Cloud TokenHub product updates", "url": "https://cloud.tencent.com/document/product/1823/130675", "type": "primary" }, { "label": "Hy3 preview complete safetensors index (pinned)", "url": "https://huggingface.co/tencent/Hy3-preview/blob/549c2b3a0fd5b9a6c6059a9935bf0d59ab69d75a/model.safetensors.index.json", "type": "primary" }, { "label": "Hy3 preview checkpoint shard headers (all 112 indexed shards, pinned)", "url": "https://huggingface.co/tencent/Hy3-preview/tree/549c2b3a0fd5b9a6c6059a9935bf0d59ab69d75a", "type": "primary" }, { "label": "HYV3 Transformer implementation at the checkpoint-specified version", "url": "https://github.com/huggingface/transformers/blob/v5.6.0/src/transformers/models/hy_v3/modeling_hy_v3.py", "type": "primary" }, { "label": "Tencent Hy3 preview training compatibility patches (pinned)", "url": "https://huggingface.co/tencent/Hy3-preview/blob/549c2b3a0fd5b9a6c6059a9935bf0d59ab69d75a/train/llama_factory_support/hy_v3_patches.py", "type": "primary" }, { "label": "HYV3 MTP implementation and weight ownership (pinned)", "url": "https://github.com/vllm-project/vllm/blob/d881812500b49bc77c8e1106664c75ffee16958e/vllm/model_executor/models/hy_v3_mtp.py", "type": "primary" }, { "label": "Hy3 preview chat template (pinned)", "url": "https://huggingface.co/tencent/Hy3-preview/blob/549c2b3a0fd5b9a6c6059a9935bf0d59ab69d75a/chat_template.jinja", "type": "primary" }, { "label": "Hy Token Plan routes Hy3 preview requests to Hy3 from July 24", "url": "https://cloud.tencent.com/announce/detail/2384", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/tencent/hy3-preview.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/tencent/Hy3-preview", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete checkpoint total 298.786140416B includes 295.033528320B backbone and 3.752612096B MTP weights; Tencent rounds those to 295B and 3.8B. Excludes 15360 routing-buffer entries. Active 21B is Tencent's nominal backbone figure; MTP active parameters are undisclosed. Configured context is 262144 tokens. The April 23, 2026 preview uses the Tencent Hy Community License Agreement. Hy Token Plan has routed hy3-preview calls to the separate Hy3 since July 24; these findings describe the original released preview checkpoint.", "active_parameters_scope": "language_backbone", "parameter_components_b": { "language_backbone": 295.03352832, "mtp": 3.752612096 }, "active_parameters_by_phase_b": { "language_backbone": 21 }, "field_status": {}, "audited_at": "2026-10-27T04:56:47.067065Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/504.json" }, { "id": "alibaba/qwen3.6-27b", "slug": "alibaba--qwen3.6-27b", "name": "Qwen3.6 27B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Qwen vision-language model for visual reasoning, documents, and agent tasks", "release_date": "2026-04-22", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 27.781427952, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/Qwen/Qwen3.6-27B", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3.6-27B", "type": "weights" }, { "label": "Official checkpoint configuration", "url": "https://huggingface.co/Qwen/Qwen3.6-27B/raw/main/config.json", "type": "primary" }, { "label": "Official checkpoint tensor index and shard headers", "url": "https://huggingface.co/Qwen/Qwen3.6-27B/raw/main/model.safetensors.index.json", "type": "primary" }, { "label": "Official checkpoint metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen3.6-27B", "type": "primary" }, { "label": "Apache 2.0 checkpoint license", "url": "https://huggingface.co/Qwen/Qwen3.6-27B/raw/main/LICENSE", "type": "primary" }, { "label": "Official April 22 release announcement in Qwen news", "url": "https://github.com/QwenLM/Qwen3.8", "type": "primary" }, { "label": "Later official Qwen3.8 comparison tables", "url": "https://huggingface.co/Qwen/Qwen3.8-27B", "type": "primary" }, { "label": "Alibaba Cloud Qwen3.6-27B launch specification", "url": "https://www.alibabacloud.com/blog/603063", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.6-27b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The full BF16 checkpoint contains 27.781427952B parameters: 26.895998464B text decoder including embeddings and LM head, 0.460730096B vision, and 0.424699392B MTP. Qwen advertises a nominal 27B language model. Active parameters are not separately disclosed. Dense FFNs coexist with hybrid linear/full attention.", "parameter_components_b": { "text_decoder": 26.895998464, "vision": 0.460730096, "mtp": 0.424699392 }, "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T09:47:39.638691Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/058.json" }, { "id": "xiaomi/mimo-v2.5", "slug": "xiaomi--mimo-v2.5", "name": "MiMo-V2.5", "developer": "Xiaomi", "developer_id": "xiaomi", "family": "mimo", "description": "Open-weight multimodal MoE model with text, image, audio and video input, text output, reasoning and tool calling.", "release_date": "2026-04-22", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 310, "active_parameters_b": 15, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "audio", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://mimo.xiaomi.com/mimo-v2-5", "source_type": "primary", "sources": [ { "label": "MiMo-V2.5 launch announcement", "url": "https://mimo.xiaomi.com/mimo-v2-5", "type": "primary" }, { "label": "MiMo-V2.5 official checkpoint card at pinned revision", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/blob/63651580ca774f8504f676040460aed3e1244ac1/README.md", "type": "weights" }, { "label": "MiMo-V2.5 checkpoint configuration", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/blob/63651580ca774f8504f676040460aed3e1244ac1/config.json", "type": "weights" }, { "label": "MiMo-V2.5 chat template", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/blob/63651580ca774f8504f676040460aed3e1244ac1/tokenizer_config.json", "type": "weights" }, { "label": "MiMo-V2.5 official model implementation", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5/blob/63651580ca774f8504f676040460aed3e1244ac1/modeling_mimo_v2.py", "type": "weights" }, { "label": "MiMo-V2.5 series open-weight release announcement", "url": "https://mimo.mi.com/docs/en-US/news/latest/v2.5-open-sourced", "type": "primary" }, { "label": "XiaomiMiMo open-source announcement via X syndication", "url": "https://cdn.syndication.twimg.com/tweet-result?id=2048821516079661561&lang=en&token=1", "type": "primary" }, { "label": "Xiaomi model release changelog", "url": "https://mimo.mi.com/docs/en-US/updates/model", "type": "primary" }, { "label": "Official MiMo-V2.5 checkpoint repository metadata", "url": "https://huggingface.co/api/models/XiaomiMiMo/MiMo-V2.5", "type": "weights" }, { "label": "Full-Pipeline Inference Optimization for MiMo-V2.5 Series", "url": "https://arxiv.org/html/2607.13095v1", "type": "primary" }, { "label": "Xiaomi model deprecation notice", "url": "https://mimo.mi.com/docs/en-US/updates/deprecate", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/xiaomi/mimo-v2.5.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Xiaomi reports nominal 310B/15B counts in its language-backbone table; the 729M vision encoder, 261M audio encoder, projectors and MTP modules are not reconciled into an exact complete-system count. Original announcement: April 22, 2026; API public-beta changelog: April 23; MIT weights announced April 27 UTC (April 28 Beijing). The checkpoint supports 1,048,576 positions; serving-example and default output limits are separate. Xiaomi documents native API identifier deprecation on October 21, 2026; open weights remain available.", "parameters_scope": "language_backbone", "parameter_components_b": { "language_backbone": 310, "vision_encoder": 0.729, "audio_encoder": 0.261 }, "active_parameters_scope": "language_decode", "active_parameters_by_phase_b": { "language_decode": 15 }, "field_status": {}, "audited_at": "2026-10-27T05:48:33.773150Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/540.json" }, { "id": "xiaomi/mimo-v2.5-pro", "slug": "xiaomi--mimo-v2.5-pro", "name": "MiMo-V2.5-Pro", "developer": "Xiaomi", "developer_id": "xiaomi", "family": "mimo", "description": "Text reasoning model with a mixture-of-experts architecture, tool calling and a 1M-token context window.", "release_date": "2026-04-22", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 1020, "active_parameters_b": 42, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "source_type": "primary", "sources": [ { "label": "MiMo-V2.5-Pro release and open-weight announcement", "url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "type": "primary" }, { "label": "MiMo-V2.5-Pro official model card at pinned revision", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/blob/21d1ecfecd7bd70f31be25ca49d7edd21f003659/README.md", "type": "weights" }, { "label": "MiMo-V2.5-Pro checkpoint configuration", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/blob/21d1ecfecd7bd70f31be25ca49d7edd21f003659/config.json", "type": "weights" }, { "label": "MiMo-V2.5-Pro chat template", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/blob/21d1ecfecd7bd70f31be25ca49d7edd21f003659/tokenizer_config.json", "type": "weights" }, { "label": "MiMo-V2.5-Pro official model implementation", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/blob/21d1ecfecd7bd70f31be25ca49d7edd21f003659/modeling_mimo_v2.py", "type": "weights" }, { "label": "Official MiMo-V2.5-Pro repository metadata", "url": "https://huggingface.co/api/models/XiaomiMiMo/MiMo-V2.5-Pro", "type": "weights" }, { "label": "MiMo-V2.5-Pro checkpoint tensor index", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro/blob/21d1ecfecd7bd70f31be25ca49d7edd21f003659/model.safetensors.index.json", "type": "weights" }, { "label": "MiMo-V2.5 series open-weight license announcement", "url": "https://mimo.mi.com/docs/en-US/news/latest/v2.5-open-sourced", "type": "primary" }, { "label": "XiaomiMiMo original V2.5 series announcement via X syndication", "url": "https://cdn.syndication.twimg.com/tweet-result?id=2046988157888209365&lang=en&token=1", "type": "primary" }, { "label": "Xiaomi model release changelog", "url": "https://mimo.mi.com/docs/en-US/updates/model", "type": "primary" }, { "label": "Xiaomi model deprecation notice", "url": "https://mimo.mi.com/docs/en-US/updates/deprecate", "type": "primary" }, { "label": "Full-Pipeline Inference Optimization for MiMo-V2.5 Series", "url": "https://arxiv.org/html/2607.13095v1", "type": "primary" }, { "label": "XiaomiMiMo open-weight announcement via X syndication", "url": "https://cdn.syndication.twimg.com/tweet-result?id=2048821516079661561&lang=en&token=1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/xiaomi/mimo-v2.5-pro.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Xiaomi reports rounded nominal totals of 1.02T parameters and 42B active; these are not exact tensor counts or a reconciled MTP breakdown. Original announcement: April 22, 2026 UTC, April 23 Beijing. MIT weights followed April 27 UTC. This checkpoint accepts and generates text; later multimodal API behavior belongs to V2.6-Pro. Xiaomi documents redirection on October 14 and API identifier expiration on October 21, 2026; public weights remain available.", "field_status": {}, "audited_at": "2026-10-27T05:53:02.664577Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/541.json" }, { "id": "google/deep-research-max-preview-04-2026", "slug": "google--deep-research-max-preview-04-2026", "name": "Deep Research Max Preview", "developer": "Google", "developer_id": "google", "family": "gemini-pro", "description": "Managed research agent built with Gemini 3.1 Pro that plans multi-step investigations and produces cited reports with optional visualizations.", "release_date": "2026-04-21", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1048576, "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text", "image" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ai.google.dev/gemini-api/docs/models/deep-research-max-preview-04-2026", "source_type": "primary", "sources": [ { "label": "Deep Research Max preview", "url": "https://ai.google.dev/gemini-api/docs/models/deep-research-max-preview-04-2026", "type": "primary" }, { "label": "Deep Research Max: a step change for autonomous research agents", "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/next-Generation-gemini-deep-research/", "type": "primary" }, { "label": "Gemini Deep Research agent", "url": "https://ai.google.dev/gemini-api/docs/deep-research", "type": "primary" }, { "label": "Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/deep-research-max-preview-04-2026.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Built with Gemini 3.1 Pro; Google has not disclosed total or active parameters for that backbone or this agent. Tool support includes remote MCP, but excludes custom function-calling tools.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "not_applicable" }, "audited_at": "2026-10-27T12:43:31.117214Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/187.json" }, { "id": "google/deep-research-preview-04-2026", "slug": "google--deep-research-preview-04-2026", "name": "Gemini Deep Research Preview", "developer": "Google", "developer_id": "google", "family": "gemini-pro", "description": "Managed research agent for autonomous multi-step investigations, synthesis, and cited reports.", "release_date": "2026-04-21", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1048576, "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text", "image" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ai.google.dev/gemini-api/docs/models/deep-research-preview-04-2026", "source_type": "primary", "sources": [ { "label": "Deep Research preview: exact agent specifications", "url": "https://ai.google.dev/gemini-api/docs/models/deep-research-preview-04-2026", "type": "primary" }, { "label": "Gemini API release notes", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "primary" }, { "label": "Deep Research Max: a step change for autonomous research agents", "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/next-generation-gemini-deep-research/", "type": "primary" }, { "label": "Gemini Deep Research agent guide", "url": "https://ai.google.dev/gemini-api/docs/deep-research", "type": "primary" }, { "label": "Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Gemini 3.1 Pro model card", "url": "https://deepmind.google/models/model-cards/gemini-3-1-pro/", "type": "primary" }, { "label": "Gemini 3 Pro model card, updated May 2026", "url": "https://deepmind.google/models/model-cards/gemini-3-pro/", "type": "primary" }, { "label": "Gemini 4 Argon model card, October 2026", "url": "https://deepmind.google/models/model-cards/gemini-4-argon/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/deep-research-preview-04-2026.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Built with Gemini 3.1 Pro, whose model card describes a sparse MoE Transformer. Google does not disclose total or active parameters for that backbone. The managed agent supports built-in tools and remote MCP servers, but excludes custom function-calling tools.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "not_applicable" }, "audited_at": "2026-10-27T12:43:17.845583Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/188.json" }, { "id": "moonshotai/kimi-k2.6", "slug": "moonshotai--kimi-k2.6", "name": "Kimi K2.6", "developer": "Moonshot AI", "developer_id": "moonshotai", "family": "kimi-k2", "description": "Open-weight multimodal MoE model with thinking and instant modes, image/video input, and multi-step tool calling.", "release_date": "2026-04-20", "date_precision": "day", "open_weights": true, "license": "Modified MIT", "parameters_b": 1026.879376368, "active_parameters_b": 32, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md", "source_type": "primary", "sources": [ { "label": "Kimi K2.6 official model card, pinned revision", "url": "https://huggingface.co/moonshotai/Kimi-K2.6/blob/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/README.md", "type": "primary" }, { "label": "Kimi K2.6 checkpoint configuration, pinned revision", "url": "https://huggingface.co/moonshotai/Kimi-K2.6/raw/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/config.json", "type": "weights" }, { "label": "Kimi K2.6 Modified MIT License, pinned revision", "url": "https://huggingface.co/moonshotai/Kimi-K2.6/raw/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/LICENSE", "type": "primary" }, { "label": "Introducing Kimi Agent: model release history", "url": "https://www.kimi.com/en/help/agent/agent-overview", "type": "primary" }, { "label": "Kimi K2.6: Advancing Open-Source Coding", "url": "https://www.kimi.com/en/blog/kimi-k2-6", "type": "primary" }, { "label": "Kimi K2.6 checkpoint tensor index, pinned revision", "url": "https://huggingface.co/moonshotai/Kimi-K2.6/resolve/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/model.safetensors.index.json", "type": "weights" }, { "label": "Kimi K2.6 language-model implementation, pinned revision", "url": "https://huggingface.co/moonshotai/Kimi-K2.6/raw/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/modeling_deepseek.py", "type": "primary" }, { "label": "Kimi K2.6 vision and projector implementation, pinned revision", "url": "https://huggingface.co/moonshotai/Kimi-K2.6/raw/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/modeling_kimi_k25.py", "type": "primary" }, { "label": "Kimi K2.6 checkpoint headers, all 64 indexed shards at pinned revision", "url": "https://huggingface.co/moonshotai/Kimi-K2.6/resolve/7eb5002f6aadc958aed6a9177b7ed26bb94011bb/model-00001-of-000064.safetensors", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/moonshotai/kimi-k2.6.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/moonshotai/Kimi-K2.6", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Exact complete count: 1026.408232448B language model + 0.416866032B MoonViT encoder + 0.054277888B projector = 1026.879376368B. All 64 pinned checkpoint headers reconcile with independent config/code arithmetic. Packed INT4 matrices are counted logically; scales and shape metadata are excluded. Router correction biases are declared parameters and included. The card’s 1T total, 400M vision encoder and 32B active figures are nominal; active remains the developer-reported figure, without an exact vision-inclusive claim. Video input is documented as experimental and official-API-only at this revision.", "parameter_components_b": { "language_backbone": 1026.408232448, "vision_encoder": 0.416866032, "vision_projector": 0.054277888 }, "field_status": {}, "audited_at": "2026-10-26T23:58:42.622Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/360.json" }, { "id": "xai/grok-4.3", "slug": "xai--grok-4.3", "name": "Grok 4.3", "developer": "xAI", "developer_id": "xai", "family": "grok", "description": "Grok model with configurable reasoning, text and image input, tool calling, and a one-million-token context window.", "release_date": "2026-04-17", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 500, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": null, "context_window": 1000000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://x.com/elonmusk/status/2045285099680252417", "source_type": "primary", "sources": [ { "label": "Grok 4.3 beta availability (Elon Musk; retrieved via X syndication)", "url": "https://x.com/elonmusk/status/2045285099680252417", "type": "primary" }, { "label": "Grok 4.3 native API documentation", "url": "https://docs.x.ai/developers/models/grok-4.3", "type": "primary" }, { "label": "Beta size clarification (Elon Musk; retrieved via X syndication)", "url": "https://x.com/elonmusk/status/2045293644693770630", "type": "primary" }, { "label": "API-launch size statement (Aditya Gupta; retrieved via X syndication)", "url": "https://x.com/ag_i_2211/status/2050089293713129772", "type": "primary" }, { "label": "Chat with Files", "url": "https://docs.x.ai/developers/model-capabilities/files/chat-with-files", "type": "primary" }, { "label": "Enterprise Terms, April 10, 2026", "url": "https://x.ai/legal/terms-of-service-enterprise/previous-2026-04-10", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/xai/grok-4.3.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "April 17, 2026 is the public beta date. Approximately 500B total parameters is the rounded developer-reported size, confirmed by Musk for the beta checkpoint and xAI researcher Aditya Gupta at the API launch. Active parameters and neural-network architecture remain undisclosed in the sources reviewed. Native xAI context is 1M; Google Cloud and Microsoft Foundry document 200k limits. PDF attachments use document-search tools. May 15 redirects from older model slugs do not change this model's release date.", "field_status": { "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T18:30:00.755572+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/531.json" }, { "id": "alibaba/qwen3.6-35b-a3b", "slug": "alibaba--qwen3.6-35b-a3b", "name": "Qwen3.6 35B-A3B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Open-weight multimodal Qwen MoE supporting text, image and video input, text output, reasoning and tool calling.", "release_date": "2026-04-16", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 35.951822704, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B", "type": "weights" }, { "label": "Official Qwen release news", "url": "https://github.com/QwenLM/Qwen3.8#news", "type": "primary" }, { "label": "Official model license", "url": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B/raw/main/LICENSE", "type": "primary" }, { "label": "Official checkpoint metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen3.6-35B-A3B", "type": "weights" }, { "label": "Official checkpoint tensor index", "url": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B/resolve/main/model.safetensors.index.json", "type": "weights" }, { "label": "Alibaba Cloud article (April 17)", "url": "https://www.alibabacloud.com/blog/qwen3-6-35b-a3b-thinking-open-weights-uncompromised-agentic-coding_603043", "type": "primary" }, { "label": "Official checkpoint history", "url": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B/commits/main", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.6-35b-a3b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Full checkpoint: 35.951822704B stored parameters including the vision encoder and auxiliary MTP. The model card reports nominal 35B total and 3B activated for the language model; 3B is not a full multimodal/MTP active count. Official weight availability is dated April 16, 2026; the Alibaba Cloud article is April 17. The checkpoint has an April 15 commit labelled Initial public release, which does not independently establish its historical visibility.", "active_parameters_scope": "language_decoder", "parameter_components_b": { "language_decoder": 34.660610688, "vision": 0.446571248, "mtp": 0.844640768 }, "field_status": {}, "audited_at": "2026-10-26T15:13:10.117625Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/059.json" }, { "id": "alibaba/qwen3.6-flash", "slug": "alibaba--qwen3.6-flash", "name": "Qwen3.6 Flash", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Alibaba's Qwen3.6 vision-language model for agentic coding and visual reasoning, available through Model Studio as the hosted counterpart of Qwen3.6-35B-A3B.", "release_date": "2026-04-16", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 35, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://help.aliyun.com/en/model-studio/qwen3-6-flash", "source_type": "primary", "sources": [ { "label": "Official API model information", "url": "https://help.aliyun.com/en/model-studio/qwen3-6-flash", "type": "primary" }, { "label": "Official regional model release dates", "url": "https://help.aliyun.com/zh/model-studio/newly-released-models", "type": "primary" }, { "label": "Official API-to-open-model mapping", "url": "https://www.alibabacloud.com/blog/603042", "type": "primary" }, { "label": "Official mapped model card", "url": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B", "type": "primary" }, { "label": "Official current snapshot mapping", "url": "https://help.aliyun.com/en/model-studio/model-pricing", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.6-flash.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Alibaba explicitly maps qwen3.6-flash to the released Qwen3.6-35B-A3B model; current pricing documentation maps the alias to qwen3.6-flash-2026-04-16. The nominal 35B total and 3B active counts refer only to its language decoder, not a measured total for the hosted multimodal deployment. Apache-2.0 describes the released weights; hosted service terms are separate. Beijing release was April 16, 2026, followed by international/global regions April 17. API operation was not tested.", "parameters_scope": "language_decoder", "parameter_components_b": { "language_decoder": 35 }, "active_parameters_scope": "language_decoder", "field_status": {}, "audited_at": "2026-10-26T15:07:22Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/060.json" }, { "id": "anthropic/claude-opus-4-7", "slug": "anthropic--claude-opus-4-7", "name": "Claude Opus 4.7", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude-opus", "description": "Anthropic Opus model for software engineering and agentic work, with adaptive reasoning, vision input, and a 1M-token context window.", "release_date": "2026-04-16", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1000000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.anthropic.com/news/claude-opus-4-7", "source_type": "primary", "sources": [ { "label": "Introducing Claude Opus 4.7", "url": "https://www.anthropic.com/news/claude-opus-4-7", "type": "primary" }, { "label": "Claude Opus 4.7 System Card", "url": "https://www-cdn.anthropic.com/037f06850df7fbe871e206dad004c3db5fd50340/Claude%20Opus%204.7%20System%20Card.pdf", "type": "primary" }, { "label": "Claude Opus 4.7 model overview", "url": "https://platform.claude.com/docs/en/models/opus-4-7/overview", "type": "primary" }, { "label": "PDF support", "url": "https://platform.claude.com/docs/en/build-with-claude/pdf-support", "type": "primary" }, { "label": "GLM-5.3 and the spread of advanced cyber capabilities", "url": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities", "type": "primary" }, { "label": "Commercial Terms of Service", "url": "https://www.anthropic.com/legal/commercial-terms", "type": "primary" }, { "label": "Claude Opus 4.8 System Card (June 3 revision)", "url": "https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea.pdf", "type": "primary" }, { "label": "Claude Opus 5.5 System Card", "url": "https://www.anthropic.com/claude-opus-5-5-system-card", "type": "primary" }, { "label": "Claude Opus 5.5 migration guide", "url": "https://platform.claude.com/docs/en/models/opus-5-5/migration-guide", "type": "primary" }, { "label": "Anthropic Transparency Hub: Claude Opus 4.7 Summary Table", "url": "https://www.anthropic.com/transparency", "type": "primary" }, { "label": "When AI builds itself", "url": "https://www.anthropic.com/institute/recursive-self-improvement", "type": "primary" }, { "label": "Claude Opus 5 System Card", "url": "https://anthropic.com/claude-opus-5-system-card", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/anthropic/claude-opus-4-7.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Nominal total and active parameter counts remain undisclosed in the reviewed Anthropic specifications and original/later system cards. No sized base-model lineage is documented. Released production architecture remains undisclosed; early-checkpoint residual-stream experiments do not establish it. PDF is an API-supported document input; output is text.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T10:39:23.485500+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/108.json" }, { "id": "inference-net/schematron-v2-small", "slug": "inference-net--schematron-v2-small", "name": "Schematron V2 Small", "developer": "Inference.net", "developer_id": "inference-net", "family": "schematron", "description": "Schema-guided HTML-to-JSON extraction model for complex schemas and long web pages.", "release_date": "2026-04-16", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 3, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": null, "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": false, "source_url": "https://inference.net/models/schematron-v2-small/", "source_type": "primary", "sources": [ { "label": "Official model page", "url": "https://inference.net/models/schematron-v2-small/", "type": "primary" }, { "label": "Official V2 launch (2026-04-16)", "url": "https://inference.net/blog/schematron-v2/", "type": "primary" }, { "label": "Official Schematron documentation", "url": "https://docs.inference.net/workhorse-models/schematron", "type": "primary" }, { "label": "Live Inference.net model API", "url": "https://api.inference.net/v1/models", "type": "primary" }, { "label": "Official tool support documentation", "url": "https://docs.inference.net/api/coding-agents", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/inference-net/schematron-v2-small.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Inference.net reports a nominal 3B total. No separate active count or architecture was found in the checked primary sources. Legacy V1 checkpoints and replacement API routing do not establish an unchanged V2 base.", "field_status": { "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T13:42:35.812182+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/247.json" }, { "id": "inference-net/schematron-v2-turbo", "slug": "inference-net--schematron-v2-turbo", "name": "Schematron V2 Turbo", "developer": "Inference.net", "developer_id": "inference-net", "family": "schematron", "description": "A 3B HTML-to-JSON extraction model in the Schematron family, optimized for throughput and guided by a supplied JSON Schema.", "release_date": "2026-04-16", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 3, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": null, "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": false, "source_url": "https://inference.net/blog/schematron-v2/", "source_type": "primary", "sources": [ { "label": "Schematron V2 launch announcement", "url": "https://inference.net/blog/schematron-v2/", "type": "primary" }, { "label": "Schematron V2 Turbo official model page", "url": "https://inference.net/models/schematron-v2-turbo/", "type": "primary" }, { "label": "Schematron API documentation", "url": "https://docs.inference.net/workhorse-models/schematron", "type": "primary" }, { "label": "Inference.net service terms", "url": "https://inference.net/terms-of-service/", "type": "primary" }, { "label": "Inference.net reasoning documentation", "url": "https://docs.inference.net/api/reasoning", "type": "primary" }, { "label": "Inference.net coding-agent compatibility", "url": "https://docs.inference.net/api/coding-agents", "type": "primary" }, { "label": "Inference.net public model API", "url": "https://api.inference.net/v1/models", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/inference-net/schematron-v2-turbo.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The developer reports a nominal 3B total. No separate active-parameter count or verified base-model lineage was found in the checked primary sources. V2 Turbo is API-only and distinct from open V1 checkpoints. Official API and product state confirm 128000 context, text input/output, and no reasoning or tool-calling capability.", "field_status": { "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T13:43:55.117961+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/248.json" }, { "id": "openai/gpt-rosalind", "slug": "openai--gpt-rosalind", "name": "GPT-Rosalind", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-rosalind", "description": null, "release_date": "2026-04-16", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-gpt-rosalind/", "source_type": "primary", "sources": [ { "label": "Initial model release", "url": "https://openai.com/index/introducing-gpt-rosalind/", "type": "primary" }, { "label": "Current eligibility and model identity", "url": "https://openai.com/rosalind/", "type": "primary" }, { "label": "Academic access announcement", "url": "https://openai.com/index/introducing-the-rosalind-academic-plan/", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:11:51.983Z", "last_seen": "2026-10-30T07:11:51.983Z", "upstream_present": false, "availability": "limited", "notes": "First externally available April16,2026 to qualified US enterprise customers through trusted access. September11 expanded eligible organizations globally; October15 introduced academic access. Approval remains required, so availability is limited. This record describes the original named language model, not the Rosalind Workbench tool package. No separately identified later trained checkpoint was verified. Parameters, context and additional native modalities remain undisclosed in reviewed launch material.", "announcement_date": "2026-04-16" }, { "id": "alibaba/qwen3.6-max-preview", "slug": "alibaba--qwen3.6-max-preview", "name": "Qwen3.6 Max Preview", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Proprietary text-only preview in the Qwen3.6 series, with thinking mode and function calling for coding and agentic tasks.", "release_date": "2026-04-14", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://help.aliyun.com/zh/model-studio/newly-released-models", "source_type": "primary", "sources": [ { "label": "Alibaba release history", "url": "https://help.aliyun.com/zh/model-studio/newly-released-models", "type": "primary" }, { "label": "Qwen preview announcement", "url": "https://qwen.ai/blog?id=qwen3.6-max-preview", "type": "primary" }, { "label": "Qwen API specifications", "url": "https://qwen.ai/apiplatform", "type": "primary" }, { "label": "Alibaba text-generation capabilities", "url": "https://www.alibabacloud.com/help/en/model-studio/text-generation-model/", "type": "primary" }, { "label": "Alibaba preview model documentation", "url": "https://www.alibabacloud.com/help/zh/model-studio/qwen3-6-max", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.6-max-preview.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Earliest documented release is the International listing on April 14, 2026; the Qwen announcement is April 18 and China listing April 20. The undated preview identifier has no verified immutable checkpoint equivalence. Total/active counts and architecture remain undisclosed by the developer; the secondary approximately 1T/MoE claim lacks primary corroboration. Context and tool support rely partly on indexed primary pages whose live content has changed; current endpoint operation was not tested.", "field_status": { "license": "undisclosed", "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T09:50:00.523746Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/061.json" }, { "id": "google/gemini-robotics-er-1.6-preview", "slug": "google--gemini-robotics-er-1.6-preview", "name": "Gemini Robotics-ER 1.6 Preview", "developer": "Google", "developer_id": "google", "family": "gemini", "description": "Vision-language model based on Gemini 3.0 Flash for spatial reasoning, task planning, success detection, and instrument reading in robotics.", "release_date": "2026-04-14", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://deepmind.google/blog/gemini-robotics-er-1-6/", "source_type": "primary", "sources": [ { "label": "Gemini Robotics-ER 1.6: Powering real-world robotics tasks through enhanced embodied reasoning", "url": "https://deepmind.google/blog/gemini-robotics-er-1-6/", "type": "primary" }, { "label": "Gemini Robotics-ER 1.6 model card", "url": "https://deepmind.google/models/model-cards/gemini-robotics-er-1-6/", "type": "primary" }, { "label": "Gemini Robotics ER 1.6 API model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-robotics-er-1.6-preview?hl=en", "type": "primary" }, { "label": "Gemini 3 Flash model card", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-Flash-Model-Card.pdf", "type": "primary" }, { "label": "Gemini 3 Pro model card, updated May 2026", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-Pro-Model-Card.pdf", "type": "primary" }, { "label": "Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Gemini deprecations", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "type": "primary" }, { "label": "Gemini Robotics ER overview", "url": "https://ai.google.dev/gemini-api/docs/robotics-overview", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemini-robotics-er-1.6-preview.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Sparse MoE Transformer based on Gemini 3 Flash, whose card names Gemini 3 Pro as its base. Total and active parameter counts are undisclosed for ER 1.6 and its named bases. These are historical preview specifications: Google marks the endpoint as shut down and lists August 31, 2026 as its shutdown date.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T13:09:26.148088+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/214.json" }, { "id": "zhipuai/glm-5.1", "slug": "zhipuai--glm-5.1", "name": "GLM-5.1", "developer": "Z.ai", "developer_id": "zhipuai", "family": "glm", "description": "Text MoE model with sparse attention, reasoning and tool calling for coding and long-running tasks.", "release_date": "2026-04-07", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 744, "active_parameters_b": 40, "parameters_source": "primary_source", "architecture": "moe", "context_window": 202752, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://z.ai/blog/glm-5.1", "source_type": "primary", "sources": [ { "label": "GLM-5.1: Towards Long-Horizon Tasks", "url": "https://z.ai/blog/glm-5.1", "type": "primary" }, { "label": "GLM-5.1 pinned official model card", "url": "https://huggingface.co/zai-org/GLM-5.1/blob/26e1bd6e011feb778d25ae34b09b07074139d92d/README.md", "type": "primary" }, { "label": "Official GLM model download table at pinned revision", "url": "https://github.com/zai-org/GLM-5/blob/83c03d317378f86f0a012f7c4cd48b6243c7e69b/README.md", "type": "primary" }, { "label": "GLM-5.1 checkpoint configuration", "url": "https://huggingface.co/zai-org/GLM-5.1/blob/26e1bd6e011feb778d25ae34b09b07074139d92d/config.json", "type": "primary" }, { "label": "GLM-5.1 checkpoint MIT license", "url": "https://huggingface.co/zai-org/GLM-5.1/blob/26e1bd6e011feb778d25ae34b09b07074139d92d/LICENSE", "type": "primary" }, { "label": "GLM-5.1 native API guide", "url": "https://docs.z.ai/guides/llm/glm-5.1", "type": "primary" }, { "label": "Z.ai model release notes", "url": "https://docs.z.ai/release-notes/new-released", "type": "primary" }, { "label": "GLM-5.1 exact checkpoint chat template", "url": "https://huggingface.co/zai-org/GLM-5.1/blob/26e1bd6e011feb778d25ae34b09b07074139d92d/chat_template.jinja", "type": "primary" }, { "label": "GLM MoE DSA implementation at checkpoint-declared version", "url": "https://github.com/huggingface/transformers/blob/v5.4.0/src/transformers/models/glm_moe_dsa/modeling_glm_moe_dsa.py", "type": "primary" }, { "label": "GLM-5.1 exact tokenizer configuration", "url": "https://huggingface.co/zai-org/GLM-5.1/blob/26e1bd6e011feb778d25ae34b09b07074139d92d/tokenizer_config.json", "type": "primary" }, { "label": "GLM-5 technical report, checked for applicability", "url": "https://arxiv.org/html/2602.15763v1", "type": "primary" }, { "label": "GLM-5.1 public checkpoint metadata", "url": "https://huggingface.co/api/models/zai-org/GLM-5.1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/zhipuai/glm-5.1.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/GLM-5.1", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Z.ai lists GLM-5.1 as nominal 744B total and 40B active. The exact download row does not define embedding, output-head or MTP accounting, so these are not exact full checkpoint counts and the GLM-5 counting scope is not transferred. The released model uses MoE with MLA and sparse attention, plus one configured MTP layer. Checkpoint and tokenizer context is 202752 tokens, rounded to 200K in native API documentation. Public API and weights release was April 7, 2026; the announcement described chat access as forthcoming.", "field_status": {}, "audited_at": "2026-10-27T06:20:03.622447+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/559.json" }, { "id": "google/gemma-4-26b-a4b-it", "slug": "google--gemma-4-26b-a4b-it", "name": "Gemma 4 26B A4B IT", "developer": "Google", "developer_id": "google", "family": "gemma", "description": "Google’s instruction-tuned Gemma 4 mixture-of-experts model with text, image and video input, text output, configurable reasoning, and native function calling.", "release_date": "2026-04-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 25.8059, "active_parameters_b": 3.8, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ai.google.dev/gemma/docs/core/model_card_4", "source_type": "primary", "sources": [ { "label": "Google Gemma 4 model card", "url": "https://ai.google.dev/gemma/docs/core/model_card_4", "type": "primary" }, { "label": "Google Gemma 4 launch announcement", "url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "type": "primary" }, { "label": "Exact checkpoint configuration", "url": "https://huggingface.co/google/gemma-4-26B-A4B-it/resolve/4d7ae4984b7db7de8f8457170b3f1a419ee76d52/config.json", "type": "weights" }, { "label": "Official checkpoint index", "url": "https://huggingface.co/google/gemma-4-26B-A4B-it/resolve/4d7ae4984b7db7de8f8457170b3f1a419ee76d52/model.safetensors.index.json", "type": "weights" }, { "label": "Official BF16 shard 1 tensor header", "url": "https://huggingface.co/google/gemma-4-26B-A4B-it/resolve/4d7ae4984b7db7de8f8457170b3f1a419ee76d52/model-00001-of-00002.safetensors", "type": "weights" }, { "label": "Official BF16 shard 2 tensor header", "url": "https://huggingface.co/google/gemma-4-26B-A4B-it/resolve/4d7ae4984b7db7de8f8457170b3f1a419ee76d52/model-00002-of-00002.safetensors", "type": "weights" }, { "label": "Gemma 4 implementation: parameters and buffers", "url": "https://github.com/huggingface/transformers/blob/6c5f8f4186c15f4bfbd67941a1736b2b347e0155/src/transformers/models/gemma4/modeling_gemma4.py", "type": "primary" }, { "label": "Gemma 4 Technical Report", "url": "https://arxiv.org/html/2607.02770v1", "type": "primary" }, { "label": "Gemma release log: date conflict", "url": "https://ai.google.dev/gemma/docs/releases", "type": "primary" }, { "label": "Google AI Edge launch corroboration", "url": "https://developers.googleblog.com/bring-state-of-the-art-agentic-skills-to-the-edge-with-gemma-4/", "type": "primary" }, { "label": "Separate MTP model documentation", "url": "https://ai.google.dev/gemma/docs/mtp/mtp", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemma-4-26b-a4b-it.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/google/gemma-4-26B-A4B-it", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "25.8059B is a rounded complete main-checkpoint estimate, including vision/projection and counting tied input/output embeddings once. Exact BF16 storage contains 25,805,936,206 elements; HF parameter-only accounting gives 25,805,933,872 after excluding 2,334 scalar/standardization buffer elements. Google JAX exposes those elements with self.param, so historical learning provenance is not independently established; both conventions round to the published value. The card’s 25.2B is the rounded text component. Active 3.8B is Google’s rounded text-decoding count, excluding vision prefill and the separate MTP assistant. Video is processed as image frames. April 2 is retained from contemporaneous Google/Hugging Face launch evidence despite the current release log’s conflicting March 31 entry.", "field_status": {}, "audited_at": "2026-10-26T20:22:31.763063Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/223.json" }, { "id": "google/gemma-4-31b-it", "slug": "google--gemma-4-31b-it", "name": "Gemma 4 31B IT", "developer": "Google", "developer_id": "google", "family": "gemma", "description": "Instruction-tuned dense Gemma 4 model for text generation from text, images and video, with configurable thinking and function calling.", "release_date": "2026-04-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 31.2731, "active_parameters_b": 31, "parameters_source": "primary_source", "architecture": "dense", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "source_type": "primary", "sources": [ { "label": "Official Gemma 4 31B IT card", "url": "https://huggingface.co/google/gemma-4-31B-it/blob/842da3794eaa0b77d5f08bae87a17459d91ff475/README.md", "type": "weights" }, { "label": "Official configuration", "url": "https://huggingface.co/google/gemma-4-31B-it/resolve/842da3794eaa0b77d5f08bae87a17459d91ff475/config.json", "type": "weights" }, { "label": "Official tensor index", "url": "https://huggingface.co/google/gemma-4-31B-it/resolve/842da3794eaa0b77d5f08bae87a17459d91ff475/model.safetensors.index.json", "type": "weights" }, { "label": "Official checkpoint shard 1 header", "url": "https://huggingface.co/google/gemma-4-31B-it/resolve/842da3794eaa0b77d5f08bae87a17459d91ff475/model-00001-of-00002.safetensors", "type": "weights" }, { "label": "Official checkpoint shard 2 header", "url": "https://huggingface.co/google/gemma-4-31B-it/resolve/842da3794eaa0b77d5f08bae87a17459d91ff475/model-00002-of-00002.safetensors", "type": "weights" }, { "label": "Google Gemma 4 transformer implementation", "url": "https://github.com/google-deepmind/gemma/blob/9e1f74131ba9d45e5fe6ae7bb3941d9f2abb39c3/gemma/gm/nn/gemma4/_modules.py", "type": "primary" }, { "label": "Transformers Gemma 4 implementation", "url": "https://github.com/huggingface/transformers/blob/a672eb5f474b8b30173e9f96eab66adf292da635/src/transformers/models/gemma4/modeling_gemma4.py", "type": "primary" }, { "label": "Gemma 4 checkpoint conversion", "url": "https://github.com/huggingface/transformers/blob/a672eb5f474b8b30173e9f96eab66adf292da635/src/transformers/models/gemma4/convert_gemma4_weights.py", "type": "primary" }, { "label": "Gemma 4 technical report", "url": "https://arxiv.org/html/2607.02770v1", "type": "primary" }, { "label": "Gemma 4 launch announcement", "url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "type": "primary" }, { "label": "Exact checkpoint release history", "url": "https://huggingface.co/api/models/google/gemma-4-31B-it/commits/main", "type": "primary" }, { "label": "Google release chronology (conflicting date)", "url": "https://ai.google.dev/gemma/docs/releases", "type": "primary" }, { "label": "Separate MTP assistant documentation", "url": "https://ai.google.dev/gemma/docs/mtp/mtp", "type": "primary" }, { "label": "Official processor configuration", "url": "https://huggingface.co/google/gemma-4-31B-it/resolve/842da3794eaa0b77d5f08bae87a17459d91ff475/processor_config.json", "type": "weights" }, { "label": "Google vision standardization implementation", "url": "https://github.com/google-deepmind/gemma/blob/9e1f74131ba9d45e5fe6ae7bb3941d9f2abb39c3/gemma/gm/nn/gemma4/vision/_layers.py", "type": "primary" }, { "label": "Alem research paper: nominal active-model-size comparison", "url": "https://arxiv.org/html/2606.08340v1", "type": "primary" }, { "label": "Vercel exact-model inference announcement", "url": "https://vercel.com/changelog/gemma-4-on-ai-gateway", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemma-4-31b-it.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/google/gemma-4-31B-it", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Full unique multimodal target checkpoint is approximately 31.2731B parameters; Google's 30.7B figure covers the text backbone. Active ~31B follows the nominal comparison convention in Tessera et al.'s Alem paper, not a Google measurement of modality-specific participation. Tied output/token embedding counted once; optional separate MTP assistant excluded. Video is processed as image frames; no native audio. Retain April 2, 2026 for the exact IT public launch; current family chronology instead lists March 31.", "field_status": {}, "audited_at": "2026-10-27T13:25:24.305201+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/224.json" }, { "id": "google/gemma-4-E2B-it", "slug": "google--gemma-4-e2b-it", "name": "Gemma 4 E2B IT", "developer": "Google", "developer_id": "google", "family": "gemma", "description": "Instruction-tuned dense Gemma 4 model with per-layer embeddings, text, image, audio and video inputs, text output, configurable thinking and function calling.", "release_date": "2026-04-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 5.1043, "active_parameters_b": 2.3, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "image", "audio", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "source_type": "primary", "sources": [ { "label": "Official Gemma 4 E2B IT model card", "url": "https://huggingface.co/google/gemma-4-E2B-it/blob/3e22461f65e89153144f8adb70e3b8c2cc9845a7/README.md", "type": "weights" }, { "label": "Exact E2B IT checkpoint configuration", "url": "https://huggingface.co/google/gemma-4-E2B-it/resolve/3e22461f65e89153144f8adb70e3b8c2cc9845a7/config.json", "type": "weights" }, { "label": "Official E2B IT safetensors header", "url": "https://huggingface.co/google/gemma-4-E2B-it/resolve/3e22461f65e89153144f8adb70e3b8c2cc9845a7/model.safetensors", "type": "weights" }, { "label": "Pinned Gemma 4 Transformers implementation", "url": "https://github.com/huggingface/transformers/blob/a672eb5f474b8b30173e9f96eab66adf292da635/src/transformers/models/gemma4/modeling_gemma4.py", "type": "primary" }, { "label": "Pinned Gemma 4 checkpoint converter", "url": "https://github.com/huggingface/transformers/blob/a672eb5f474b8b30173e9f96eab66adf292da635/src/transformers/models/gemma4/convert_gemma4_weights.py", "type": "primary" }, { "label": "Google Gemma 4 attention and transformer blocks", "url": "https://github.com/google-deepmind/gemma/blob/9e1f74131ba9d45e5fe6ae7bb3941d9f2abb39c3/gemma/gm/nn/gemma4/_modules.py", "type": "primary" }, { "label": "Gemma 4 Technical Report", "url": "https://arxiv.org/html/2607.02770v1", "type": "primary" }, { "label": "Gemma 4 launch announcement", "url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "type": "primary" }, { "label": "Gemma release chronology", "url": "https://ai.google.dev/gemma/docs/releases", "type": "primary" }, { "label": "Gemma 4 multi-token prediction documentation", "url": "https://ai.google.dev/gemma/docs/mtp/mtp", "type": "primary" }, { "label": "A Visual Guide to Gemma 4", "url": "https://gemma.ai/tutorials/visual-guides/gemma-4/", "type": "primary" }, { "label": "Jais 3 original comparison table: Gemma 4 E2B activated parameters", "url": "https://huggingface.co/inception42/Jais-3-3.7B/blob/351b5ad88efe27531926621931a069b4a9e0d09d/README.md", "type": "primary" }, { "label": "Jais 3 release and comparison methodology", "url": "https://ifm.ai/blog/jais3/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemma-4-E2B-it.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/google/gemma-4-E2B-it", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Approximately 5.1043B complete usable-model parameters, including PLE, audio, vision and projectors; stored checkpoint values total about 5.1232B because unused shared-layer tensors remain. Active 2.3B applies IFM/Inception42’s E2B comparison convention to IT through Google’s documented shared base/IT architecture; the comparison omits the IT suffix. Google calls 2.3B effective. Exact per-token and modality coverage are unspecified. Tied embeddings count once; separate MTP assistant excluded. Video uses image frames. April 2, 2026 is retained from contemporaneous release evidence; the current family chronology says March 31.", "field_status": {}, "audited_at": "2026-10-27T13:28:26.733423+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/225.json" }, { "id": "google/gemma-4-E4B-it", "slug": "google--gemma-4-e4b-it", "name": "Gemma 4 E4B IT", "developer": "Google", "developer_id": "google", "family": "gemma", "description": "Instruction-tuned dense Gemma 4 model with per-layer embeddings, text, image, audio and video input, text output, configurable thinking and function calling.", "release_date": "2026-04-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 7.9411, "active_parameters_b": 4, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "image", "audio", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/google/gemma-4-E4B-it", "source_type": "primary", "sources": [ { "label": "Exact Gemma 4 E4B IT model card", "url": "https://huggingface.co/google/gemma-4-E4B-it/blob/ee0ef6023621cff504d758262d4e04895a5af4a2/README.md", "type": "primary" }, { "label": "Exact Gemma 4 E4B IT configuration", "url": "https://huggingface.co/google/gemma-4-E4B-it/resolve/ee0ef6023621cff504d758262d4e04895a5af4a2/config.json", "type": "weights" }, { "label": "Exact E4B BF16 checkpoint header", "url": "https://huggingface.co/google/gemma-4-E4B-it/resolve/ee0ef6023621cff504d758262d4e04895a5af4a2/model.safetensors", "type": "weights" }, { "label": "Gemma 4 Transformers implementation", "url": "https://github.com/huggingface/transformers/blob/a672eb5f474b8b30173e9f96eab66adf292da635/src/transformers/models/gemma4/modeling_gemma4.py", "type": "primary" }, { "label": "Gemma 4 checkpoint conversion", "url": "https://github.com/huggingface/transformers/blob/a672eb5f474b8b30173e9f96eab66adf292da635/src/transformers/models/gemma4/convert_gemma4_weights.py", "type": "primary" }, { "label": "Google Gemma 4 block and attention implementation", "url": "https://github.com/google-deepmind/gemma/blob/9e1f74131ba9d45e5fe6ae7bb3941d9f2abb39c3/gemma/gm/nn/gemma4/_modules.py", "type": "primary" }, { "label": "Gemma 4 Technical Report", "url": "https://arxiv.org/html/2607.02770v1", "type": "primary" }, { "label": "Gemma 4 launch announcement", "url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/", "type": "primary" }, { "label": "Gemma release chronology", "url": "https://ai.google.dev/gemma/docs/releases", "type": "primary" }, { "label": "Gemma 4 multi-token prediction documentation", "url": "https://ai.google.dev/gemma/docs/mtp/mtp", "type": "primary" }, { "label": "Gemma 4 model overview and memory planning", "url": "https://ai.google.dev/gemma/docs/core.md", "type": "primary" }, { "label": "Current official Gemma 4 model card", "url": "https://ai.google.dev/gemma/docs/core/model_card_4", "type": "primary" }, { "label": "Official named Gemma 4 E4B base card", "url": "https://huggingface.co/google/gemma-4-E4B/blob/main/README.md", "type": "primary" }, { "label": "ALEM: Benchmarking Open-Ended Multi-Agent Coordination in Language Agents", "url": "https://arxiv.org/html/2606.08340v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemma-4-E4B-it.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/google/gemma-4-E4B-it", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Approximately 7.9411B parameters in the complete usable model under the pinned Transformers implementation, including per-layer embeddings and audio/vision components. Google reports a nominal 8B total; stored weights are approximately 7.9962B before removing redundant shared-layer KV weights. Approximately 4B active follows Tessera et al.'s ALEM benchmark convention for E4B IT; Google separately reports 4.5B effective. Separate MTP assistant excluded. Video uses image frames. April 2 public launch retained; current release chronology lists March 31.", "field_status": {}, "audited_at": "2026-10-27T13:28:58.521403+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/226.json" }, { "id": "stepfun/step-3.5-flash-2603", "slug": "stepfun--step-3.5-flash-2603", "name": "Step 3.5 Flash 2603", "developer": "StepFun", "developer_id": "stepfun", "family": "step", "description": "Agent-tuned text model in the Step 3.5 Flash family, with low/high reasoning modes and tool calling.", "release_date": "2026-04-02", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 196, "active_parameters_b": 11, "parameters_source": "primary_source", "architecture": "moe", "context_window": 256000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/stepfun/step-3.5-flash-2603.toml", "source_type": "catalog", "sources": [ { "label": "StepFun model guide and named variants", "url": "https://platform.stepfun.ai/docs/en/guides/models/step-3.5-flash.md", "type": "primary" }, { "label": "StepFun exact model overview", "url": "https://platform.stepfun.ai/docs/en/guides/models/overview", "type": "primary" }, { "label": "StepFun API service terms", "url": "https://platform.stepfun.ai/docs/en/agreement/userservice.md", "type": "primary" }, { "label": "Contemporaneous April 2 availability report", "url": "https://www.nbd.com.cn/articles/2026-04-02/4323121.html", "type": "secondary" }, { "label": "Hugging Face", "url": "https://huggingface.co/stepfun-ai/Step-3.5-Flash", "type": "weights" }, { "label": "StepFun public checkpoint inventory", "url": "https://huggingface.co/api/models?author=stepfun-ai&limit=100", "type": "primary" }, { "label": "Exact Step 3.5 Flash 2603 weight search", "url": "https://huggingface.co/api/models?search=Step-3.5-Flash-2603&limit=100", "type": "primary" }, { "label": "Original Step 3.5 Flash checkpoint branches and tags", "url": "https://huggingface.co/api/models/stepfun-ai/Step-3.5-Flash/refs", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/stepfun/step-3.5-flash-2603.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "196B total and 11B active are nominal figures from StepFun's shared model guide listing 2603 as a tuned variant; they are not an exact dated-checkpoint or full-MTP count. No public 2603 weights were found; the original Step 3.5 Flash checkpoint does not establish equivalence. Proprietary describes the API terms, not a discovered weight license. Context is advertised as 256K. April 2, 2026 availability is corroborated by contemporaneous press; no authenticated primary release announcement was verified. Classified closed for the 2603 API variant because no matching public checkpoint was found.", "field_status": {}, "audited_at": "2026-10-28T08:06:27.019825Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/495.json" }, { "id": "alibaba/qwen3.6-plus", "slug": "alibaba--qwen3.6-plus", "name": "Qwen3.6 Plus", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Alibaba Qwen model for agentic coding and multimodal understanding, with a one-million-token context window.", "release_date": "2026-04-01", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1000000, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.6-plus.toml", "source_type": "catalog", "sources": [ { "label": "Alibaba Unveils Qwen3.6-Plus to Accelerate Agentic AI Deployment for Enterprises and Alibaba’s AI Applications", "url": "https://www.alibabacloud.com/en/press-room/alibaba-unveils-qwen3-6-plus-to-accelerate-agentic?_p_lc=1", "type": "primary" }, { "label": "qwen3.6-plus — Alibaba Cloud Model Studio", "url": "https://docs.modelstudio.console.alibabacloud.com/en/model-studio/qwen3-6-plus", "type": "primary" }, { "label": "Model lifecycle and updates", "url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "type": "primary" }, { "label": "Text generation models", "url": "https://www.alibabacloud.com/help/en/model-studio/text-generation-model", "type": "primary" }, { "label": "Qwen3.6-Plus: Towards Real World Agents", "url": "https://www.alibabacloud.com/blog/alibaba-unveils-qwen3-6-plus-to-accelerate-agentic-ai-deployment-for-enterprises-and-alibaba%E2%80%99s-ai-applications_603005", "type": "primary" }, { "label": "Qwen3.6-35B-A3B: Agentic Coding Power, Now Open to All", "url": "https://www.alibabacloud.com/blog/qwen3-6-35b-a3b-thinking-open-weights-uncompromised-agentic-coding_603043", "type": "primary" }, { "label": "Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Model", "url": "https://www.alibabacloud.com/blog/603063", "type": "primary" }, { "label": "Qwen3.7-Plus: Multimodal Agent Intelligence", "url": "https://www.alibabacloud.com/blog/qwen3-7-plus-multimodal-agent-intelligence_603206", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=qwen3.6-plus&limit=100", "label": "Official Qwen inventory search for the exact hosted model", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=Qwen3.6&limit=1000", "label": "Official Qwen Qwen3.6 weight inventory", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.6-plus.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Model Studio records April 1, 2026 availability, preceding the April 2 launch announcement. The alias maps to qwen3.6-plus-2026-04-02. Total and active parameters remain undisclosed, with no established lineage to a quantified base; the published 35B/3B figures identify the separate Flash/35B-A3B model. Architecture and weight license are undisclosed. No public Plus weight release is available. Classified as closed weights: no public release of this hosted variant was found in the developer documentation and weight inventory.", "field_status": { "license": "undisclosed", "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-28T08:01:10.205724+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/062.json" }, { "id": "arcee-ai/trinity-large-thinking", "slug": "arcee-ai--trinity-large-thinking", "name": "Trinity Large Thinking", "developer": "Arcee AI", "developer_id": "arcee-ai", "family": "trinity", "description": "Open-weight Trinity MoE language model with dedicated thinking and structured tool calling.", "release_date": "2026-04-01", "date_precision": "day", "open_weights": true, "license": "OpenMDW-1.1", "parameters_b": 398.635286016, "active_parameters_b": 13, "parameters_source": "primary_source", "architecture": "moe", "context_window": 512000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.arcee.ai/blog/trinity-large-thinking", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Thinking", "type": "weights" }, { "label": "Announcement", "url": "https://www.arcee.ai/blog/trinity-large-thinking", "type": "announcement" }, { "label": "Trinity-Large-Thinking pinned configuration", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Thinking/raw/dc6a99b6b74202880e31491a445a94884c97890c/config.json", "type": "primary" }, { "label": "Trinity-Large-Thinking OpenMDW-1.1 license", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Thinking/raw/dc6a99b6b74202880e31491a445a94884c97890c/LICENSE", "type": "primary" }, { "label": "Arcee Trinity Large Technical Report", "url": "https://arxiv.org/pdf/2602.17004v1", "type": "primary" }, { "label": "Trinity-Large-Thinking pinned model implementation", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Thinking/raw/dc6a99b6b74202880e31491a445a94884c97890c/modeling_afmoe.py", "type": "primary" }, { "label": "Trinity-Large-Thinking pinned checkpoint index", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Thinking/raw/dc6a99b6b74202880e31491a445a94884c97890c/model.safetensors.index.json", "type": "primary" }, { "label": "Trinity-Large-Thinking complete pinned safetensors headers", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Thinking/tree/dc6a99b6b74202880e31491a445a94884c97890c", "type": "primary" }, { "label": "Trinity is moving to OpenMDW-1.1", "url": "https://www.arcee.ai/blog/trinity-is-moving-to-openmdw-1-1", "type": "primary" }, { "label": "Trinity-Large-Thinking repository history", "url": "https://huggingface.co/api/models/arcee-ai/Trinity-Large-Thinking/commits/main", "type": "primary" }, { "label": "Trinity-Large-Thinking pinned chat template", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Thinking/raw/dc6a99b6b74202880e31491a445a94884c97890c/chat_template.jinja", "type": "primary" }, { "label": "Trinity-Large-Thinking developer documentation", "url": "https://docs.arcee.ai/arcee-language-models/trinity-family/trinity-large-thinking.md", "type": "primary" }, { "label": "Arcee vLLM deployment guide", "url": "https://docs.arcee.ai/quick-deploys/inference-engines/vllm", "type": "primary" }, { "label": "Genesis-Science-1", "url": "https://www.arcee.ai/blog/genesis-science-1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/arcee-ai/trinity-large-thinking.toml", "type": "catalog" }, { "label": "License", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Thinking/blob/main/LICENSE", "type": "license" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Full checkpoint contains 398.635286016B parameters; approximately 398B/400B are developer nominal descriptions. Approximately 13B active is the developer per-token figure, consistent with four routed experts plus shared and dense components. OpenMDW-1.1 superseded launch Apache-2.0 wording in May 2026. Context 512000 is decimal normalization of the exact Thinking card/docs advertised 512k extension, not a literal primary-source integer or verified exact boundary; released configuration separately sets 262144.", "parameter_components_b": { "routed_experts": 391.378894848, "shared_experts": 1.528823808, "attention_including_qk_norms": 3.77488896, "dense_mlp": 0.679477248, "router_weights": 0.042467328, "input_embeddings": 0.614989824, "output_head": 0.614989824, "other_norms": 0.000740352, "expert_balancing_bias": 0.000013824 }, "field_status": {}, "audited_at": "2026-10-26T17:21:35.053129Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/119.json" }, { "id": "zhipuai/glm-5v-turbo", "slug": "zhipuai--glm-5v-turbo", "name": "GLM-5V-Turbo", "developer": "Z.ai", "developer_id": "zhipuai", "family": "glm", "description": "Vision-language Transformer with CogViT, multimodal token prediction, reasoning and tool calling.", "release_date": "2026-04-01", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 202752, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.zhipuai.cn/zh/research/156", "source_type": "primary", "sources": [ { "label": "Original GLM-5V-Turbo release announcement", "url": "https://www.zhipuai.cn/zh/research/156", "type": "docs" }, { "label": "Original international GLM-5V-Turbo launch post", "url": "https://x.com/Zai_org/status/2039371126984360085", "type": "docs" }, { "label": "GLM-5V-Turbo technical report", "url": "https://arxiv.org/html/2604.26752v1", "type": "paper" }, { "label": "GLM-5V-Turbo native API guide", "url": "https://docs.z.ai/guides/vlm/glm-5v-turbo", "type": "docs" }, { "label": "Exact GLM-5V-Turbo Z.AI route metadata", "url": "https://openrouter.ai/api/v1/models/z-ai/glm-5v-turbo/endpoints", "type": "docs" }, { "label": "BigModel release log", "url": "https://docs.bigmodel.cn/cn/update/new-releases.md", "type": "docs" }, { "label": "Z.ai hosted API terms", "url": "https://docs.z.ai/legal-agreement/terms-of-use.md", "type": "docs" }, { "label": "Official public checkpoint inventory", "url": "https://huggingface.co/api/models?author=zai-org&limit=1000", "type": "docs" }, { "label": "Official GLM-V release history and downloads", "url": "https://github.com/zai-org/GLM-V/blob/11201646b9233db4068b7a98f0122a7f6ce85098/README.md", "type": "docs" }, { "label": "Later GLM-5.3-Flash original card", "url": "https://huggingface.co/zai-org/GLM-5.3-Flash/raw/main/README.md", "type": "docs" }, { "label": "Later GLM-5.3 original card", "url": "https://huggingface.co/zai-org/GLM-5.3/raw/main/README.md", "type": "docs" }, { "label": "Later GLM-5.2 original card", "url": "https://huggingface.co/zai-org/GLM-5.2/raw/main/README.md", "type": "docs" }, { "label": "Revised GLM-V technical report", "url": "https://arxiv.org/html/2507.01006v6", "type": "paper" }, { "label": "Original GLM-5 family comparison and download table", "url": "https://github.com/zai-org/GLM-5/blob/main/README.md", "type": "docs" }, { "label": "P3D-Bench original evaluation, Table 2", "url": "https://arxiv.org/html/2606.11152v1", "type": "paper" }, { "label": "Original GLM-5 model card", "url": "https://huggingface.co/zai-org/GLM-5/raw/main/README.md", "type": "docs" }, { "label": "GLM-5 technical report, Appendix A Table 10", "url": "https://arxiv.org/html/2602.15763v1", "type": "paper" }, { "label": "Original GLM-5-Turbo named-variant launch", "url": "https://x.com/Zai_org/status/2033221428640674015", "type": "docs" }, { "label": "Current official GLM-V release history and downloads", "url": "https://raw.githubusercontent.com/zai-org/GLM-V/main/README.md", "type": "docs" }, { "label": "Z.ai exact GLM 5V Turbo API guide", "url": "https://docs.z.ai/guides/vlm/glm-5v-turbo.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/zhipuai/glm-5v-turbo.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released April 1, 2026 via native API and chat; BigModel and the GLM-V repository use April 2. Production total/active counts remain undisclosed. The report names GLM-5-Turbo as its language-only base but adds CogViT/MLP and multimodal MTP; the base family's 744B/40B cannot establish this model's counts. Its 0.5B ablation is not the released model. No exact public 5V Turbo checkpoint was found; proprietary describes hosted API terms. Native-provider context is 202752 tokens, advertised as 200K, with a separate 131072 output cap. Files contain text and visual inputs; output is text. Classified closed because no public downloadable GLM-5V-Turbo checkpoint was found.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-28T08:06:49.472255Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/563.json" }, { "id": "bytedance-seed/seed-character", "slug": "bytedance-seed--seed-character", "name": "Seed Character", "developer": "ByteDance", "developer_id": "bytedance-seed", "family": "seed", "description": "ByteDance Seed model for roleplay and narrative dialogue, with text generation, image and video understanding, thinking, and tool calling in its 260628 revision.", "release_date": "2026-04", "date_precision": "month", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 128000, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/bytedance-seed/seed-character.toml", "source_type": "catalog", "sources": [ { "label": "Ark exact-version model specifications", "url": "https://console.volcengine.com/ark/region:cn-beijing/docs/model-list", "type": "primary" }, { "label": "Ark model release announcements", "url": "https://console.volcengine.com/ark/region:cn-beijing/docs/model-release-announcement", "type": "primary" }, { "label": "Doubao model service agreement", "url": "https://docs.volcengine.com/docs/ark/doubao-model-service-agreement?lang=en", "type": "primary" }, { "label": "Ark model list - complete official Markdown", "url": "https://ark.volcengine.com/region:cn-beijing/docs/model-list.md", "type": "primary" }, { "label": "Ark release announcements - complete official Markdown", "url": "https://ark.volcengine.com/region:cn-beijing/docs/model-release-announcement.md", "type": "primary" }, { "label": "Seed2.0 Model Card: Towards Intelligence Frontier for Real-World Complexity", "url": "https://arxiv.org/abs/2607.00248", "type": "primary" }, { "label": "Seed2.1 officially released: advancing AI productivity", "url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/bytedance-seed/seed-character.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Release month refers to the earliest explicitly documented Seed Character release (251128, April 2026). Capabilities and 128k context describe the later 260628 revision announced in June 2026. API revision suffixes are not launch dates. Nominal total and active parameter counts remain undisclosed; no publicly sized base relationship is documented. Architecture remains undisclosed.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T11:04:15.289532+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/138.json" }, { "id": "cognitivecomputations/dolphin-mistral-24b-venice-edition-v1.2", "slug": "cognitivecomputations--dolphin-mistral-24b-venice-edition-v1.2", "name": "Dolphin Mistral 24B Venice Edition 1.2", "developer": "Dolphin (Cognitive Computations) and Venice", "developer_id": "cognitivecomputations", "family": "mistral", "description": "Vision-language revision of the Dolphin and Venice conversational model.", "release_date": "2026-04", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 24.01136128, "active_parameters_b": 24, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://venice.ai/models/venice-uncensored-1-2", "source_type": "primary", "sources": [ { "label": "Venice 1.2 model datasheet", "url": "https://venice.ai/models/venice-uncensored-1-2", "type": "primary" }, { "label": "Developer replaces weights with version 1.2", "url": "https://huggingface.co/dphn/Dolphin-Mistral-24B-Venice-Edition/commit/337ce042026e598256f4bbb8178116b59c9e302d", "type": "primary" }, { "label": "Version 1.2 model card", "url": "https://huggingface.co/dphn/Dolphin-Mistral-24B-Venice-Edition/blob/b23ff40b4a25e4f93de432bd896d4464479a9f79/README.md", "type": "primary" }, { "label": "Version 1.2 configuration", "url": "https://huggingface.co/dphn/Dolphin-Mistral-24B-Venice-Edition/blob/b23ff40b4a25e4f93de432bd896d4464479a9f79/config.json", "type": "primary" }, { "label": "Version 1.2 weight index", "url": "https://huggingface.co/dphn/Dolphin-Mistral-24B-Venice-Edition/blob/b23ff40b4a25e4f93de432bd896d4464479a9f79/model.safetensors.index.json", "type": "primary" }, { "label": "Venice announces the newly trained revision", "url": "https://www.linkedin.com/posts/veniceai_venice-uncensored-12-is-now-live-developed-activity-7450602742793859075-II3o", "type": "primary" }, { "label": "Native Venice model capabilities", "url": "https://api.venice.ai/api/v1/models", "type": "primary" }, { "label": "Checkpoint revision history", "url": "https://huggingface.co/api/models/dphn/Dolphin-Mistral-24B-Venice-Edition/commits/main", "type": "primary" }, { "label": "Base model nominal active count", "url": "https://docs.mistral.ai/models/mistral-small-3-0-25-01", "type": "primary" } ], "aliases": [ "venice/venice-uncensored-1-2" ], "first_seen": "2026-10-29T11:32:22.060Z", "last_seen": "2026-10-29T11:32:22.060Z", "upstream_present": false, "availability": "public", "active_parameters_scope": "language_model", "parameter_components_b": { "language_model": 23.5724032, "vision_encoder": 0.403305472, "projector": 0.035652608 }, "notes": "Distinct revision trained with supervised fine-tuning and KTO reinforcement learning. Complete 24.01136128B checkpoint count comprises 23.5724032B language, 0.403305472B vision and 0.035652608B projector weights; nominal 24B active is scoped to the language component, not the full multimodal model. Weight configuration supports 131072 context; the Venice API serves 128000. April 16, 2026 is the verified weight replacement and public release post; the current publisher datasheet and API timestamp claim April 1. Month precision avoids overstating the first-access day." }, { "id": "mistral/mistral-medium-3-5", "slug": "mistral--mistral-medium-3-5", "name": "Mistral Medium 3.5", "developer": "Mistral", "developer_id": "mistral", "family": "mistral-medium", "description": "Mistral model for multilingual chat, reasoning, and tool-assisted workflows", "release_date": "2026-04", "date_precision": "month", "open_weights": true, "license": "Modified MIT", "parameters_b": 128, "active_parameters_b": 128, "parameters_source": "primary_source", "architecture": "dense", "context_window": 262144, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://docs.mistral.ai/models/mistral-medium-3-5-26-04", "source_type": "primary", "sources": [ { "label": "Alias documentation", "url": "https://docs.mistral.ai/models/mistral-medium-3-5-26-04", "type": "docs" }, { "label": "Official Medium 3.5 model card", "url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/blob/22b2b868a15677cfa6061277ed2f653d1349a9ab/README.md", "type": "weights" }, { "label": "Official Medium 3.5 native configuration", "url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/blob/22b2b868a15677cfa6061277ed2f653d1349a9ab/params.json", "type": "weights" }, { "label": "Official Medium 3.5 Transformers configuration", "url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/blob/22b2b868a15677cfa6061277ed2f653d1349a9ab/config.json", "type": "weights" }, { "label": "Official comparison with total and active counts", "url": "https://docs.mistral.ai/inference/model-selection-guide?models=mistral-medium-3-5-26-04", "type": "primary" }, { "label": "Exact Modified MIT license", "url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/blob/22b2b868a15677cfa6061277ed2f653d1349a9ab/LICENSE", "type": "weights" }, { "label": "Official legal model lifecycle with differing day", "url": "https://legal.mistral.ai/ai-governance/models/mistral-medium-3/", "type": "docs" }, { "label": "Official release changelog", "url": "https://docs.mistral.ai/resources/changelogs", "type": "primary" }, { "label": "Remote agents announcement with later displayed date", "url": "https://mistral.ai/news/vibe-remote-agents-mistral-medium-3-5/", "type": "primary" }, { "label": "Document QnA OCR preprocessing", "url": "https://docs.mistral.ai/studio/document-processing/document_qna", "type": "primary" }, { "label": "April 29 public deployment pull request", "url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/discussions/1", "type": "primary" }, { "label": "Official 128B model weights", "url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/mistral-medium-3-5.toml", "type": "catalog" } ], "aliases": [ "mistral/mistral-medium-2604" ], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "128B total and 128B active are Mistral's nominal model-level disclosures; the active value is explicitly published in its comparison table, not inferred from density or an exact per-token census. The dense multimodal checkpoint accepts text and images; PDF QnA uses separate OCR preprocessing. Native context is 262144 tokens. The Modified MIT license contains a company/employer revenue restriction above $20 million in the preceding month. Release remains April 2026: inference documentation and changelog say April 28, legal lifecycle says April 29, and public repository activity is confirmed April 29. The linked blog currently displays May 22, a later date.", "field_status": {}, "audited_at": "2026-10-26T23:22:26.349540Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/338.json", "alias_sources": { "mistral/mistral-medium-2604": { "url": "https://docs.mistral.ai/models/mistral-medium-3-5-26-04", "note": "Mistral Medium 3.5 and its April 2026 catalog name refer to one release, officially version 26.04." } } }, { "id": "liquid/lfm2.5-350m", "slug": "liquid--lfm2.5-350m", "name": "LFM2.5-350M", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm", "description": null, "release_date": "2026-03-31", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 0.35, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "hybrid", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://www.liquid.ai/blog/lfm2-5-350m-no-size-left-behind", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/lfm2-5-350m-no-size-left-behind", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/LiquidAI/LFM2.5-350M/blob/9e6c6ccf47cd318696e137d381a7ded8fe4df09f/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/LiquidAI/LFM2.5-350M/blob/9e6c6ccf47cd318696e137d381a7ded8fe4df09f/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:23.459Z", "last_seen": "2026-10-29T10:17:23.459Z", "upstream_present": false, "availability": "public", "notes": "Post-trained checkpoint explicitly included in the developer release. Parameter and context values follow the exact model card; MoE variants also use the LFM convolution/attention hybrid backbone." }, { "id": "liquid/lfm2.5-350m-base", "slug": "liquid--lfm2.5-350m-base", "name": "LFM2.5-350M-Base", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm", "description": null, "release_date": "2026-03-31", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 0.35, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "hybrid", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.liquid.ai/blog/lfm2-5-350m-no-size-left-behind", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/lfm2-5-350m-no-size-left-behind", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/LiquidAI/LFM2.5-350M-Base/blob/9960764e30892e01f29a6dc23df2533fcd8bd5ae/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/LiquidAI/LFM2.5-350M-Base/blob/9960764e30892e01f29a6dc23df2533fcd8bd5ae/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:23.459Z", "last_seen": "2026-10-29T10:17:23.459Z", "upstream_present": false, "availability": "public", "notes": "Pretrained checkpoint explicitly included in the developer release. Parameter and context values follow the exact model card; MoE variants also use the LFM convolution/attention hybrid backbone." }, { "id": "inception/mercury-edit-2", "slug": "inception--mercury-edit-2", "name": "Mercury Edit 2", "developer": "Inception", "developer_id": "inception", "family": "mercury", "description": "Code editing dLLM for autocomplete (FIM) and next-edit suggestions", "release_date": "2026-03-30", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "diffusion", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": false, "source_url": "https://www.inceptionlabs.ai/blog/introducing-mercury-edit-2", "source_type": "primary", "sources": [ { "label": "Introducing Mercury Edit 2", "url": "https://www.inceptionlabs.ai/blog/introducing-mercury-edit-2", "type": "primary" }, { "label": "Inception Blog: Mercury Edit 2 launch index", "url": "https://www.inceptionlabs.ai/blog", "type": "primary" }, { "label": "Models - Inception Platform", "url": "https://docs.inceptionlabs.ai/get-started/models", "type": "primary" }, { "label": "Terms of Use - Inception", "url": "https://www.inceptionlabs.ai/docs/terms-of-use", "type": "primary" }, { "label": "Inception API OpenAPI specification", "url": "https://api.inceptionlabs.ai/openapi.json", "type": "primary" }, { "label": "Reasoning Efforts - Inception Platform", "url": "https://docs.inceptionlabs.ai/capabilities/reasoning-efforts", "type": "primary" }, { "label": "Live Inception FIM model listing", "url": "https://api.inceptionlabs.ai/v1/fim/completions/models", "type": "primary" }, { "label": "Live Inception Edit model listing", "url": "https://api.inceptionlabs.ai/v1/edit/completions/models", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/inception/mercury-edit-2.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Mercury Edit 2 is a diffusion code-editing model. No total or active parameter count was found in the checked primary sources; its denoiser backbone and dense/MoE status are undisclosed. Context is unresolved: the model documentation specifies 32K for FIM and NextEdit, but both public live model listings return 128,000 context and 32,000 output tokens; the documented request schemas limit output to 8,192. Public release is March 30, 2026, not the inconsistent April 1, 2025 metadata creation timestamp. No dedicated reasoning mode or tool calling is exposed on its documented FIM/edit endpoints. Current inference availability was not tested.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "context_window": "unverified" }, "audited_at": "2026-10-27T13:41:18.561193+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/241.json" }, { "id": "google/gemini-3.1-flash-live-preview", "slug": "google--gemini-3.1-flash-live-preview", "name": "Gemini 3.1 Flash Live Preview", "developer": "Google", "developer_id": "google", "family": "gemini-flash", "description": "Preview Live API model for real-time voice and video dialogue, with text and audio responses, configurable thinking, and synchronous function calling.", "release_date": "2026-03-26", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text", "audio" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-live-preview", "source_type": "primary", "sources": [ { "label": "Gemini 3.1 Flash Live Preview API documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-live-preview", "type": "primary" }, { "label": "Gemini 3.1 Flash Live original model card", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-1-Flash-Live-Model-Card.pdf", "type": "primary" }, { "label": "Gemini 3.1 Flash Audio model card", "url": "https://deepmind.google/models/model-cards/gemini-3-1-flash-audio/", "type": "primary" }, { "label": "Referenced Gemini 3 Pro architecture", "url": "https://deepmind.google/models/model-cards/gemini-3-pro/", "type": "primary" }, { "label": "Google launch announcement", "url": "https://blog.google/innovation-and-ai/technology/developers-tools/build-with-gemini-3-1-flash-live/", "type": "primary" }, { "label": "Gemini API release notes", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "primary" }, { "label": "Gemini API terms", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Gemini deprecations", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemini-3.1-flash-live-preview.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Sparse MoE Transformer based on Gemini 3 Pro; neither the Live variant nor its named base has a published total or active parameter count in the reviewed sources. Exact Live specifications retain 131,072 input tokens and text/audio outputs. The current API input summary omits video, while both model cards and API migration instructions support it. March 26, 2026 is the explicit public launch and changelog date; the deprecations table inconsistently lists March 11. As of this audit, Google schedules shutdown for November 17, 2026; endpoint operation was not tested.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T13:00:23.426793+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/204.json" }, { "id": "google/lyria-3-clip-preview", "slug": "google--lyria-3-clip-preview", "name": "Lyria 3 Clip Preview", "developer": "Google", "developer_id": "google", "family": "lyria", "description": "Google music-generation model that produces 30-second clips and lyrics from text or image prompts.", "release_date": "2026-03-25", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "latent-diffusion", "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text", "audio" ] }, "reasoning": false, "tool_call": false, "source_url": "https://ai.google.dev/gemini-api/docs/models/lyria-3-clip-preview", "source_type": "primary", "sources": [ { "label": "Lyria 3 Clip preview API specifications", "url": "https://ai.google.dev/gemini-api/docs/models/lyria-3-clip-preview", "type": "primary" }, { "label": "Gemini API release notes, March 25, 2026", "url": "https://ai.google.dev/gemini-api/docs/changelog#march-25-2026", "type": "primary" }, { "label": "Lyria 3 model card, updated March 2026", "url": "https://deepmind.google/models/model-cards/lyria-3/", "type": "primary" }, { "label": "Build with Lyria 3, our newest music generation model", "url": "https://blog.google/innovation-and-ai/technology/developers-tools/lyria-3-developers/", "type": "primary" }, { "label": "Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Music generation guide", "url": "https://ai.google.dev/gemini-api/docs/music-generation", "type": "primary" }, { "label": "Lyria 3.5 model card, July 2026", "url": "https://deepmind.google/models/model-cards/lyria-3-5/", "type": "primary" }, { "label": "Google Cloud Lyria 3 model specifications", "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/lyria/lyria-3", "type": "primary" }, { "label": "Gemini API model lifecycle", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/lyria-3-clip-preview.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Lyria 3 uses latent diffusion over audio latents. Google has not published total or active parameter counts for Clip. The 131,072-token value is the API input limit; Clip output is fixed at 30 seconds. The exact API page says 48 kHz, while the current generation guide and Google Cloud specification say 44.1 kHz. The March 25 date is the public Clip API preview release.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T13:28:04.283084+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/227.json" }, { "id": "google/lyria-3-pro-preview", "slug": "google--lyria-3-pro-preview", "name": "Lyria 3 Pro Preview", "developer": "Google", "developer_id": "google", "family": "lyria", "description": "Music generation model for songs and lyrics from text or image prompts, with vocals and controls for song structure.", "release_date": "2026-03-25", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "latent-diffusion", "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text", "audio" ] }, "reasoning": false, "tool_call": false, "source_url": "https://ai.google.dev/gemini-api/docs/changelog", "source_type": "primary", "sources": [ { "label": "Lyria 3 Pro Preview API specification", "url": "https://ai.google.dev/gemini-api/docs/models/lyria-3-pro-preview", "type": "primary" }, { "label": "Gemini API changelog: March 25, 2026", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "primary" }, { "label": "Lyria 3 Pro: Create longer tracks in more Google products", "url": "https://blog.google/innovation-and-ai/technology/ai/lyria-3-pro/", "type": "primary" }, { "label": "Lyria 3 model card, updated March 2026", "url": "https://deepmind.google/models/model-cards/lyria-3/", "type": "primary" }, { "label": "Lyria 3.5 model card, July 2026", "url": "https://deepmind.google/models/model-cards/lyria-3-5/", "type": "primary" }, { "label": "Google Cloud Lyria 3: exact Pro Preview section", "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/lyria/lyria-3", "type": "primary" }, { "label": "Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Gemini API model lifecycle: Lyria", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/lyria-3-pro-preview.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "No total or active parameter count was found in the checked primary sources. The Lyria 3 card describes latent diffusion and covers subsequent versions. The Pro API page documents 131,072 input tokens but mistakenly labels its model-code row as Clip; its title, section and version list identify Pro. Gemini and Cloud documentation disagree on audio sample rate, so no rate is asserted. Lyria 3.5 is the recommended replacement; no shutdown date is announced.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T13:30:27.163380+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/228.json" }, { "id": "nvidia/nemotron-cascade-2-30b-a3b", "slug": "nvidia--nemotron-cascade-2-30b-a3b", "name": "Nemotron Cascade 2 30B A3B", "developer": "NVIDIA", "developer_id": "nvidia", "family": "nemotron", "description": "Text model combining Mamba-2, attention and sparse MoE layers, post-trained with Cascade RL and multi-domain on-policy distillation for reasoning and tool use.", "release_date": "2026-03-19", "date_precision": "day", "open_weights": true, "license": "NVIDIA Open Model License", "parameters_b": 31.577937344, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "hybrid-ssm-transformer-moe", "context_window": 1000000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/nvidia/Nemotron-Cascade-2-30B-A3B/blob/6327cdbcf907e1c7cec9cb29fb6e6cebdf8feaf7/README.md", "source_type": "primary", "sources": [ { "label": "Nemotron-Cascade-2-30B-A3B model card", "url": "https://huggingface.co/nvidia/Nemotron-Cascade-2-30B-A3B/blob/6327cdbcf907e1c7cec9cb29fb6e6cebdf8feaf7/README.md", "type": "primary" }, { "label": "Nemotron-Cascade 2 technical report, revision 2", "url": "https://arxiv.org/html/2603.19220v2", "type": "primary" }, { "label": "Exact Cascade 2 checkpoint configuration", "url": "https://huggingface.co/nvidia/Nemotron-Cascade-2-30B-A3B/blob/6327cdbcf907e1c7cec9cb29fb6e6cebdf8feaf7/config.json", "type": "primary" }, { "label": "Exact Cascade 2 model implementation", "url": "https://huggingface.co/nvidia/Nemotron-Cascade-2-30B-A3B/blob/6327cdbcf907e1c7cec9cb29fb6e6cebdf8feaf7/modeling_nemotron_h.py", "type": "primary" }, { "label": "Cascade 2 BF16 checkpoint index and bounded shard headers", "url": "https://huggingface.co/nvidia/Nemotron-Cascade-2-30B-A3B/blob/6327cdbcf907e1c7cec9cb29fb6e6cebdf8feaf7/model.safetensors.index.json", "type": "primary" }, { "label": "NVIDIA Open Model License Agreement", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/nvidia/nemotron-cascade-2-30b-a3b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Total is an independently reconstructed complete learned BF16 checkpoint count (31.577937344B), excluding 23 routing buffers; active 3B remains the developer’s nominal exact-model claim. Context is the advertised nominal 1M maximum; checkpoint config and example serving commands use 262144. The technical report evaluates this exact model at 1M. March 19 is the explicit model-card release date; NVIDIA’s research page bears a March 16 publication date.", "field_status": {}, "audited_at": "2026-10-27T00:56:02.227066Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/389.json" }, { "id": "minimax/MiniMax-M2.7", "slug": "minimax--minimax-m2.7", "name": "MiniMax-M2.7", "developer": "MiniMax", "developer_id": "minimax", "family": "minimax", "description": "MiniMax M2-series language model with publicly available weights, designed for coding agents, tool use, and office workflows.", "release_date": "2026-03-18", "date_precision": "day", "open_weights": true, "license": "MiniMax Non-Commercial License", "parameters_b": 228.69, "active_parameters_b": 10, "parameters_source": "primary_source", "architecture": "moe", "context_window": 204800, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.minimax.io/news/minimax-m27-en", "source_type": "primary", "sources": [ { "label": "MiniMax M2.7: Early Echoes of Self-Evolution", "url": "https://www.minimax.io/news/minimax-m27-en", "type": "primary" }, { "label": "MiniMax model release notes", "url": "https://platform.minimax.io/docs/release-notes/models", "type": "primary" }, { "label": "Hugging Face", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.7", "type": "weights" }, { "label": "MiniMax-M2.7 current released configuration", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.7/blob/0a6536713b0b03f3be97494e5cf8d22f668a3164/config.json", "type": "weights" }, { "label": "MiniMax-M2.7 reference implementation", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.7/blob/0a6536713b0b03f3be97494e5cf8d22f668a3164/modeling_minimax_m2.py", "type": "weights" }, { "label": "MiniMax-M2.7 Non-Commercial License", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.7/blob/0a6536713b0b03f3be97494e5cf8d22f668a3164/LICENSE", "type": "weights" }, { "label": "The MiniMax-M2 Series: Mini Activations Unleashing Max Real-World Intelligence", "url": "https://arxiv.org/html/2605.26494v2", "type": "primary" }, { "label": "MiniMax-M2.7 checkpoint index and bounded Safetensors header census", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.7/blob/0a6536713b0b03f3be97494e5cf8d22f668a3164/model.safetensors.index.json", "type": "weights" }, { "label": "MiniMax-M2.7 Tool Calling Guide", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.7/blob/0a6536713b0b03f3be97494e5cf8d22f668a3164/docs/tool_calling_guide.md", "type": "weights" }, { "label": "MiniMax-M2.7 chat template", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.7/blob/0a6536713b0b03f3be97494e5cf8d22f668a3164/chat_template.jinja", "type": "weights" }, { "label": "MiniMax model retirement notice and model list", "url": "https://platform.minimax.io/docs/faq/models", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/minimax/MiniMax-M2.7.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Total 228.69B is the complete released-checkpoint parameter count rounded to 0.01B, excluding FP8 scale metadata; the 15872 routing-bias values do not affect this precision. No MTP tensors are present in the release. Active approximately 10B is the developer nominal claim explicitly for M2.7, not the report M2-only 9.8B. Context 204800 is the current checkpoint configuration; current API operation was not established. Weights use the custom MiniMax Non-Commercial License, requiring prior authorization for commercial use. Current FAQ gives conflicting retirement timing (November 18 notice versus already-retired table); it says M Plan/Token Plan calls route to M3.1 Flash from October 19.", "field_status": {}, "audited_at": "2026-10-26T22:33:23.486792Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/305.json" }, { "id": "minimax/MiniMax-M2.7-highspeed", "slug": "minimax--minimax-m2.7-highspeed", "name": "MiniMax-M2.7-highspeed", "developer": "MiniMax", "developer_id": "minimax", "family": "minimax", "description": "Higher-throughput API version of MiniMax M2.7 for text reasoning, coding and tool-using agents, advertised by MiniMax as producing identical results.", "release_date": "2026-03-18", "date_precision": "day", "open_weights": true, "license": "MiniMax M2.7 Non-Commercial License", "parameters_b": 230, "active_parameters_b": 10, "parameters_source": "primary_source", "architecture": "moe", "context_window": 204800, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://platform.minimax.io/docs/release-notes/models", "source_type": "primary", "sources": [ { "label": "MiniMax API model release notes", "url": "https://platform.minimax.io/docs/release-notes/models", "type": "primary" }, { "label": "MiniMax M2.7 product page", "url": "https://www.minimax.io/models/text/m27", "type": "primary" }, { "label": "MiniMax-M2.7 official model card", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.7/blob/0a6536713b0b03f3be97494e5cf8d22f668a3164/README.md", "type": "weights" }, { "label": "MiniMax-M2.7 Non-Commercial License", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.7/blob/0a6536713b0b03f3be97494e5cf8d22f668a3164/LICENSE", "type": "weights" }, { "label": "MiniMax-M2.7 checkpoint configuration", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.7/blob/0a6536713b0b03f3be97494e5cf8d22f668a3164/config.json", "type": "weights" }, { "label": "MiniMax-M2.7 reference model implementation", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.7/blob/0a6536713b0b03f3be97494e5cf8d22f668a3164/modeling_minimax_m2.py", "type": "weights" }, { "label": "MiniMax-M2.7 published checkpoint index and shard headers", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.7/blob/0a6536713b0b03f3be97494e5cf8d22f668a3164/model.safetensors.index.json", "type": "weights" }, { "label": "The MiniMax-M2 Series: Mini Activations Unleashing Max Real-World Intelligence", "url": "https://arxiv.org/html/2605.26494v1", "type": "primary" }, { "label": "MiniMax M2.7 Advances Scalable Agentic Workflows on NVIDIA Platforms for Complex AI Applications", "url": "https://developer.nvidia.com/blog/minimax-m2-7-advances-scalable-agentic-workflows-on-nvidia-platforms-for-complex-ai-applications/", "type": "primary" }, { "label": "MiniMax OpenAI-compatible SDK documentation", "url": "https://platform.minimax.io/docs/api-reference/text-openai-api", "type": "primary" }, { "label": "MiniMax Anthropic-compatible SDK documentation", "url": "https://platform.minimax.io/docs/api-reference/text-anthropic-api", "type": "primary" }, { "label": "MiniMax M2.7 High Speed on Vercel AI Gateway", "url": "https://vercel.com/ai-gateway/models/minimax-m2.7-highspeed", "type": "primary" }, { "label": "MiniMax model list and M2.5/M2.7 retirement notice", "url": "https://platform.minimax.io/docs/faq/models", "type": "primary" }, { "label": "MiniMax-M2.7 Hugging Face checkpoint metadata", "url": "https://huggingface.co/api/models/MiniMaxAI/MiniMax-M2.7", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/minimax/MiniMax-M2.7-highspeed.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.7", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Highspeed is documented as the faster M2.7 API version with identical results. 230B total is NVIDIA’s nominal M2.7 deployment specification; approximately 10B active is explicitly disclosed for M2.7 by MiniMax’s later technical report. Public M2.7 weights are approximately 228.69B, excluding 13,880,064 FP8 inverse scales; 15,872 routing-bias values are stored but their original M2.7 training status is unresolved. Neither checkpoint storage nor the M2-only 229.9B/9.8B counts establish an exact hosted highspeed census. The shared weights use a non-commercial license. Context 204800 is verified for Vercel’s MiniMax highspeed route. MiniMax’s notice routes subscription calls to M3.1 Flash from October 19, 2026 and schedules existing pay-as-you-go retirement for November 18, although its retired-model table already labels highspeed retired; endpoint operation was not tested.", "field_status": {}, "audited_at": "2026-10-26T22:37:16.353893Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/306.json" }, { "id": "openai/gpt-5.4-mini", "slug": "openai--gpt-5.4-mini", "name": "GPT-5.4 Mini", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5.4", "description": "OpenAI model for coding and computer use, with configurable reasoning, text and image inputs, and tool calling.", "release_date": "2026-03-17", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 400000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", "source_type": "primary", "sources": [ { "label": "Introducing GPT-5.4 mini and nano", "url": "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", "type": "primary" }, { "label": "GPT-5.4 Mini API model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.4-mini.md", "type": "primary" }, { "label": "GPT-5.4 Thinking System Card, Appendix: GPT-5.4 mini", "url": "https://deploymentsafety.openai.com/gpt-5-4-thinking/gpt-5-4-thinking.pdf", "type": "primary" }, { "label": "Investigating the consequences of accidentally grading CoT during RL", "url": "https://alignment.openai.com/accidental-cot-grading/", "type": "primary" }, { "label": "Self-replicating prompt injections exist", "url": "https://alignment.openai.com/misalignment-reports/self-replicating-prompt-injections-exist", "type": "primary" }, { "label": "GPT-5.5 System Card", "url": "https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf", "type": "primary" }, { "label": "GPT-5.6 System Card", "url": "https://deploymentsafety.openai.com/gpt-5-6/gpt-5-6.pdf", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201.md", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "OpenAI API changelog", "url": "https://developers.openai.com/api/docs/changelog.md", "type": "primary" }, { "label": "Model Release Notes", "url": "https://help.openai.com/en/articles/9624314-model-release-notes", "type": "primary" }, { "label": "OpenAI API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "OpenAI model inventory", "url": "https://developers.openai.com/api/docs/models/all.md", "type": "primary" }, { "label": "GPT-5.4 Thinking System Card — GPT-5.4 mini appendix", "url": "https://deploymentsafety.openai.com/gpt-5-4-thinking/appendix-gpt-5.4-mini", "type": "primary" }, { "label": "Incompressible Knowledge Probes: Estimating Black-Box LLM Parameter Counts via Factual Capacity, v2", "url": "https://arxiv.org/html/2604.24827v2", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5.4-mini.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released March 17, 2026 in the API and Codex; a March 18 release note describes the ChatGPT rollout. Current default API snapshot is gpt-5.4-mini-2026-03-17. The 400,000-token context has separate 272,000-token input and 128,000-token output maxima. Reasoning supports none, low, medium, high and xhigh. PDF denotes API extraction into text and page images; image generation uses a hosted tool. Transformer is a qualified inference from exact mini prompt-caching support and OpenAI pre-GPT-5.6 KV-cache documentation. Detailed topology and total/active parameter counts remain undisclosed; calibrated effective-knowledge estimates are not model specifications. No exact API deprecation notice was found; current endpoint operation was not tested.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T16:57:21.211102Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/431.json" }, { "id": "openai/gpt-5.4-nano", "slug": "openai--gpt-5.4-nano", "name": "GPT-5.4 nano", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5.4", "description": "GPT-5.4 model for classification, data extraction and ranking, with text and image input.", "release_date": "2026-03-17", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 400000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", "source_type": "primary", "sources": [ { "label": "Introducing GPT-5.4 mini and nano", "url": "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", "type": "primary" }, { "label": "GPT-5.4 nano API model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.4-nano.md", "type": "primary" }, { "label": "OpenAI API changelog", "url": "https://developers.openai.com/api/docs/changelog.md", "type": "primary" }, { "label": "OpenAI API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201.md", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "Public Summary of Training Content for GPT-5.4 nano", "url": "https://cdn.openai.com/pdf/gpt-5-4-nano-eu-ai-act-public-summary-of-training-content.pdf", "type": "primary" }, { "label": "GPT-5.4 Thinking System Card", "url": "https://deploymentsafety.openai.com/gpt-5-4-thinking/gpt-5-4-thinking.pdf", "type": "primary" }, { "label": "GPT-5.5 System Card", "url": "https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf", "type": "primary" }, { "label": "GPT-5.6 System Card", "url": "https://deploymentsafety.openai.com/gpt-5-6/gpt-5-6.pdf", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "OpenAI model inventory", "url": "https://developers.openai.com/api/docs/models/all.md", "type": "primary" }, { "label": "Much Ado about Prompting: LLM Classification of Text Messages from Experiments", "url": "https://can-celebi.github.io/llmClassification.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5.4-nano.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Public API launch and default snapshot are March 17, 2026. A July training summary separately lists Union-market placement March 12; this unresolved date/scope discrepancy does not override the explicit public launch. Context is 400,000 tokens, with 272,000 maximum input and 128,000 maximum output. PDF denotes API extraction into text and page images, not native PDF processing. Transformer is a qualified broad inference from exact-model prompt caching and OpenAI pre-GPT-5.6 KV documentation; detailed topology remains undisclosed. Total and active parameter counts are unpublished in the checked OpenAI sources; the June 2026 original comparison by Celebi/Penczynski explicitly leaves both blank. No prior-model size is inherited, and behavioral capacity estimates are not treated as model parameters. Deprecated October 1, 2026, with API shutdown scheduled April 1, 2027; endpoint operation was not tested.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T16:57:26.871933Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/432.json" }, { "id": "mistral/mistral-small-2603", "slug": "mistral--mistral-small-2603", "name": "Mistral Small 4", "developer": "Mistral", "developer_id": "mistral", "family": "mistral-small", "description": "Open-weight multimodal mixture-of-experts model for instruction following, coding and configurable reasoning, with native function calling.", "release_date": "2026-03-16", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 119, "active_parameters_b": 6.5, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://docs.mistral.ai/models/mistral-small-4-0-26-03", "source_type": "primary", "sources": [ { "label": "Mistral Small 4 official model documentation", "url": "https://docs.mistral.ai/models/mistral-small-4-0-26-03", "type": "primary" }, { "label": "Mistral-Small-4-119B-2603 model card, pinned revision", "url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/raw/a11f36bebf709121056b1dbcc943d1c6afbe494d/README.md", "type": "weights" }, { "label": "Mistral Small 4 native parameters, pinned revision", "url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/raw/a11f36bebf709121056b1dbcc943d1c6afbe494d/params.json", "type": "weights" }, { "label": "Mistral Small 4 Transformers configuration, pinned revision", "url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603/raw/a11f36bebf709121056b1dbcc943d1c6afbe494d/config.json", "type": "weights" }, { "label": "Introducing Mistral Small 4", "url": "https://mistral.ai/news/mistral-small-4/", "type": "primary" }, { "label": "Official Mistral Small 4 model schema, pinned documentation revision", "url": "https://raw.githubusercontent.com/mistralai/platform-docs-public/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/schema/models/models/mistral-small-4-0-26-03.ts", "type": "primary" }, { "label": "Official model weights table column labels, pinned documentation revision", "url": "https://raw.githubusercontent.com/mistralai/platform-docs-public/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/app/%5Blocale%5D/(docs)/models/model-cards/%5Bslug%5D/components/weights-table.tsx", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/mistral-small-2603.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Mistral-Small-4-119B-2603", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "119B total and 6.5B active are official nominal counts from the current card and labeled documentation table. The launch announcement instead reports 6B active, or 8B including embedding/output layers; those conventions are not reconciled. Context is the card-supported 262,144-token serving length (256k); the native 1,048,576 positional setting is not a supported-context claim, and the API documentation does not spell out its exact endpoint ceiling.", "field_status": {}, "audited_at": "2026-10-26T23:33:37.638426Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/344.json" }, { "id": "nvidia/nemotron-voicechat", "slug": "nvidia--nemotron-voicechat", "name": "Nemotron 3 VoiceChat", "developer": "NVIDIA", "developer_id": "nvidia", "family": "nemotron", "description": "Early-access speech-to-speech model combining a Fast Conformer encoder, Nemotron Nano v2 language backbone and speech decoder for simultaneous listening and speaking.", "release_date": "2026-03-16", "date_precision": "day", "open_weights": false, "license": "NVIDIA Software and Model Evaluation License", "parameters_b": 12, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "hybrid-ssm-transformer", "context_window": null, "modalities": { "input": [ "text", "audio" ], "output": [ "text", "audio" ] }, "reasoning": false, "tool_call": null, "source_url": "https://build.nvidia.com/nvidia/nemotron-voicechat/modelcard", "source_type": "primary", "sources": [ { "label": "NVIDIA Nemotron 3 VoiceChat model card", "url": "https://build.nvidia.com/nvidia/nemotron-voicechat.md", "type": "primary" }, { "label": "NVIDIA Nemotron 3 VoiceChat early access", "url": "https://developer.nvidia.com/nemotron-voicechat-early-access", "type": "primary" }, { "label": "NVIDIA Software and Model Evaluation License", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-software-and-model-evaluation-license/", "type": "primary" }, { "label": "NVIDIA March 2026 VoiceChat launch details", "url": "https://developer.nvidia.com/blog/building-nvidia-nemotron-3-agents-for-reasoning-multimodal-rag-voice-and-safety/", "type": "primary" }, { "label": "NVIDIA public VoiceChat weight inventory", "url": "https://huggingface.co/api/models?author=nvidia&search=VoiceChat&limit=100", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/nvidia/nemotron-voicechat.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "March 2026 early-access Nemotron 3 VoiceChat: NVIDIA reports 12B total parameters, including a named 9B Nano v2 language backbone. An active count is not disclosed. The later public 11B NemotronLabs release is distinct. Training audio context is up to two minutes; a token limit and supported tool calling remain unverified. Exact 12B weights remain restricted to the qualified-developer early-access program. Classified closed because no public download of the exact 12B early-access checkpoint was found.", "field_status": { "active_parameters_b": "undisclosed", "context_window": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-28T08:03:07.166394Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/394.json" }, { "id": "zhipuai/glm-5-turbo", "slug": "zhipuai--glm-5-turbo", "name": "GLM-5-Turbo", "developer": "Z.ai", "developer_id": "zhipuai", "family": "glm", "description": "Text model trained for OpenClaw workflows, with optional reasoning and tool calling.", "release_date": "2026-03-15", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 744, "active_parameters_b": 40, "parameters_source": "primary_source", "architecture": null, "context_window": 202752, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://x.com/Zai_org/status/2033221428640674015", "source_type": "primary", "sources": [ { "label": "Original GLM-5-Turbo launch announcement", "url": "https://x.com/Zai_org/status/2033221428640674015", "type": "primary" }, { "label": "Developer explanation of closed distribution", "url": "https://x.com/Zai_org/status/2033221440267280696", "type": "primary" }, { "label": "Official model and product release history", "url": "https://docs.bigmodel.cn/cn/update/new-releases.md", "type": "primary" }, { "label": "GLM-5-Turbo native API guide", "url": "https://docs.z.ai/guides/llm/glm-5-turbo.md", "type": "primary" }, { "label": "Exact Z.AI provider route metadata", "url": "https://openrouter.ai/api/v1/models/z-ai/glm-5-turbo/endpoints", "type": "provider" }, { "label": "Z.ai API terms and model ownership", "url": "https://docs.z.ai/legal-agreement/terms-of-use.md", "type": "primary" }, { "label": "Official GLM-5 family download table", "url": "https://github.com/zai-org/GLM-5/blob/main/README.md", "type": "primary" }, { "label": "GLM-5 technical report", "url": "https://arxiv.org/html/2602.15763v1", "type": "primary" }, { "label": "IndexCache original research report", "url": "https://arxiv.org/html/2603.12201v1", "type": "primary" }, { "label": "GLM-5.1 later official card", "url": "https://huggingface.co/zai-org/GLM-5.1/raw/main/README.md", "type": "primary" }, { "label": "GLM-5.2 later official card", "url": "https://huggingface.co/zai-org/GLM-5.2/raw/main/README.md", "type": "primary" }, { "label": "GLM-5.3 later official card", "url": "https://huggingface.co/zai-org/GLM-5.3/raw/main/README.md", "type": "primary" }, { "label": "GLM-5.3-Flash later official card", "url": "https://huggingface.co/zai-org/GLM-5.3-Flash/raw/main/README.md", "type": "primary" }, { "label": "Original Chinese GLM-5-Turbo launch and training explanation", "url": "https://www.zhipuai.cn/zh/research/155", "type": "primary" }, { "label": "Exact official GLM-5 base model card", "url": "https://huggingface.co/zai-org/GLM-5/raw/main/README.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/zhipuai/glm-5-turbo.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released March 15, 2026 UTC (March 16 in Beijing). Z.ai explicitly calls Turbo a GLM-5 variant optimized for OpenClaw. Nominal 744B total / 40B active are inferred from that named base; no size-changing modification was found in original launch/training sources. The base report includes MTP but excludes embeddings/output; Turbo has no separate tensor census. Experimental closed-source distribution; exact topology remains undisclosed. The native Z.AI route specifies 202752 context and a separate 131072-token output cap.", "field_status": { "architecture": "undisclosed" }, "audited_at": "2026-10-27T18:37:05.981826+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/558.json" }, { "id": "nvidia/nemotron-3-super-120b-a12b", "slug": "nvidia--nemotron-3-super-120b-a12b", "name": "Nemotron 3 Super 120B A12B", "developer": "NVIDIA", "developer_id": "nvidia", "family": "nemotron", "description": "NVIDIA text model with Mamba-2, attention and LatentMoE layers, configurable reasoning, tool calling and a 1M-token context window.", "release_date": "2026-03-11", "date_precision": "day", "open_weights": true, "license": "NVIDIA Nemotron Open Model License", "parameters_b": 123.611012096, "active_parameters_b": 12.7, "parameters_source": "primary_source", "architecture": "hybrid-ssm-transformer-moe", "context_window": 1048576, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://developer.nvidia.com/blog/introducing-nemotron-3-super-an-open-hybrid-mamba-transformer-moe-for-agentic-reasoning/", "source_type": "primary", "sources": [ { "label": "NVIDIA launch announcement", "url": "https://developer.nvidia.com/blog/introducing-nemotron-3-super-an-open-hybrid-mamba-transformer-moe-for-agentic-reasoning/", "type": "primary" }, { "label": "Exact post-trained model card", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16/blob/2dc98e2afe4face0e4ce40972a915c45368bd34a/README.md", "type": "weights" }, { "label": "NVIDIA technical report", "url": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Super-Technical-Report.pdf", "type": "primary" }, { "label": "Checkpoint configuration", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16/blob/2dc98e2afe4face0e4ce40972a915c45368bd34a/config.json", "type": "weights" }, { "label": "Checkpoint tensor index", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16/blob/2dc98e2afe4face0e4ce40972a915c45368bd34a/model.safetensors.index.json", "type": "weights" }, { "label": "NVIDIA model implementation", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16/blob/2dc98e2afe4face0e4ce40972a915c45368bd34a/modeling_nemotron_h.py", "type": "weights" }, { "label": "MTP sharing implementation", "url": "https://github.com/vllm-project/vllm/blob/f235b89bdf62b5fc6356fc8d5e9d59572e48b145/vllm/model_executor/models/nemotron_h_mtp.py", "type": "primary" }, { "label": "NVIDIA Nemotron license", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-nemotron-open-model-license/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/nvidia/nemotron-3-super-120b-a12b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete BF16 checkpoint: 123.611012096B learned parameters, comprising a 120.66868736B main model and 2.942324736B MTP module. MTP shares the target input embedding and output head; neither is counted twice. Excludes 20992 non-parameter router-buffer elements. Active 12.7B is NVIDIA's nominal main-forward figure including embeddings, not an exact count or the cost of MTP drafting. The card's 256K configuration is a memory-saving default; supported context is 1048576. Evidence is for the original post-trained release, not Base, separate MTPv2 or 3.5 Super.", "parameter_components_b": { "main_model": 120.66868736, "mtp": 2.942324736 }, "active_parameters_scope": "main_model", "field_status": {}, "audited_at": "2026-10-27T00:43:35.249774+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/383.json" }, { "id": "rekaai/reka-edge", "slug": "rekaai--reka-edge", "name": "Reka Edge", "developer": "Reka AI", "developer_id": "rekaai", "family": "reka", "description": "Vision-language model for image and video understanding, object grounding and tool calling, with a dense language decoder and ConvNeXt V2 vision encoder.", "release_date": "2026-03-11", "date_precision": "day", "open_weights": true, "license": "Reka Edge Business Source License 1.1", "parameters_b": 7.128509568, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 16384, "modalities": { "input": [ "image", "text", "video" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://reka.ai/labs/research/reka-edge-frontier-level-edge-intelligence-for-physical-ai", "source_type": "primary", "sources": [ { "label": "Reka Edge: Frontier-Level Edge Intelligence for Physical AI — March 11, 2026", "url": "https://reka.ai/labs/research/reka-edge-frontier-level-edge-intelligence-for-physical-ai", "type": "primary" }, { "label": "Reka Edge 2603 pinned model card", "url": "https://huggingface.co/RekaAI/reka-edge-2603/blob/492c81c225fbf5f3263a8245b00827721b119a13/README.md", "type": "primary" }, { "label": "Reka Edge 2603 pinned configuration", "url": "https://huggingface.co/RekaAI/reka-edge-2603/blob/492c81c225fbf5f3263a8245b00827721b119a13/config.json", "type": "primary" }, { "label": "Selected Yasa2 multimodal implementation", "url": "https://huggingface.co/RekaAI/reka-edge-2603/blob/492c81c225fbf5f3263a8245b00827721b119a13/modeling_yasa2.py", "type": "primary" }, { "label": "Yasa2 configuration defaults", "url": "https://huggingface.co/RekaAI/reka-edge-2603/blob/492c81c225fbf5f3263a8245b00827721b119a13/configuration_yasa2.py", "type": "primary" }, { "label": "Complete checkpoint index and all three pinned BF16 shard headers", "url": "https://huggingface.co/RekaAI/reka-edge-2603/blob/492c81c225fbf5f3263a8245b00827721b119a13/model.safetensors.index.json", "type": "primary" }, { "label": "Reka Edge 2603 Business Source License 1.1", "url": "https://huggingface.co/RekaAI/reka-edge-2603/blob/492c81c225fbf5f3263a8245b00827721b119a13/LICENSE", "type": "primary" }, { "label": "Reka Edge 2603 chat template", "url": "https://huggingface.co/RekaAI/reka-edge-2603/blob/492c81c225fbf5f3263a8245b00827721b119a13/chat_template.jinja", "type": "primary" }, { "label": "Reka API model table and exact Edge2603 example", "url": "https://docs.reka.ai/chat/models", "type": "primary" }, { "label": "Reka Gateway reasoning documentation", "url": "https://developer.reka.ai/chat/reasoning", "type": "primary" }, { "label": "Reka EdgeQ, On-Device — September 22, 2026", "url": "https://reka.ai/labs/research/reka-edgeq", "type": "primary" }, { "label": "Reka Edge 2603 pinned Hugging Face parameter metadata — October 27 recheck", "url": "https://huggingface.co/api/models/RekaAI/reka-edge-2603/revision/492c81c225fbf5f3263a8245b00827721b119a13", "type": "primary" }, { "label": "Reka Labs current Edge model specification — October 27 recheck", "url": "https://reka.ai/labs", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/rekaai/reka-edge.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "March 2026 reka-edge-2603, distinct from the earlier Edge series. Reka publishes a nominal 7B total; the exact pinned BF16 count is 7.128509568B: 6.442717184B language/head, 0.65747264B vision encoder and 0.028319744B projector. Fresh Hub metadata confirms the prior complete 673-tensor reconciliation. The total includes 5,632 pooling-norm parameters bypassed in generation. No supported active-count convention is published; vision encoding runs on visual prefill. Context 16384 uses YaRN from 8192 positions. The model has no dedicated reasoning mode. Model materials use Reka’s Business Source License 1.1 with additional commercial/production/hosted-service restrictions; the stated Apache 2.0 change occurs two years after release.", "parameter_components_b": { "language_with_head": 6.442717184, "vision_encoder": 0.65747264, "vision_projector": 0.028319744 }, "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T17:42:35.860630Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/478.json" }, { "id": "xiaomi/mimo-v2-omni", "slug": "xiaomi--mimo-v2-omni", "name": "MiMo-V2-Omni", "developer": "Xiaomi", "developer_id": "xiaomi", "family": "mimo", "description": "Multimodal model with text, image, audio and video input, text output, and structured tool calling.", "release_date": "2026-03-11", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 262144, "modalities": { "input": [ "text", "image", "audio", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openrouter.ai/openrouter/healer-alpha/pricing", "source_type": "primary", "sources": [ { "label": "MiMo-V2-Omni launch", "url": "https://mimo.xiaomi.com/mimo-v2-omni", "type": "primary" }, { "label": "Model Release changelog", "url": "https://mimo.mi.com/docs/en-US/updates/model", "type": "primary" }, { "label": "Model Deprecation", "url": "https://mimo.mi.com/docs/updates/deprecate", "type": "primary" }, { "label": "Xiaomi MiMo Service Agreement", "url": "https://mimo.mi.com/docs/quick-start/terms/user-agreement", "type": "primary" }, { "label": "Service Agreement public document content", "url": "https://mimo.mi.com/static/1995.8b548d0e0b571f7c.chunk.js", "type": "primary" }, { "label": "OpenClaw v2026.3.23 native Xiaomi provider catalog", "url": "https://raw.githubusercontent.com/openclaw/openclaw/v2026.3.23/extensions/xiaomi/provider-catalog.ts", "type": "primary" }, { "label": "MiMo-V2-Omni historical provider record", "url": "https://openrouter.ai/xiaomi/mimo-v2-omni", "type": "primary" }, { "label": "MiMo-V2-Omni endpoint metadata", "url": "https://openrouter.ai/api/v1/models/xiaomi/mimo-v2-omni/endpoints", "type": "primary" }, { "label": "Healer Alpha historical provider listing and MiMo-V2-Omni identity notice", "url": "https://openrouter.ai/openrouter/healer-alpha/pricing", "type": "primary" }, { "label": "Healer Alpha historical uptime page", "url": "https://openrouter.ai/openrouter/healer-alpha/uptime", "type": "primary" }, { "label": "Healer Alpha endpoint metadata", "url": "https://openrouter.ai/api/v1/models/openrouter/healer-alpha/endpoints", "type": "primary" }, { "label": "P3D-Bench: Benchmarking MLLMs for Parametric 3D Generation and Structural Reasoning, Table 2", "url": "https://arxiv.org/html/2606.11152v1", "type": "primary" }, { "label": "MiMo-V2.5 original launch and model table", "url": "https://mimo.xiaomi.com/mimo-v2-5", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/xiaomi/mimo-v2-omni.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Public preview began March 11, 2026 as Healer Alpha: OpenRouter explicitly identifies it as an early testing version of MiMo-V2-Omni. Xiaomi dates the named release March 18. The preview-to-launch relation does not establish identical weights; capability specifications here describe the named original release. Xiaomi states 256K context; contemporaneous native-Xiaomi integration and OpenRouter document 262144 tokens. Reasoning is documented by the March 23 OpenClaw native integration. Total and active parameters remain undisclosed; no counted base or unchanged lineage is identified for Omni. Later V2.5/Flash/Pro sizes do not establish Omni counts. Dedicated image/video/audio encoders share one backbone, whose network class remains undisclosed. Native inputs exclude PDF; external tools produce any media in agent demos. Xiaomi routed this identifier to V2.5 on June 1, 2026 and retired it June 30 (00:00 UTC+8); original specifications are retained, not replacement specifications.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T18:36:18.285107+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/538.json" }, { "id": "xiaomi/mimo-v2-pro", "slug": "xiaomi--mimo-v2-pro", "name": "MiMo-V2-Pro", "developer": "Xiaomi", "developer_id": "xiaomi", "family": "mimo", "description": "Xiaomi text model for reasoning, coding and tool use, with hybrid attention and a 1M-token context window.", "release_date": "2026-03-11", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 1020, "active_parameters_b": 42, "parameters_source": "primary_source", "architecture": "hybrid-attention", "context_window": 1048576, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://mimo.xiaomi.com/mimo-v2-pro", "source_type": "primary", "sources": [ { "label": "Xiaomi original MiMo-V2-Pro launch", "url": "https://mimo.xiaomi.com/mimo-v2-pro", "type": "primary" }, { "label": "Xiaomi model release log", "url": "https://mimo.mi.com/docs/en-US/updates/model", "type": "primary" }, { "label": "Xiaomi later comparison: exact MiMo-V2-Pro 1.02T/42B column", "url": "https://mimo.xiaomi.com/mimo-v2-5-pro", "type": "primary" }, { "label": "OpenRouter Hunter Alpha preview release and identity", "url": "https://openrouter.ai/openrouter/hunter-alpha", "type": "primary" }, { "label": "OpenRouter exact V2-Pro modality metadata", "url": "https://openrouter.ai/api/v1/models/xiaomi/mimo-v2-pro/endpoints", "type": "primary" }, { "label": "OpenRouter historical V2-Pro context limit", "url": "https://openrouter.ai/xiaomi/mimo-v2-pro-20260318", "type": "primary" }, { "label": "Xiaomi model deprecation schedule", "url": "https://mimo.mi.com/docs/en-US/updates/deprecate", "type": "primary" }, { "label": "Xiaomi service agreement document content", "url": "https://mimo.mi.com/static/1995.8b548d0e0b571f7c.chunk.js", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/xiaomi/mimo-v2-pro.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Release date is the March 11, 2026 public Hunter Alpha preview, explicitly identified by Xiaomi and OpenRouter as an early MiMo-V2-Pro build; the named API launch followed March 18. Parameters are Xiaomi’s nominal 1.02T/42B from a later comparison column explicitly labeled MiMo-V2-Pro, consistent with the original >1T/42B disclosure; they are not exact tensor counts. The original uses 7:1 sliding-window/global attention and text input/output. V2.5-Pro’s architecture, MIT license and open weights are not transferred. Xiaomi routed the old API name to V2.5-Pro June 1 and deprecated it June 30, 2026.", "field_status": {}, "audited_at": "2026-10-27T05:48:55Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/539.json" }, { "id": "xai/grok-4.20-0309-non-reasoning", "slug": "xai--grok-4.20-0309-non-reasoning", "name": "Grok 4.20 (Non-Reasoning)", "developer": "xAI", "developer_id": "xai", "family": "grok", "description": "Non-reasoning Grok 4.20 model with text and image input, text output, function calling, and a 1 million token context window.", "release_date": "2026-03-10", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 500, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": null, "context_window": 1000000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://docs.x.ai/developers/release-notes#grok-420-and-grok-420-multi-agent-are-live", "source_type": "primary", "sources": [ { "label": "Grok 4.20 0309 Non Reasoning", "url": "https://docs.x.ai/developers/models/grok-4.20-0309-non-reasoning", "type": "primary" }, { "label": "Release notes: Grok 4.20 and Grok 4.20 Multi-agent are live", "url": "https://docs.x.ai/developers/release-notes#grok-420-and-grok-420-multi-agent-are-live", "type": "primary" }, { "label": "Responses API reference", "url": "https://docs.x.ai/developers/rest-api-reference/inference/responses", "type": "primary" }, { "label": "Grok 4.20 System Card, April 7, 2026", "url": "https://data.x.ai/2026-04-07-grok-4-20-model-card.pdf", "type": "primary" }, { "label": "Files", "url": "https://docs.x.ai/developers/files", "type": "primary" }, { "label": "Enterprise Terms of Service, June 27, 2025", "url": "https://x.ai/legal/terms-of-service-enterprise/previous-2025-06-27", "type": "primary" }, { "label": "Elon Musk: Grok 4.20 has 0.5T total parameters, April 9, 2026", "url": "https://cdn.syndication.twimg.com/tweet-result?id=2042123561666855235&lang=en&token=1", "type": "primary" }, { "label": "Elon Musk confirms Grok 4.2 is 0.5T, April 8, 2026", "url": "https://cdn.syndication.twimg.com/tweet-result?id=2041893772683047215&lang=en&token=1", "type": "primary" }, { "label": "xAI Grok 4.20 model modes", "url": "https://docs.oracle.com/en-us/iaas/Content/generative-ai/xai-grok-4-20.htm", "type": "provider" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/xai/grok-4.20-0309-non-reasoning.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "xAI dates the API launch March 10, 2026; the 0309 identifier alone does not establish a March 9 release. Nominal approximately 500B total parameters comes from Elon Musk's explicit Grok 4.20 total-size disclosure, applied to its documented non-reasoning mode. Active parameters and neural architecture remain undisclosed. PDF attachments use server-side search, not a direct model modality.", "field_status": { "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T18:26:43.933359+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/529.json" }, { "id": "xai/grok-4.20-0309-reasoning", "slug": "xai--grok-4.20-0309-reasoning", "name": "Grok 4.20 (Reasoning)", "developer": "xAI", "developer_id": "xai", "family": "grok", "description": "Grok 4.20 reasoning API model with text and image inputs, text output, and function calling.", "release_date": "2026-03-10", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 500, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": null, "context_window": 1000000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://docs.x.ai/developers/models/grok-4.20-0309-reasoning", "source_type": "primary", "sources": [ { "label": "Exact Grok 4.20 reasoning API model", "url": "https://docs.x.ai/developers/models/grok-4.20-0309-reasoning", "type": "primary" }, { "label": "March 10 API launch", "url": "https://docs.x.ai/developers/release-notes#grok-420-and-grok-420-multi-agent-are-live", "type": "primary" }, { "label": "Grok 4.20 System Card", "url": "https://data.x.ai/2026-04-07-grok-4-20-model-card.pdf", "type": "primary" }, { "label": "Document attachment processing", "url": "https://docs.x.ai/developers/model-capabilities/files/chat-with-files", "type": "primary" }, { "label": "xAI enterprise service terms", "url": "https://x.ai/legal/terms-of-service-enterprise/previous-2025-06-27", "type": "primary" }, { "label": "Founder confirms Grok 4.20 nominal size", "url": "https://x.com/elonmusk/status/2041893772683047215", "type": "primary" }, { "label": "Founder clarifies 0.5T total parameters", "url": "https://x.com/elonmusk/status/2042123561666855235", "type": "primary" }, { "label": "Grok 4.20 reasoning and non-reasoning modes", "url": "https://docs.oracle.com/en-us/iaas/Content/generative-ai/xai-grok-4-20.htm", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/xai/grok-4.20-0309-reasoning.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The 0309 identifier denotes the snapshot; xAI dates the API launch March 10, 2026. The founder disclosed nominal 500B total for Grok 4.20, whose reasoning mode is linked to the dated identifier by official documentation. This rounded model size is not a separate tensor census of the snapshot. Active parameters and neural architecture remain undisclosed. Native context is 1,000,000 tokens; provider limits can differ. PDF attachments use document search.", "field_status": { "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T18:28:16.195933+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/530.json" }, { "id": "sarvam/sarvam-30b", "slug": "sarvam--sarvam-30b", "name": "Sarvam 30B", "developer": "Sarvam AI", "developer_id": "sarvam", "family": "sarvam", "description": "Mixture-of-experts language model trained from scratch for Indian-language conversation, reasoning, coding and tool use.", "release_date": "2026-03-06", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 32.152648064, "active_parameters_b": 2.4, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.sarvam.ai/blogs/sarvam-30b-105b", "source_type": "primary", "sources": [ { "label": "Sarvam-30B model card", "url": "https://huggingface.co/sarvamai/sarvam-30b/blob/071ae95e933605ca1104a6b4524a36a98488efa4/README.md", "type": "primary" }, { "label": "Exact published checkpoint configuration", "url": "https://huggingface.co/sarvamai/sarvam-30b/blob/071ae95e933605ca1104a6b4524a36a98488efa4/config.json", "type": "primary" }, { "label": "Exact model implementation", "url": "https://huggingface.co/sarvamai/sarvam-30b/blob/071ae95e933605ca1104a6b4524a36a98488efa4/modeling_sarvam_moe.py", "type": "primary" }, { "label": "Complete original F32 checkpoint index", "url": "https://huggingface.co/sarvamai/sarvam-30b/blob/071ae95e933605ca1104a6b4524a36a98488efa4/model.safetensors.index.json", "type": "primary" }, { "label": "Shipped chat template", "url": "https://huggingface.co/sarvamai/sarvam-30b/blob/071ae95e933605ca1104a6b4524a36a98488efa4/chat_template.jinja", "type": "primary" }, { "label": "Open-Sourcing Sarvam 30B and 105B, March 6, 2026", "url": "https://www.sarvam.ai/blogs/sarvam-30b-105b", "type": "primary" }, { "label": "Sarvam 30B archived API specification", "url": "https://docs.sarvam.ai/api/getting-started/models/sarvam-30b", "type": "primary" }, { "label": "August 18, 2026 API changelog", "url": "https://docs.sarvam.ai/changelog/2026/8/18", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/sarvam/sarvam-30b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete learned checkpoint: 32.152648064B parameters, including separate input and output vocabulary matrices; the nominal 30B label corresponds approximately to non-embedding weights. Active 2.4B is the author-rounded non-embedding count (configuration-derived 2.373089664B); it excludes both vocabulary matrices. The 26 original F32 shards contain 2304 additional frozen router-bias values, excluded as auxiliary balancing state. Checkpoint context is configured at 131072; the retired Sarvam API documented 64K, and published evaluations used 65536. March 6, 2026 is the explicit weights/API release; February 18 was technical disclosure and earlier Samvaad use is undated. Sarvam API retirement was documented on August 18, 2026; downloadable weights remain available.", "parameter_components_b": { "input_embeddings": 1.073741824, "output_head": 1.073741824, "decoder_layers": 30.00516032, "final_norm": 0.000004096 }, "active_parameters_scope": "non_embedding", "active_parameters_by_phase_b": { "non_embedding": 2.4 }, "field_status": {}, "audited_at": "2026-10-27T04:32:25.172371Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/491.json" }, { "id": "openai/gpt-5.4", "slug": "openai--gpt-5.4", "name": "GPT-5.4", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5.4", "description": "OpenAI reasoning model for coding and general tasks, with text and image input, tool calling and computer use.", "release_date": "2026-03-05", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 1050000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-gpt-5-4/", "source_type": "primary", "sources": [ { "label": "Introducing GPT-5.4", "url": "https://openai.com/index/introducing-gpt-5-4/", "type": "primary" }, { "label": "GPT-5.4 API model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.4.md", "type": "primary" }, { "label": "GPT-5.4 Thinking System Card", "url": "https://deploymentsafety.openai.com/gpt-5-4-thinking/gpt-5-4-thinking.pdf", "type": "primary" }, { "label": "GPT-5.5 System Card", "url": "https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf", "type": "primary" }, { "label": "Investigating the consequences of accidentally grading CoT during RL", "url": "https://alignment.openai.com/accidental-cot-grading/", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201.md", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "OpenAI API changelog", "url": "https://developers.openai.com/api/docs/changelog.md", "type": "primary" }, { "label": "OpenAI API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "OpenAI model inventory", "url": "https://developers.openai.com/api/docs/models/all.md", "type": "primary" }, { "label": "Much Ado about Prompting: LLM Classification of Text Messages from Experiments", "url": "https://can-celebi.github.io/llmClassification.pdf", "type": "primary" }, { "label": "Incompressible Knowledge Probes: Estimating Black-Box LLM Parameter Counts via Factual Capacity", "url": "https://arxiv.org/html/2604.24827v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5.4.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released March 5, 2026 in ChatGPT as GPT-5.4 Thinking, Codex and the API; the API default snapshot is gpt-5.4-2026-03-05. The API context is 1,050,000 tokens with a 128,000-token maximum output; ChatGPT and Codex product limits differ. OpenAI documented an image-encoder fix on March 13. PDF means API extraction into text and page images, and hosted image generation is a tool rather than native model output. Transformer is a qualified broad inference from exact-model prompt caching support and OpenAI pre-GPT-5.6 KV-cache documentation. Detailed topology and total/active parameter counts remain undisclosed in the checked sources; a later original study also leaves both counts blank. Black-box knowledge-capacity estimates are not model weight counts. No exact-model API shutdown notice was found in the current deprecation page; live endpoint operation was not tested.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T16:55:45.720646Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/430.json" }, { "id": "openai/gpt-5.4-pro", "slug": "openai--gpt-5.4-pro", "name": "GPT-5.4 Pro", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5.4", "description": "OpenAI reasoning model using additional inference compute, with text and image input and tool calling.", "release_date": "2026-03-05", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1050000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-gpt-5-4/", "source_type": "primary", "sources": [ { "label": "Introducing GPT-5.4", "url": "https://openai.com/index/introducing-gpt-5-4/", "type": "primary" }, { "label": "GPT-5.4 Pro API model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.4-pro.md", "type": "primary" }, { "label": "GPT-5.4 Thinking System Card", "url": "https://deploymentsafety.openai.com/gpt-5-4-thinking/gpt-5-4-thinking.pdf", "type": "primary" }, { "label": "Introducing GPT-5.5", "url": "https://openai.com/index/introducing-gpt-5-5/", "type": "primary" }, { "label": "GeneBench: Assessing AI Agents for Multi-Stage Inference Problems in Genomics and Quantitative Biology", "url": "https://cdn.openai.com/pdf/6dc7175d-d9e7-4b8d-96b8-48fe5798cd5b/oai_genebench_benchmark.pdf", "type": "primary" }, { "label": "GPT-5.5 System Card", "url": "https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "OpenAI API changelog", "url": "https://developers.openai.com/api/docs/changelog.md", "type": "primary" }, { "label": "OpenAI API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "Prompt caching", "url": "https://developers.openai.com/api/docs/guides/prompt-caching.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "OpenAI model inventory", "url": "https://developers.openai.com/api/docs/models/all.md", "type": "primary" }, { "label": "Using GPT-5.4 — model, API, and feature updates", "url": "https://developers.openai.com/api/docs/guides/latest-model?model=gpt-5.4", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5.4-pro.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released March 5, 2026 in ChatGPT and the Responses API; default API snapshot gpt-5.4-pro-2026-03-05. API context is 1,050,000 tokens with 128,000 maximum output. Reasoning effort supports medium (default), high and xhigh. PDF is an API input format extracted into text and images; image generation is a hosted tool rather than native model output. Total and active parameter counts remain undisclosed after the October 27 recheck; the named GPT-5.4 version relationship supplies no published base count to inherit. Additional inference compute and behavioral capacity estimates do not establish parameter sizes. Neural architecture remains unverified. No exact API shutdown notice was found in the current deprecation page; ChatGPT availability and live API operation were not verified.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "unverified" }, "audited_at": "2026-10-27T16:59:28.498802Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/433.json" }, { "id": "google/gemini-3.1-flash-lite-preview", "slug": "google--gemini-3.1-flash-lite-preview", "name": "Gemini 3.1 Flash-Lite Preview", "developer": "Google", "developer_id": "google", "family": "gemini-flash-lite", "description": "March 2026 preview of Gemini 3.1 Flash-Lite with multimodal input, thinking and function calling.", "release_date": "2026-03-03", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-lite-preview", "source_type": "primary", "sources": [ { "label": "Gemini 3.1 Flash-Lite preview - Gemini API", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-lite-preview", "type": "primary" }, { "label": "Gemini API release notes", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "primary" }, { "label": "Gemini 3.1 Flash-Lite model card - March 2026", "url": "https://deepmind.google/models/model-cards/gemini-3-1-flash-lite/", "type": "primary" }, { "label": "Gemini 3 Pro Model Card - updated May 2026", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-Pro-Model-Card.pdf", "type": "primary" }, { "label": "Gemini 3.1 Flash-Lite Model Card - updated May 2026", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-1-Flash-Lite-Model-Card.pdf", "type": "primary" }, { "label": "Gemini 3.5 Flash-Lite Model Card - July 21, 2026", "url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "type": "primary" }, { "label": "Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Gemini 3.1 Flash-Lite: Built for intelligence at scale", "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-flash-lite/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemini-3.1-flash-lite-preview.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Preview launched March 3, 2026 and shut down May 25; stable gemini-3.1-flash-lite launched May 7. Specifications describe the retired preview. Its model card names Gemini 3 Pro as the base and architecture reference. Total and active parameter counts are undisclosed for the preview and its named base in the reviewed sources.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T12:58:04.093011+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/203.json" }, { "id": "openai/gpt-5.3-chat-latest", "slug": "openai--gpt-5.3-chat-latest", "name": "GPT-5.3 Chat", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5.3", "description": "API alias for GPT-5.3 Instant, accepting text, images and PDF files and producing text with function calling.", "release_date": "2026-03-03", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 128000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/gpt-5-3-instant/", "source_type": "primary", "sources": [ { "label": "GPT-5.3 Instant: Smoother, more useful everyday conversations", "url": "https://openai.com/index/gpt-5-3-instant/", "type": "primary" }, { "label": "GPT-5.3 Chat model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.3-chat-latest", "type": "primary" }, { "label": "GPT-5.3 Instant System Card", "url": "https://deploymentsafety.openai.com/gpt-5-3-instant/gpt-5-3-instant.pdf", "type": "primary" }, { "label": "GPT-5.5 Instant System Card", "url": "https://deploymentsafety.openai.com/gpt-5-5-instant/gpt-5-5-instant.pdf", "type": "primary" }, { "label": "Investigating the consequences of accidentally grading CoT during RL", "url": "https://alignment.openai.com/accidental-cot-grading/", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "Deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "OpenAI API changelog", "url": "https://developers.openai.com/api/docs/changelog.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201.md", "type": "primary" }, { "label": "Prompt caching", "url": "https://developers.openai.com/api/docs/guides/prompt-caching.md", "type": "primary" }, { "label": "GPT-5.5 Instant: smarter, clearer, and more personalized", "url": "https://openai.com/index/gpt-5-5-instant/", "type": "primary" }, { "label": "GPT-5.3 Instant System Card web page", "url": "https://deploymentsafety.openai.com/gpt-5-3-instant", "type": "primary" }, { "label": "GPT-5.3 Chat model documentation, Markdown", "url": "https://developers.openai.com/api/docs/models/gpt-5.3-chat-latest.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5.3-chat-latest.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Mutable alias for GPT-5.3 Instant: launched March 3, 2026 and updated March 16. The system card evaluates a February 26 version; its HTML March 2 label differs from the March 3 PDF and public launch. OpenAI lists API shutdown on August 10, 2026; ChatGPT replacement and legacy access followed a separate May 5 announcement. Technical fields describe documented capabilities, not current endpoint availability. Reasoning means limited internal chain of thought, not an exposed effort control. PDF is an API input format converted to text and images. Context remains 128,000; a conflicting 272,000 maximum-input line is not adopted. Total and active parameter counts are undisclosed in the checked original specifications and later exact-model comparisons; no counted-base lineage is established. Structural architecture remains unverified.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "unverified" }, "audited_at": "2026-10-27T16:52:02.490729Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/427.json" }, { "id": "alibaba/qwen3.5-0.8b", "slug": "alibaba--qwen3.5-0.8b", "name": "Qwen3.5 0.8B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Dense Qwen3.5 vision-language model with optional thinking and tool calling, intended for prototyping and task-specific fine-tuning.", "release_date": "2026-03-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.873438784, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/Qwen/Qwen3.5-0.8B", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3.5-0.8B", "type": "weights" }, { "label": "Official architecture configuration", "url": "https://huggingface.co/Qwen/Qwen3.5-0.8B/raw/main/config.json", "type": "primary" }, { "label": "Official checkpoint tensor header", "url": "https://huggingface.co/Qwen/Qwen3.5-0.8B/resolve/main/model.safetensors-00001-of-00001.safetensors", "type": "weights" }, { "label": "Official open-weight release announcement", "url": "https://github.com/QwenLM/Qwen3.8#news", "type": "primary" }, { "label": "Official repository parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen3.5-0.8B", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.5-0.8b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The official 0.8B label is a nominal language-model size. The full checkpoint contains 0.873438784B stored parameters: language model 0.752393024B, vision 0.100592896B, MTP 0.020452864B. Dense FFNs use hybrid Gated DeltaNet and gated attention. A separate active-parameter count is not disclosed.", "parameter_components_b": { "language_model": 0.752393024, "vision": 0.100592896, "mtp": 0.020452864 }, "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T09:42:02.197542Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/046.json" }, { "id": "alibaba/qwen3.5-2b", "slug": "alibaba--qwen3.5-2b", "name": "Qwen3.5 2B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Post-trained Qwen3.5 vision-language model with dense feed-forward layers, hybrid attention, optional thinking and tool calling.", "release_date": "2026-03-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 2.274069824, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/Qwen/Qwen3.5-2B", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3.5-2B", "type": "weights" }, { "label": "Qwen3.5-2B configuration", "url": "https://huggingface.co/Qwen/Qwen3.5-2B/raw/main/config.json", "type": "primary" }, { "label": "Official checkpoint repository metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen3.5-2B", "type": "primary" }, { "label": "Qwen3.5-2B Apache License", "url": "https://huggingface.co/Qwen/Qwen3.5-2B/blob/main/LICENSE", "type": "primary" }, { "label": "Official Qwen repository release news", "url": "https://github.com/QwenLM/Qwen3.8#news", "type": "primary" }, { "label": "Qwen3.5-2B checkpoint tensor index", "url": "https://huggingface.co/Qwen/Qwen3.5-2B/raw/main/model.safetensors.index.json", "type": "primary" }, { "label": "Alibaba Cloud Smart Studio supported training models", "url": "https://smartstudio.console.alibabacloud.com/guides/model-serving/docs/model-training/training", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.5-2b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Full checkpoint metadata reports 2.274069824B stored parameters, including vision and MTP; the card scopes nominal 2B to the language model. Dense FFNs use hybrid linear/full attention. Qwen has not separately disclosed an active count. Thinking is optional and off by default. Release date is the official open-weight announcement.", "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T09:44:40.221138Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/049.json" }, { "id": "alibaba/qwen3.5-4b", "slug": "alibaba--qwen3.5-4b", "name": "Qwen3.5 4B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Post-trained dense vision-language model with hybrid Gated DeltaNet and gated attention, switchable thinking, and tool calling.", "release_date": "2026-03-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 4.659865088, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.5-4b.toml", "source_type": "catalog", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3.5-4B", "type": "weights" }, { "label": "Qwen3.5-4B configuration", "url": "https://huggingface.co/Qwen/Qwen3.5-4B/blob/main/config.json", "type": "primary" }, { "label": "Qwen3.5-4B repository safetensors metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen3.5-4B", "type": "weights" }, { "label": "Official Qwen repository release news", "url": "https://github.com/QwenLM/Qwen3.8#news", "type": "primary" }, { "label": "Qwen3.5-4B LICENSE", "url": "https://huggingface.co/Qwen/Qwen3.5-4B/blob/main/LICENSE", "type": "primary" }, { "label": "Qwen3.5-4B checkpoint tensor index", "url": "https://huggingface.co/Qwen/Qwen3.5-4B/resolve/main/model.safetensors.index.json", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.5-4b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Full checkpoint: 4,659,865,088 stored parameters, including vision and auxiliary MTP weights. The 4B label is the nominal language-model size; active parameters are not separately disclosed by Qwen. Native context is 262,144; optional YaRN extension is 1,010,000. Release date denotes public weights.", "parameter_components_b": { "language_model": 4 }, "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T09:45:01.543132Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/053.json" }, { "id": "alibaba/qwen3.5-9b", "slug": "alibaba--qwen3.5-9b", "name": "Qwen3.5 9B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Open-weight multimodal Qwen model with hybrid attention, optional thinking, and tool use.", "release_date": "2026-03-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 9.653104368, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/Qwen/Qwen3.5-9B", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3.5-9B", "type": "weights" }, { "label": "Qwen3.5-9B configuration", "url": "https://huggingface.co/Qwen/Qwen3.5-9B/raw/main/config.json", "type": "primary" }, { "label": "Official Qwen release history", "url": "https://github.com/QwenLM/Qwen3.8", "type": "primary" }, { "label": "Checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen3.5-9B", "type": "primary" }, { "label": "Checkpoint tensor manifest", "url": "https://huggingface.co/Qwen/Qwen3.5-9B/raw/main/model.safetensors.index.json", "type": "primary" }, { "label": "Checkpoint license", "url": "https://huggingface.co/Qwen/Qwen3.5-9B/raw/main/LICENSE", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.5-9b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "9B is the nominal language-model size. Complete checkpoint: 9.653104368B parameters, comprising language model and LM head 8.953803264B, vision 0.456010480B and MTP 0.243290624B. Qwen does not separately specify active parameters. Native context is 262144; larger context requires extension configuration.", "parameter_components_b": { "language_model": 8.953803264, "vision": 0.45601048, "mtp": 0.243290624 }, "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T09:48:06.491Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/054.json" }, { "id": "alibaba/qwen3.5-9b-base", "slug": "alibaba--qwen3.5-9b-base", "name": "Qwen3.5 9B Base", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Pre-trained Qwen3.5 base checkpoint for fine-tuning and research, without instruction tuning", "release_date": "2026-03-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 9.653104368, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://x.com/Alibaba_Qwen/status/2028460046510965160", "source_type": "primary", "sources": [ { "label": "Exact Base model card", "url": "https://huggingface.co/Qwen/Qwen3.5-9B-Base/raw/main/README.md", "type": "model_card" }, { "label": "Exact Base configuration", "url": "https://huggingface.co/Qwen/Qwen3.5-9B-Base/raw/main/config.json", "type": "config" }, { "label": "Official checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen3.5-9B-Base", "type": "weights" }, { "label": "Official checkpoint tensor index", "url": "https://huggingface.co/Qwen/Qwen3.5-9B-Base/raw/main/model.safetensors.index.json", "type": "weights" }, { "label": "Exact Base tokenizer and template", "url": "https://huggingface.co/Qwen/Qwen3.5-9B-Base/raw/main/tokenizer_config.json", "type": "config" }, { "label": "Official Qwen release news", "url": "https://github.com/QwenLM/Qwen3.8#news", "type": "announcement" }, { "label": "Exact Base commit history", "url": "https://huggingface.co/api/models/Qwen/Qwen3.5-9B-Base/commits/main", "type": "repository" }, { "label": "Qwen-Scope dense backbone classification", "url": "https://qwen.ai/blog?id=qwen-scope", "type": "technical_report" }, { "label": "Qwen3.5-9B comparison tables", "url": "https://huggingface.co/Qwen/Qwen3.5-9B/raw/main/README.md", "type": "model_card" }, { "label": "Qwen-Scope backbone comparison table", "url": "https://arxiv.org/html/2605.11887v1", "type": "technical_report" }, { "label": "Qwen March 2 small-model and Base release announcement", "url": "https://x.com/Alibaba_Qwen/status/2028460046510965160", "type": "primary" }, { "label": "Tongyi Lab original weekly release announcement", "url": "https://tongyilab.substack.com/p/mar-6-2026-the-tongyi-weekly", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.5-9b-base.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3.5-9B-Base", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete checkpoint: 9.653104368B parameters, including 8.953803264B language model with output head (nominally 9B), 0.456010480B vision and 0.243290624B MTP. Active parameters are not separately disclosed. Dense FFNs accompany hybrid DeltaNet/full attention. Base weights are pretrained-only; generic card claims and template branches do not establish trained reasoning or tool capabilities. Qwen announced release of the small-series Base checkpoints on March 2, 2026. The February 26 repository creation and February 27 weight upload were preparation dates.", "parameter_components_b": { "language_model": 8.953803264, "vision": 0.45601048, "mtp": 0.243290624 }, "field_status": { "active_parameters_b": "undisclosed", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T19:29:18.594022Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/055.json" }, { "id": "google/gemini-3.1-flash-image", "slug": "google--gemini-3.1-flash-image", "name": "Gemini 3.1 flash image", "developer": "Google", "developer_id": "google", "family": "gemini", "description": null, "release_date": "2026-02-26", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 131072, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text", "image" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.google.dev/gemini-api/docs/changelog", "source_type": "primary", "sources": [ { "label": "Dated publisher API releases", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "documentation" }, { "label": "Publisher interleaved text and image capability", "url": "https://ai.google.dev/gemini-api/docs/image-generation", "type": "documentation" }, { "label": "Publisher model specifications", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-image", "type": "documentation" } ], "aliases": [ "google/gemini-3.1-flash-image-preview" ], "first_seen": "2026-10-29T10:31:17.277Z", "last_seen": "2026-10-29T10:31:17.277Z", "upstream_present": false, "availability": "public", "notes": "Generates interleaved text and images. Date records first preview availability; general availability followed 2026-05-28. Preview and stable serving names are retained as aliases, without evidence of a separate trained checkpoint. Context and capabilities describe current published model documentation; preview limits may have differed." }, { "id": "alibaba/qwen3.5-122b-a10b", "slug": "alibaba--qwen3.5-122b-a10b", "name": "Qwen3.5 122B-A10B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Qwen vision-language model for visual reasoning, documents, and agent tasks", "release_date": "2026-02-24", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 125.086497008, "active_parameters_b": 10, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.5-122b-a10b.toml", "source_type": "catalog", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3.5-122B-A10B", "type": "weights" }, { "label": "Official configuration", "url": "https://huggingface.co/Qwen/Qwen3.5-122B-A10B/raw/main/config.json", "type": "primary" }, { "label": "Official release timeline", "url": "https://github.com/QwenLM/Qwen3.8/blob/main/README.md", "type": "primary" }, { "label": "Official checkpoint metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen3.5-122B-A10B", "type": "weights" }, { "label": "Official tensor index", "url": "https://huggingface.co/Qwen/Qwen3.5-122B-A10B/raw/main/model.safetensors.index.json", "type": "weights" }, { "label": "Official Apache 2.0 license", "url": "https://huggingface.co/Qwen/Qwen3.5-122B-A10B/blob/main/LICENSE", "type": "primary" }, { "label": "Official publication history", "url": "https://huggingface.co/Qwen/Qwen3.5-122B-A10B/commits/main", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.5-122b-a10b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Full checkpoint: 125.086497008B tensor parameters, including vision and MTP weights. Qwen reports a nominal 122B language model with 10B active; active count applies to the language decoder. Native context is 262,144 tokens.", "active_parameters_scope": "language_decoder", "parameter_components_b": { "language_decoder": 122 }, "field_status": {}, "audited_at": "2026-10-26T14:47:13.960689Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/047.json" }, { "id": "alibaba/qwen3.5-27b", "slug": "alibaba--qwen3.5-27b", "name": "Qwen3.5 27B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Dense Qwen3.5 vision-language model with optional thinking and tool calling", "release_date": "2026-02-24", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 27.781427952, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.5-27b.toml", "source_type": "catalog", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3.5-27B", "type": "weights" }, { "label": "Official checkpoint configuration", "url": "https://huggingface.co/Qwen/Qwen3.5-27B/raw/main/config.json", "type": "primary" }, { "label": "Official checkpoint metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen3.5-27B", "type": "primary" }, { "label": "Official Qwen release timeline", "url": "https://raw.githubusercontent.com/QwenLM/Qwen3.8/main/README.md", "type": "primary" }, { "label": "Official tensor index", "url": "https://huggingface.co/Qwen/Qwen3.5-27B/raw/main/model.safetensors.index.json", "type": "primary" }, { "label": "Checkpoint Apache-2.0 license", "url": "https://huggingface.co/Qwen/Qwen3.5-27B/raw/main/LICENSE", "type": "primary" }, { "label": "Later Qwen developer comparison tables", "url": "https://huggingface.co/Qwen/Qwen3.6-27B", "type": "primary" }, { "label": "Alibaba model release documentation", "url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.5-27b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete checkpoint: 27.781427952B parameters, comprising 26.895998464B language model including embeddings/output head, 0.460730096B vision and 0.424699392B MTP. The card’s 27B is a nominal language-model count; a separate developer active count remains undisclosed. Dense FFNs with hybrid Gated DeltaNet/full attention; generic family MoE wording does not describe this checkpoint. Native context 262144; documented extension to 1010000.", "parameter_components_b": { "language_model": 26.895998464, "vision": 0.460730096, "mtp": 0.424699392 }, "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T09:44:34.688728Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/048.json" }, { "id": "alibaba/qwen3.5-35b-a3b", "slug": "alibaba--qwen3.5-35b-a3b", "name": "Qwen3.5 35B-A3B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Qwen vision-language model for visual reasoning, documents, and agent tasks", "release_date": "2026-02-24", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 35.951822704, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B", "type": "weights" }, { "label": "Official checkpoint configuration", "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B/raw/main/config.json", "type": "primary" }, { "label": "Official checkpoint tensor metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen3.5-35B-A3B", "type": "primary" }, { "label": "Official checkpoint weight index", "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B/raw/main/model.safetensors.index.json", "type": "primary" }, { "label": "Official open-weight release announcement", "url": "https://github.com/QwenLM/Qwen3.8#news", "type": "primary" }, { "label": "Alibaba API release listing", "url": "https://help.aliyun.com/en/model-studio/newly-released-models", "type": "primary" }, { "label": "Checkpoint license", "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B/blob/main/LICENSE", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.5-35b-a3b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Open weights released February 24, 2026; Alibaba’s API listing is dated February 23. Full checkpoint count is 35.951822704B from published Safetensors metadata, including vision and MTP tensors. The model card’s nominal 35B total and 3B activated figures describe the language model; 3B is not a whole-multimodal active count.", "parameter_components_b": { "language_decoder": 35 }, "active_parameters_scope": "language_decoder", "field_status": {}, "audited_at": "2026-10-26T14:51:29.516623Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/050.json" }, { "id": "alibaba/qwen3.5-35b-a3b-base", "slug": "alibaba--qwen3.5-35b-a3b-base", "name": "Qwen3.5 35B-A3B Base", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Pre-trained Qwen3.5 base checkpoint for fine-tuning and research, without instruction tuning", "release_date": "2026-02-24", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 35.951822704, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B-Base/commit/cbb3419314a02b3597a1a8640a25c53dba36b0b6", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B-Base", "type": "weights" }, { "label": "Exact Base checkpoint metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen3.5-35B-A3B-Base", "type": "primary" }, { "label": "Exact Base weight index", "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B-Base/raw/main/model.safetensors.index.json", "type": "primary" }, { "label": "Exact Base configuration", "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B-Base/raw/main/config.json", "type": "primary" }, { "label": "Exact Base commit history", "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B-Base/commits/main", "type": "primary" }, { "label": "Exact Base checkpoint initial weight publication", "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B-Base/commit/cbb3419314a02b3597a1a8640a25c53dba36b0b6", "type": "primary" }, { "label": "Same-day public Base release discussion", "url": "https://huggingface.co/api/models/Qwen/Qwen3.5-35B-A3B-Base/discussions/1", "type": "primary" }, { "label": "Release-day Base model card", "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B-Base/raw/19a222a7378ddd5ff85d4caf570225d66fc1a2e9/README.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.5-35b-a3b-base.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Full checkpoint: 35.951822704B stored parameters, including vision and MTP. The published 35B total and 3B active figures describe the language model; 3B is not a full multimodal active count. The Base checkpoint was released on February 24, 2026, as its original weight publication and same-day public discussion establish. Trained reasoning/tool support remains unverified.", "active_parameters_scope": "language_decoder", "parameter_components_b": { "language_decoder": 35 }, "field_status": { "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T19:26:56.803246Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/051.json" }, { "id": "inception/mercury-2", "slug": "inception--mercury-2", "name": "Mercury 2", "developer": "Inception", "developer_id": "inception", "family": "mercury", "description": "Mercury 2 is Inception's text-only diffusion language model with adjustable reasoning, tool calling, and a 128,000-token context window.", "release_date": "2026-02-24", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.inceptionlabs.ai/blog/introducing-mercury-2", "source_type": "primary", "sources": [ { "label": "Mercury 2 announcement", "url": "https://www.inceptionlabs.ai/blog/introducing-mercury-2", "type": "primary" }, { "label": "Official dated launch index", "url": "https://www.inceptionlabs.ai/blog", "type": "primary" }, { "label": "Mercury 2 Transformer and diffusion explanation", "url": "https://www.inceptionlabs.ai/blog/mercury-2-the-first-reasoning-model-fast-enough-to-pick-up-the-phone", "type": "primary" }, { "label": "Inception model specifications", "url": "https://docs.inceptionlabs.ai/get-started/models", "type": "primary" }, { "label": "Exact model object and context length", "url": "https://docs.inceptionlabs.ai/api-reference/models/list-available-models", "type": "primary" }, { "label": "Inception Terms of Use", "url": "https://www.inceptionlabs.ai/docs/terms-of-use", "type": "primary" }, { "label": "Mercury 2 commercial Azure license", "url": "https://www.inceptionlabs.ai/blog/mercury-2-on-azure-foundry", "type": "primary" }, { "label": "Later relative model-size comparison", "url": "https://www.inceptionlabs.ai/blog/introducing-mercury-3", "type": "primary" }, { "label": "Current access and retirement notice", "url": "https://www.inceptionlabs.ai/models", "type": "primary" }, { "label": "Later Uno research comparison (size undisclosed)", "url": "https://arxiv.org/pdf/2609.04010", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/inception/mercury-2.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Diffusion generation uses a Transformer denoising network. No total or active parameter count was found in the checked primary sources; the later Mercury size chart is relative only. Dense/MoE details remain undisclosed. As of October 26, 2026, Inception documents Mercury 2 but restricts it to existing customers until November 15, with enterprise support governed by agreements; endpoint operation was not tested.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T13:38:40.020375+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/238.json" }, { "id": "liquid/lfm2-24b-a2b", "slug": "liquid--lfm2-24b-a2b", "name": "LFM2-24B-A2B", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm2", "description": "Open-weight LFM2 instruct language model with hybrid gated convolutions and grouped-query attention, sparse MoE routing, a 32,768-token context window and native function calling.", "release_date": "2026-02-24", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 23.843659008, "active_parameters_b": 2.3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://www.liquid.ai/blog/lfm2-24b-a2b", "source_type": "primary", "sources": [ { "label": "LFM2-24B-A2B: Scaling Up the LFM2 Architecture", "url": "https://www.liquid.ai/blog/lfm2-24b-a2b", "type": "primary" }, { "label": "Hugging Face", "url": "https://huggingface.co/LiquidAI/LFM2-24B-A2B", "type": "weights" }, { "label": "LFM2-24B-A2B released checkpoint configuration", "url": "https://huggingface.co/LiquidAI/LFM2-24B-A2B/blob/a3bbacd91a678b97712f0e323e52f8c24ba29542/config.json", "type": "primary" }, { "label": "LFM Open License v1.0", "url": "https://huggingface.co/LiquidAI/LFM2-24B-A2B/blob/a3bbacd91a678b97712f0e323e52f8c24ba29542/LICENSE", "type": "primary" }, { "label": "LFM2.5-24B-A2B: Built for Fast On-Device Agents", "url": "https://www.liquid.ai/blog/lfm2-5-24b-a2b", "type": "primary" }, { "label": "LFM2-24B-A2B documentation", "url": "https://docs.liquid.ai/lfm/models/lfm2-24b-a2b", "type": "primary" }, { "label": "LFM2-24B-A2B BF16 safetensors header", "url": "https://huggingface.co/LiquidAI/LFM2-24B-A2B/resolve/a3bbacd91a678b97712f0e323e52f8c24ba29542/model.safetensors", "type": "weights" }, { "label": "LFM2 MoE Transformers implementation at v5.0.0rc1", "url": "https://raw.githubusercontent.com/huggingface/transformers/v5.0.0rc1/src/transformers/models/lfm2_moe/modeling_lfm2_moe.py", "type": "primary" }, { "label": "LFM2 MoE configuration implementation at v5.0.0rc1", "url": "https://raw.githubusercontent.com/huggingface/transformers/v5.0.0rc1/src/transformers/models/lfm2_moe/configuration_lfm2_moe.py", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/liquid/lfm2-24b-a2b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Original February 24, 2026 LFM2 instruct checkpoint; now deprecated in favor of LFM2.5. 24B is nominal: complete learned count is 23.843659008B, excluding 2,432 routing-bias buffer elements and counting tied embeddings once. Developer reports 2.3B active per token. Hybrid short-convolution/GQA backbone with sparse MoE routing. Documented context is 32,768 despite config max_position_embeddings=128000; the 131,072 context and reasoning-only behavior belong to the later LFM2.5 successor. Custom LFM Open License v1.0 has commercial-use conditions.", "field_status": {}, "audited_at": "2026-10-26T21:11:29.345158Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/250.json" }, { "id": "alibaba/qwen3.5-flash", "slug": "alibaba--qwen3.5-flash", "name": "Qwen3.5 Flash", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Hosted Qwen3.5 vision-language model with text, image and video inputs, optional thinking, function calling and a one-million-token context window.", "release_date": "2026-02-23", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 35, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://help.aliyun.com/en/model-studio/qwen3-5-flash", "source_type": "primary", "sources": [ { "label": "Exact Flash model documentation", "url": "https://help.aliyun.com/en/model-studio/qwen3-5-flash", "type": "documentation" }, { "label": "Official API release log", "url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "type": "announcement" }, { "label": "Qwen hosted-counterpart mapping and decoder specifications", "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B", "type": "model_card" }, { "label": "Flash commercial thinking support", "url": "https://www.alibabacloud.com/help/en/model-studio/deep-thinking", "type": "documentation" }, { "label": "Flash product capabilities", "url": "https://www.qwencloud.com/models/qwen3.5-flash", "type": "documentation" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.5-flash.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "35B total and 3B active describe the language decoder of Qwen’s explicitly corresponding released model. The complete hosted deployment total is undisclosed. The documented alias maps to the 2026-02-23 snapshot. Open weights and Apache-2.0 refer to the corresponding checkpoint; hosted service terms are separate.", "parameters_scope": "language_decoder", "active_parameters_scope": "language_decoder", "parameter_components_b": { "language_decoder": 35 }, "field_status": {}, "audited_at": "2026-10-26T14:59:39Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/056.json" }, { "id": "openai/gpt-audio-1.5", "slug": "openai--gpt-audio-1.5", "name": "GPT-audio-1.5", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-audio", "description": null, "release_date": "2026-02-23", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 128000, "modalities": { "input": [ "text", "audio" ], "output": [ "text", "audio" ] }, "reasoning": null, "tool_call": null, "source_url": "https://developers.openai.com/api/docs/changelog", "source_type": "primary", "sources": [ { "label": "Dated publisher API releases", "url": "https://developers.openai.com/api/docs/changelog", "type": "documentation" }, { "label": "Publisher model documentation and snapshot mapping", "url": "https://developers.openai.com/api/docs/models/gpt-audio-1.5", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:31:17.277Z", "last_seen": "2026-10-29T10:31:17.277Z", "upstream_present": false, "availability": "public", "notes": "Original publisher conversational audio model. " }, { "id": "openai/gpt-realtime-1.5", "slug": "openai--gpt-realtime-1.5", "name": "GPT-realtime-1.5", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-realtime", "description": null, "release_date": "2026-02-23", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 32000, "modalities": { "input": [ "text", "audio", "image" ], "output": [ "text", "audio" ] }, "reasoning": null, "tool_call": null, "source_url": "https://developers.openai.com/api/docs/changelog", "source_type": "primary", "sources": [ { "label": "Dated publisher API releases", "url": "https://developers.openai.com/api/docs/changelog", "type": "documentation" }, { "label": "Publisher model documentation and snapshot mapping", "url": "https://developers.openai.com/api/docs/models/gpt-realtime-1.5", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:31:17.277Z", "last_seen": "2026-10-29T10:31:17.277Z", "upstream_present": false, "availability": "public", "notes": "Original publisher conversational audio model. " }, { "id": "sarvam/sarvam-105b", "slug": "sarvam--sarvam-105b", "name": "Sarvam 105B", "developer": "Sarvam AI", "developer_id": "sarvam", "family": "sarvam", "description": "Multilingual text model with MoE routing, MLA attention, reasoning and tool calling, trained from scratch by Sarvam AI.", "release_date": "2026-02-20", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 106.031763456, "active_parameters_b": 10.3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/sarvamai/sarvam-105b/blob/abe264bdb8fb2244daadaa59ebc2020b3c139e89/README.md", "source_type": "primary", "sources": [ { "label": "Sarvam-105B model card", "url": "https://huggingface.co/sarvamai/sarvam-105b/blob/abe264bdb8fb2244daadaa59ebc2020b3c139e89/README.md", "type": "primary" }, { "label": "Introducing Indus — February 20, 2026", "url": "https://www.sarvam.ai/blogs/introducing-indus", "type": "primary" }, { "label": "Open-Sourcing Sarvam 30B and 105B — March 6, 2026", "url": "https://www.sarvam.ai/blogs/sarvam-30b-105b", "type": "primary" }, { "label": "Sarvam-105B configuration", "url": "https://huggingface.co/sarvamai/sarvam-105b/blob/abe264bdb8fb2244daadaa59ebc2020b3c139e89/config.json", "type": "primary" }, { "label": "Sarvam-105B complete checkpoint index and all indexed headers", "url": "https://huggingface.co/sarvamai/sarvam-105b/blob/abe264bdb8fb2244daadaa59ebc2020b3c139e89/model.safetensors.index.json", "type": "primary" }, { "label": "Sarvam-105B model implementation", "url": "https://huggingface.co/sarvamai/sarvam-105b/blob/abe264bdb8fb2244daadaa59ebc2020b3c139e89/modeling_sarvam_moe.py", "type": "primary" }, { "label": "Sarvam-105B configuration implementation", "url": "https://huggingface.co/sarvamai/sarvam-105b/blob/abe264bdb8fb2244daadaa59ebc2020b3c139e89/configuration_sarvam_moe.py", "type": "primary" }, { "label": "Sarvam-105B chat template", "url": "https://huggingface.co/sarvamai/sarvam-105b/blob/abe264bdb8fb2244daadaa59ebc2020b3c139e89/chat_template.jinja", "type": "primary" }, { "label": "Sarvam-105B API documentation", "url": "https://docs.sarvam.ai/api/getting-started/models/sarvam-105b", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/sarvam/sarvam-105b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "First documented access was the Indus limited beta on 2026-02-20; open weights followed on 2026-03-06. Complete checkpoint learned-parameter count: 106.031763456B, excluding 3968 persistent router buffer elements from 106031767424 F32 stored elements. The developer’s 10.3B active figure is nominal and its counting convention is unstated; it agrees with rounded non-vocabulary active capacity, rather than counting both full vocabulary matrices. Counts describe the original public checkpoint, not later conversational/work-agent revisions.", "field_status": {}, "audited_at": "2026-10-27T04:25:53.359871Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/490.json" }, { "id": "venice/venice-uncensored-role-play", "slug": "venice--venice-uncensored-role-play", "name": "Venice Role Play Uncensored", "developer": "Venice AI", "developer_id": "venice", "family": "venice-uncensored", "description": null, "release_date": "2026-02-20", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 24, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": null, "context_window": 128000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://cdn.venice.ai/models/venice-uncensored-role-play", "source_type": "primary", "sources": [ { "label": "Publisher model datasheet and release date", "url": "https://cdn.venice.ai/models/venice-uncensored-role-play", "type": "documentation" }, { "label": "Publisher API model inventory and exact hosted capabilities", "url": "https://api.venice.ai/api/v1/models?type=text", "type": "documentation" }, { "label": "Publisher-linked model checkpoint", "url": "https://huggingface.co/dphnAI/24B-3.2-RP-K2-final", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T11:32:22.750Z", "last_seen": "2026-10-29T11:32:22.750Z", "upstream_present": false, "availability": "public", "notes": "Venice identifies this as its company fine-tune based on dphnAI 24B-3.2-RP-K2-final, released February 20, 2026. The nominal 24B total, Apache-2.0 license and open-weights status come from its datasheet; the linked checkpoint returned HTTP 401 during this audit. The live API confirms 128,000-token hosted context, image input and function calling, and reports no reasoning, audio or video input. Exact parameter tensors and architecture remain unverified." }, { "id": "google/gemini-3.1-pro-preview", "slug": "google--gemini-3.1-pro-preview", "name": "Gemini 3.1 Pro Preview", "developer": "Google", "developer_id": "google", "family": "gemini-pro", "description": "Gemini 3.1 Pro preview with multimodal input, text output, thinking and function calling.", "release_date": "2026-02-19", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/3-1-pro", "source_type": "primary", "sources": [ { "label": "Gemini 3.1 Pro preview — Gemini API model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-pro-preview", "type": "primary" }, { "label": "Gemini 3.1 Pro — Google Cloud model documentation", "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/3-1-pro", "type": "primary" }, { "label": "Gemini 3.1 Pro — Model Card", "url": "https://deepmind.google/models/model-cards/gemini-3-1-pro/", "type": "primary" }, { "label": "Gemini 3 Pro Model Card (updated May 2026)", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-Pro-Model-Card.pdf", "type": "primary" }, { "label": "Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Gemini API release notes — February 19, 2026", "url": "https://ai.google.dev/gemini-api/docs/changelog#february-19-2026", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemini-3.1-pro-preview.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Sparse MoE Transformer based on Gemini 3 Pro. Neither this preview nor its named base has a published total or active parameter count in the reviewed sources.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T13:00:39.581355+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/205.json" }, { "id": "google/gemini-3.1-pro-preview-customtools", "slug": "google--gemini-3.1-pro-preview-customtools", "name": "Gemini 3.1 Pro Preview Custom Tools", "developer": "Google", "developer_id": "google", "family": "gemini-pro", "description": "Variant of Gemini 3.1 Pro Preview tuned to prefer custom tools in workflows that also use bash.", "release_date": "2026-02-19", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-pro-preview", "source_type": "primary", "sources": [ { "label": "Gemini 3.1 Pro preview — Gemini API model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3.1-pro-preview", "type": "primary" }, { "label": "Gemini API release notes — February 19, 2026", "url": "https://ai.google.dev/gemini-api/docs/changelog#february-19-2026", "type": "primary" }, { "label": "Gemini 3.1 Pro model card", "url": "https://deepmind.google/models/model-cards/gemini-3-1-pro/", "type": "primary" }, { "label": "Gemini 3 Pro model card — May 2026 update", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-Pro-Model-Card.pdf", "type": "primary" }, { "label": "Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Gemini Interactions API — model options", "url": "https://ai.google.dev/api/interactions-api", "type": "primary" }, { "label": "Gemini CLI configuration — model definitions", "url": "https://github.com/google-gemini/gemini-cli/blob/main/docs/reference/configuration.md", "type": "primary" }, { "label": "Gemini 4 Argon Safety & Security Report", "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemini-3.1-pro-preview-customtools.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Customtools is the documented Gemini 3.1 Pro variant for prioritizing custom tools alongside bash. Its shared card identifies Gemini 3 Pro as the base and describes sparse MoE architecture. Total and active parameter counts are undisclosed for the variant and its named bases in the reviewed sources.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T13:01:26.121387Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/206.json" }, { "id": "trendyol/asure-12b", "slug": "trendyol--asure-12b", "name": "Trendyol Asure 12B", "developer": "Trendyol", "developer_id": "trendyol", "family": "asure", "description": "Turkish-English multimodal instruct model based on Gemma 3 12B, tuned for e-commerce summarization, question answering and structured extraction.", "release_date": "2026-02-19", "date_precision": "day", "open_weights": true, "license": "Gemma", "parameters_b": 12.18732504, "active_parameters_b": 12, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/Trendyol/Trendyol-LLM-Asure-12B/blob/df1164d37f47080ba5fbc1ff4defa0c7a1a218a5/README.md", "source_type": "primary", "sources": [ { "label": "Asure 12B original model card, pinned", "url": "https://huggingface.co/Trendyol/Trendyol-LLM-Asure-12B/blob/df1164d37f47080ba5fbc1ff4defa0c7a1a218a5/README.md", "type": "primary" }, { "label": "Asure 12B released configuration, pinned", "url": "https://huggingface.co/Trendyol/Trendyol-LLM-Asure-12B/blob/df1164d37f47080ba5fbc1ff4defa0c7a1a218a5/config.json", "type": "primary" }, { "label": "Asure 12B complete checkpoint index, pinned", "url": "https://huggingface.co/Trendyol/Trendyol-LLM-Asure-12B/blob/df1164d37f47080ba5fbc1ff4defa0c7a1a218a5/model.safetensors.index.json", "type": "primary" }, { "label": "All five Asure 12B bounded shard headers, pinned", "url": "https://huggingface.co/Trendyol/Trendyol-LLM-Asure-12B/tree/df1164d37f47080ba5fbc1ff4defa0c7a1a218a5", "type": "primary" }, { "label": "Gemma3 implementation at configured version 4.57.3", "url": "https://github.com/huggingface/transformers/blob/v4.57.3/src/transformers/models/gemma3/modeling_gemma3.py", "type": "primary" }, { "label": "Gemma3 configuration defaults at 4.57.3", "url": "https://github.com/huggingface/transformers/blob/v4.57.3/src/transformers/models/gemma3/configuration_gemma3.py", "type": "primary" }, { "label": "Transformers embedding tying at 4.57.3", "url": "https://github.com/huggingface/transformers/blob/v4.57.3/src/transformers/modeling_utils.py", "type": "primary" }, { "label": "SigLIP vision implementation at 4.57.3", "url": "https://github.com/huggingface/transformers/blob/v4.57.3/src/transformers/models/siglip/modeling_siglip.py", "type": "primary" }, { "label": "Asure released chat template, pinned", "url": "https://huggingface.co/Trendyol/Trendyol-LLM-Asure-12B/blob/df1164d37f47080ba5fbc1ff4defa0c7a1a218a5/chat_template.jinja", "type": "primary" }, { "label": "Asure public-release announcement", "url": "https://tr.linkedin.com/posts/cakmak-yusuf_trendyoltrendyol-llm-asure-12b-hugging-activity-7430267282997809153-YSp1", "type": "primary" }, { "label": "Open-source AI Meetup speaker announcement, May4 2026", "url": "https://cosmos.yildiz.edu.tr/news", "type": "primary" }, { "label": "Trendyol official model collections", "url": "https://huggingface.co/Trendyol/collections", "type": "primary" }, { "label": "Asure exact revision metadata, October 27 recheck", "url": "https://huggingface.co/api/models/Trendyol/Trendyol-LLM-Asure-12B/revision/df1164d37f47080ba5fbc1ff4defa0c7a1a218a5", "type": "primary" }, { "label": "Qwen3 Technical Report, Table 5: Gemma 3 12B Base", "url": "https://arxiv.org/html/2505.09388v1#S3.T5", "type": "primary" }, { "label": "Qwen3 Technical Report, Table 18: Gemma 3 12B IT", "url": "https://arxiv.org/html/2505.09388v1#S4.T18", "type": "primary" }, { "label": "Gemma 3 Technical Report, architecture and instruction tuning", "url": "https://storage.googleapis.com/deepmind-media/gemma/Gemma3Report.pdf", "type": "primary" }, { "label": "Kimi-VL Technical Report, Table 3: alternative counting convention", "url": "https://arxiv.org/html/2504.07491v1#S3.T3", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/trendyol/asure-12b.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/Trendyol/Trendyol-LLM-Asure-12B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete checkpoint: 12.18732504B learned parameters, including 11.766034176B language, 0.416866032B vision and 0.004424832B projector. Nominal 12B active is inferred from documented Gemma 3-12B ancestry under Qwen's published comparison convention; it is not an exact per-token count. February 19, 2026 is supported by the developer public-availability announcement; February 20 is a README update. Configured context is 131072 tokens. Reasoning and tool-calling flags are unverified.", "parameter_components_b": { "language_backbone": 11.766034176, "vision_encoder": 0.416866032, "multimodal_projector": 0.004424832 }, "field_status": { "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T18:08:52.175970+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/511.json" }, { "id": "anthropic/claude-sonnet-4-6", "slug": "anthropic--claude-sonnet-4-6", "name": "Claude Sonnet 4.6", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude-sonnet", "description": "Claude Sonnet model for coding, computer use and knowledge work, with adaptive reasoning and a 1M-token context window.", "release_date": "2026-02-17", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1000000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.anthropic.com/news/claude-sonnet-4-6", "source_type": "primary", "sources": [ { "label": "Introducing Claude Sonnet 4.6", "url": "https://www.anthropic.com/news/claude-sonnet-4-6", "type": "primary" }, { "label": "Claude Sonnet 4.6 - Claude Platform Docs", "url": "https://platform.claude.com/docs/en/models/sonnet-4-6/overview", "type": "primary" }, { "label": "Claude Sonnet 4.6 System Card", "url": "https://www.anthropic.com/claude-sonnet-4-6-system-card", "type": "primary" }, { "label": "Claude Sonnet 5 System Card", "url": "https://www.anthropic.com/claude-sonnet-5-system-card", "type": "primary" }, { "label": "Claude Sonnet 5.5 System Card", "url": "https://www.anthropic.com/claude-sonnet-5-5-system-card", "type": "primary" }, { "label": "Verbalizable Representations Form a Global Workspace in Language Models", "url": "https://transformer-circuits.pub/2026/workspace/index.html", "type": "primary" }, { "label": "GLM-5.3 and the spread of advanced cyber capabilities", "url": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities", "type": "primary" }, { "label": "Commercial Terms of Service", "url": "https://www.anthropic.com/legal/commercial-terms", "type": "primary" }, { "label": "PDF support", "url": "https://platform.claude.com/docs/en/build-with-claude/pdf-support", "type": "primary" }, { "label": "Tool use with Claude", "url": "https://platform.claude.com/docs/en/agents-and-tools/tool-use/overview", "type": "primary" }, { "label": "Claude Opus 5.5 System Card", "url": "https://www.anthropic.com/claude-opus-5-5-system-card", "type": "primary" }, { "label": "Claude Haiku 5.5 System Card", "url": "https://www.anthropic.com/claude-haiku-5-5-system-card", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/anthropic/claude-sonnet-4-6.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Nominal total and active parameter counts remain undisclosed; no publicly sized base lineage is established. Third-party capacity and throughput estimates are not specifications. Sonnet 4.6 architecture remains undisclosed. PDF input denotes document processing through text and vision.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T10:49:25.453032Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/114.json" }, { "id": "cohere/tiny-aya-earth", "slug": "cohere--tiny-aya-earth", "name": "Tiny Aya Earth", "developer": "Cohere", "developer_id": "cohere", "family": "tiny-aya", "description": "Compact multilingual model specialized for West Asian and African languages", "release_date": "2026-02-17", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-4.0", "parameters_b": 3.35, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://cohere.com/blog/cohere-labs-tiny-aya", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/tiny-aya-earth", "type": "weights" }, { "label": "Tiny Aya technical report", "url": "https://arxiv.org/html/2603.11510v1", "type": "primary" }, { "label": "Cohere Tiny Aya documentation", "url": "https://docs.cohere.com/docs/tiny-aya", "type": "primary" }, { "label": "Cohere Tiny Aya launch", "url": "https://cohere.com/blog/cohere-labs-tiny-aya", "type": "primary" }, { "label": "Official Earth repository metadata", "url": "https://huggingface.co/api/models/CohereLabs/tiny-aya-earth", "type": "primary" }, { "label": "Cohere2 implementation", "url": "https://raw.githubusercontent.com/huggingface/transformers/v4.57.1/src/transformers/models/cohere2/modeling_cohere2.py", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/cohere/tiny-aya-earth.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Tiny Aya Earth is the Africa/West Asia regional variant with a nominal 3.35B total and 8192-token input context. Active parameters are not separately disclosed; the report's 2.8B non-embedding count is a different measure. The card's isolated 111B terms boilerplate conflicts with its model specification. Dedicated thinking and native tool-calling support remain unverified for Earth. CC-BY-NC-4.0 includes acceptable-use conditions.", "field_status": { "active_parameters_b": "undisclosed", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T12:30:44.147384+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/157.json" }, { "id": "cohere/tiny-aya-fire", "slug": "cohere--tiny-aya-fire", "name": "Tiny Aya Fire", "developer": "Cohere", "developer_id": "cohere", "family": "tiny-aya", "description": "Compact multilingual model specialized for South Asian languages", "release_date": "2026-02-17", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-4.0", "parameters_b": 3.35, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://cohere.com/blog/cohere-labs-tiny-aya", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/tiny-aya-fire", "type": "weights" }, { "label": "Tiny Aya technical report", "url": "https://arxiv.org/html/2603.11510v1", "type": "primary" }, { "label": "Tiny Aya launch announcement", "url": "https://cohere.com/blog/cohere-labs-tiny-aya", "type": "primary" }, { "label": "Tiny Aya Fire repository metadata", "url": "https://huggingface.co/api/models/CohereLabs/tiny-aya-fire", "type": "primary" }, { "label": "Cohere2 implementation", "url": "https://github.com/huggingface/transformers/blob/main/src/transformers/models/cohere2/modeling_cohere2.py", "type": "primary" }, { "label": "Tiny Aya documentation", "url": "https://docs.cohere.com/docs/tiny-aya", "type": "primary" }, { "label": "Tiny Aya Base model card", "url": "https://huggingface.co/CohereLabs/tiny-aya-base", "type": "primary" }, { "label": "Building Multilingual Bridges: Data Mixing as the Pillar of Generalization for In-Language Reasoning", "url": "https://arxiv.org/html/2609.10445v1", "type": "primary" }, { "label": "How small AI models can make a big impact for enterprises", "url": "https://cohere.com/blog/how-small-models-can-make-a-big-impact-for-enterprises", "type": "primary" }, { "label": "Multilingual Bridges: How Data Mixing Unlocks In-Language Reasoning", "url": "https://cohere.com/blog/building-multilingual-bridges", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/cohere/tiny-aya-fire.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Tiny Aya Fire is the South Asian regional instruction-tuned model. 3.35B is Cohere's nominal total; active parameters are not separately disclosed. Input context is 8,192 tokens, with output separately specified as 8,192. The native template has no thinking or tool-call branches; dedicated reasoning and native tool calling remain unverified. The model card's 111B terms boilerplate conflicts with its explicit 3.35B specification.", "field_status": { "active_parameters_b": "undisclosed", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T12:32:32.919173+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/158.json" }, { "id": "cohere/tiny-aya-global", "slug": "cohere--tiny-aya-global", "name": "Tiny Aya Global", "developer": "Cohere", "developer_id": "cohere", "family": "tiny-aya", "description": "Compact instruction-tuned multilingual model designed to balance performance across languages and regions.", "release_date": "2026-02-17", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-4.0", "parameters_b": 3.35, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://cohere.com/blog/cohere-labs-tiny-aya", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/tiny-aya-global", "type": "weights" }, { "label": "Tiny Aya developer technical report", "url": "https://arxiv.org/html/2603.11510v1", "type": "primary" }, { "label": "Tiny Aya documentation", "url": "https://docs.cohere.com/docs/tiny-aya", "type": "primary" }, { "label": "Tiny Aya launch announcement", "url": "https://cohere.com/blog/cohere-labs-tiny-aya", "type": "primary" }, { "label": "Exact Global metadata and chat template", "url": "https://huggingface.co/api/models/CohereLabs/tiny-aya-global", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/cohere/tiny-aya-global.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "3.35B is Cohere's nominal total; active parameters are not separately disclosed. The report's 2.8B non-embedding count is not an active count. First-party language totals differ (67 in launch, 70 in documentation). CC-BY-NC-4.0 includes an acceptable-use addendum. Dedicated reasoning and native/API tool support remain unverified despite a direct-response chat template.", "field_status": { "active_parameters_b": "undisclosed", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T12:33:15.752537+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/159.json" }, { "id": "cohere/tiny-aya-water", "slug": "cohere--tiny-aya-water", "name": "Tiny Aya Water", "developer": "Cohere", "developer_id": "cohere", "family": "tiny-aya", "description": "Compact multilingual model specialized for European and Asia-Pacific languages", "release_date": "2026-02-17", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-4.0", "parameters_b": 3.35, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://cohere.com/blog/cohere-labs-tiny-aya", "source_type": "primary", "sources": [ { "label": "Cohere Labs launch announcement", "url": "https://cohere.com/blog/cohere-labs-tiny-aya", "type": "primary" }, { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/tiny-aya-water", "type": "weights" }, { "label": "Tiny Aya technical report", "url": "https://arxiv.org/html/2603.11510v1", "type": "primary" }, { "label": "Cohere Tiny Aya documentation", "url": "https://docs.cohere.com/docs/tiny-aya", "type": "primary" }, { "label": "Exact checkpoint metadata and chat template", "url": "https://huggingface.co/api/models/CohereLabs/tiny-aya-water", "type": "primary" }, { "label": "Cohere model overview", "url": "https://docs.cohere.com/docs/models", "type": "primary" }, { "label": "Building Multilingual Bridges: Data Mixing as the Pillar of Generalization for In-Language Reasoning", "url": "https://arxiv.org/html/2609.10445v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/cohere/tiny-aya-water.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "3.35B is Cohere’s nominal total; active parameters are not separately disclosed. The report’s 2.8B figure excludes embeddings and is not an active count. The released chat template has no native tool or thinking handling; hosted API support for those features remains unverified.", "field_status": { "active_parameters_b": "undisclosed", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T12:33:11.336109+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/160.json" }, { "id": "alibaba/qwen3.5-397b-a17b", "slug": "alibaba--qwen3.5-397b-a17b", "name": "Qwen3.5 397B-A17B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Open-weight Qwen vision-language MoE combining Gated DeltaNet and full attention, with thinking and tool calling.", "release_date": "2026-02-16", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 403.397928944, "active_parameters_b": 17, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.5-397b-a17b.toml", "source_type": "catalog", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "type": "weights" }, { "label": "Checkpoint configuration", "url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B/raw/main/config.json", "type": "primary" }, { "label": "Official checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen3.5-397B-A17B", "type": "primary" }, { "label": "Checkpoint tensor index", "url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B/raw/main/model.safetensors.index.json", "type": "primary" }, { "label": "Official release history", "url": "https://github.com/QwenLM/Qwen3.8#news", "type": "primary" }, { "label": "Alibaba release announcement", "url": "https://home.alibabagroup.com/en-US/document-1960233590314762240", "type": "primary" }, { "label": "Qwen release blog (indexed date conflict)", "url": "https://qwen.ai/blog?id=qwen3.5", "type": "primary" }, { "label": "Apache-2.0 license", "url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B/blob/main/LICENSE", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.5-397b-a17b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Full checkpoint contains 403.397928944B stored parameters including vision and MTP tensors; 397B total and 17B active are nominal language-decoder figures. Native context is 262144 tokens. Official repository and Alibaba date the first release February 16, 2026; the indexed Qwen blog displays February 15.", "parameter_components_b": { "language_decoder": 397 }, "active_parameters_scope": "language_decoder", "field_status": {}, "audited_at": "2026-10-26T14:54:00.665900Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/052.json" }, { "id": "alibaba/qwen3.5-plus", "slug": "alibaba--qwen3.5-plus", "name": "Qwen3.5 Plus", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Hosted Qwen3.5 vision-language model with hybrid linear attention and sparse MoE, a one-million-token context, optional thinking, and function calling.", "release_date": "2026-02-15", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 397, "active_parameters_b": 17, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-5-plus", "source_type": "primary", "sources": [ { "label": "Exact Plus model documentation", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-5-plus", "type": "documentation" }, { "label": "Official API release log", "url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "type": "announcement" }, { "label": "Official hosted-counterpart mapping and decoder counts", "url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B/blob/main/README.md", "type": "model_card" }, { "label": "Plus hybrid-thinking support", "url": "https://www.alibabacloud.com/help/en/model-studio/deep-thinking", "type": "documentation" }, { "label": "Alibaba announcement", "url": "https://www.alibabagroup.com/en-US/document-1960233590314762240", "type": "announcement" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3.5-plus.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "397B total and 17B active describe the language decoder of Qwen’s explicitly corresponding released model. The complete hosted deployment total is undisclosed. The documented alias maps to the 2026-02-15 snapshot. Open weights and Apache-2.0 refer to the corresponding checkpoint; hosted service terms are separate.", "parameters_scope": "language_decoder", "parameter_components_b": { "language_decoder": 397 }, "active_parameters_scope": "language_decoder", "field_status": {}, "audited_at": "2026-10-26T15:05:30Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/057.json" }, { "id": "bytedance-seed/seed-2.0-code", "slug": "bytedance-seed--seed-2.0-code", "name": "Seed 2.0 Code", "developer": "ByteDance", "developer_id": "bytedance-seed", "family": "seed", "description": "ByteDance Seed 2.0 model specialized for coding, with visual understanding, configurable reasoning, and tool calling.", "release_date": "2026-02-14", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 256000, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://seed.bytedance.com/en/blog/seed-2-0-official-launch", "source_type": "primary", "sources": [ { "label": "Seed 2.0 Official Launch", "url": "https://seed.bytedance.com/en/blog/seed-2-0-official-launch", "type": "primary" }, { "label": "Model list", "url": "https://docs.volcengine.com/docs/ark/model-list?lang=zh", "type": "primary" }, { "label": "Deep thinking", "url": "https://docs.volcengine.com/docs/ark/deep-thinking?lang=en", "type": "primary" }, { "label": "Video understanding", "url": "https://docs.volcengine.com/docs/ark/video-understanding?lang=zh", "type": "primary" }, { "label": "Doubao model service agreement", "url": "https://docs.volcengine.com/docs/ark/doubao-model-service-agreement?lang=zh", "type": "primary" }, { "label": "Coding Plan model release announcements", "url": "https://docs.volcengine.com/docs/ark/coding-plan-personal-model-release?lang=zh", "type": "primary" }, { "label": "Seed2.0 current first-party model comparison", "url": "https://seed.bytedance.com/en/seed2", "type": "primary" }, { "label": "Seed2.1 Officially Released: Advancing AI Productivity", "url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/bytedance-seed/seed-2.0-code.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Official API revision: doubao-seed-2-0-code-preview-260215. Earliest documented release: 2026-02-14. Volcano Engine documents 256k context (represented as nominal decimal 256000), 224k maximum input, and 128k maximum output. An exact 262144-token context boundary is not established. The exact Code revision is marked as approaching retirement; no endpoint availability test was performed. Nominal total and active counts remain undisclosed; coding specialization does not establish a publicly sized base. Architecture also remains undisclosed.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T10:55:15.005945+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/132.json" }, { "id": "bytedance-seed/seed-2.0-lite", "slug": "bytedance-seed--seed-2.0-lite", "name": "Seed 2.0 Lite", "developer": "ByteDance", "developer_id": "bytedance-seed", "family": "seed", "description": "ByteDance Seed 2.0 model with multimodal reasoning and tool calling. The late-April 2026 Lite update added audio input.", "release_date": "2026-02-14", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 256000, "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://seed.bytedance.com/en/seed2", "source_type": "primary", "sources": [ { "label": "Official Seed2.0 launch announcement", "url": "https://seed.bytedance.com/en/blog/seed2-0-%E6%AD%A3%E5%BC%8F%E5%8F%91%E5%B8%83", "type": "primary" }, { "label": "Official Seed2.0 model page and Lite update", "url": "https://seed.bytedance.com/en/seed2", "type": "primary" }, { "label": "BytePlus exact-version model specifications", "url": "https://docs.byteplus.com/en/docs/modelark/model-list", "type": "primary" }, { "label": "BytePlus model release notes", "url": "https://docs.byteplus.com/en/docs/modelark/model-release-announcement", "type": "primary" }, { "label": "BytePlus model service terms", "url": "https://docs.byteplus.com/en/docs/legal/docs-service-specific-terms", "type": "primary" }, { "label": "Later first-party Seed2.0 model card", "url": "https://arxiv.org/abs/2607.00248", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/bytedance-seed/seed-2.0-lite.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Original public launch: February 14, 2026. Audio input describes the late-April Lite 0428 upgrade, not the original February Lite; BytePlus lists the 260428 rollout under May. This undated record summarizes current documented model capabilities without verifying reseller alias routing. Context is nominal 256K, stored as 256000; an exact token ceiling is not independently established. Nominal total and active counts remain undisclosed; no publicly sized base relationship is established for either Lite revision. Production architecture remains undisclosed. The API cookbook Apache-2.0 license does not license model weights.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T10:56:07.718193+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/133.json" }, { "id": "bytedance-seed/seed-2.0-mini", "slug": "bytedance-seed--seed-2.0-mini", "name": "Seed 2.0 Mini", "developer": "ByteDance", "developer_id": "bytedance-seed", "family": "seed", "description": "ByteDance Seed 2.0 Mini model with reasoning and tool calling. Revision 260428 accepts text, images, audio, and video and generates text.", "release_date": "2026-02-14", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 256000, "modalities": { "input": [ "text", "image", "audio", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://seed.bytedance.com/en/blog/seed2-0-%E6%AD%A3%E5%BC%8F%E5%8F%91%E5%B8%83", "source_type": "primary", "sources": [ { "label": "Seed 2.0 Official Launch", "url": "https://seed.bytedance.com/en/blog/seed2-0-%E6%AD%A3%E5%BC%8F%E5%8F%91%E5%B8%83", "type": "primary" }, { "label": "Seed2.0 model overview", "url": "https://seed.bytedance.com/en/seed2", "type": "primary" }, { "label": "Seed2.0 Model Card: Towards Intelligence Frontier for Real-World Complexity", "url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "type": "primary" }, { "label": "Seed2.1 Model Card: Agentic Intelligence for Productivity", "url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "type": "primary" }, { "label": "Model list", "url": "https://docs.byteplus.com/en/docs/modelark/model-list", "type": "primary" }, { "label": "Model release announcements", "url": "https://docs.byteplus.com/en/docs/modelark/model-release-announcement", "type": "primary" }, { "label": "Model fine-tuning overview", "url": "https://docs.byteplus.com/en/docs/modelark/fine-tuning-overview", "type": "primary" }, { "label": "Service Specific Terms — ModelArk and Model Services", "url": "https://docs.byteplus.com/en/docs/legal/docs-service-specific-terms", "type": "primary" }, { "label": "Official Seed2.0 repository", "url": "https://github.com/ByteDance-Seed/Seed2.0", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/bytedance-seed/seed-2.0-mini.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "First publicly released February 14, 2026. Current modality support describes Mini revision 260428, listed in BytePlus release announcements in May 2026; its identifier alone does not prove an April 28 public release. Context is nominal 256K, represented as 256000 because the inspected primary documentation does not define the K multiplier; an exact token ceiling is unverified. Nominal total and active counts remain undisclosed; no publicly sized base relationship is documented. Architecture remains undisclosed. The official repository Apache-2.0 license covers repository materials, not distributed Mini weights.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T10:57:59.781963+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/134.json" }, { "id": "bytedance-seed/seed-2.0-pro", "slug": "bytedance-seed--seed-2.0-pro", "name": "Seed 2.0 Pro", "developer": "ByteDance", "developer_id": "bytedance-seed", "family": "seed", "description": "ByteDance Seed 2.0 model for multimodal reasoning and tool-assisted tasks.", "release_date": "2026-02-14", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 256000, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://seed.bytedance.com/en/blog/seed-2-0-official-launch", "source_type": "primary", "sources": [ { "label": "Seed 2.0 Official Launch", "url": "https://seed.bytedance.com/en/blog/seed-2-0-official-launch", "type": "primary" }, { "label": "Seed2.0 product page", "url": "https://seed.bytedance.com/en/seed2", "type": "primary" }, { "label": "Seed2.0 Model Card: Towards Intelligence Frontier for Real-World Complexity", "url": "https://arxiv.org/pdf/2607.00248", "type": "primary" }, { "label": "Seed2.1 Model Card: Agentic Intelligence for Productivity", "url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "type": "primary" }, { "label": "Volcano Ark model list", "url": "https://docs.volcengine.com/docs/ark/model-list?lang=en", "type": "primary" }, { "label": "Doubao model service agreement", "url": "https://docs.volcengine.com/docs/ark/doubao-model-service-agreement?lang=en", "type": "primary" }, { "label": "Official Seed2.0 API cookbook repository", "url": "https://github.com/ByteDance-Seed/Seed2.0", "type": "primary" }, { "label": "Public Summary of Training Data Content for Seed 2.0 Pro", "url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/transparency/20260803/Public%20Summary%20of%20Training%20Data%20Content%20for%20Seed%202.0%20Pro%20.pdf", "type": "primary" }, { "label": "DescribeBaasAIModels official API example: Doubao-Seed-2.0-pro", "url": "https://api.volcengine.com/api-explorer/debug?action=DescribeBaasAIModels&groupName=BaaS+AI+%E7%AE%A1%E7%90%86&serviceCode=aidap&version=2025-10-01", "type": "primary" }, { "label": "Seed2.0 original February Model Card", "url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/bytedance-seed/seed-2.0-pro.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Provider context is advertised as 256k; 256000 is nominal decimal normalization. The original doubao-seed-2-0-pro-260215 endpoint is marked impending retirement in current Ark documentation. Nominal total and active counts remain undisclosed; no publicly sized base relationship is documented. Architecture remains undisclosed. The cookbook Apache-2.0 license does not license model weights.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T10:59:32.916204+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/135.json" }, { "id": "minimax/MiniMax-M2.5", "slug": "minimax--minimax-m2.5", "name": "MiniMax-M2.5", "developer": "MiniMax", "developer_id": "minimax", "family": "minimax", "description": "Open-weight MiniMax M2-series language model trained with reinforcement learning for coding, tool use, search, and office workflows.", "release_date": "2026-02-12", "date_precision": "day", "open_weights": true, "license": "MiniMax Model License", "parameters_b": 230, "active_parameters_b": 9.8, "parameters_source": "primary_source", "architecture": "moe", "context_window": 196608, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.minimax.io/news/minimax-m25", "source_type": "primary", "sources": [ { "label": "MiniMax M2.5: Built for Real-World Productivity", "url": "https://www.minimax.io/news/minimax-m25", "type": "primary" }, { "label": "MiniMax-M2.5 released configuration", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5/blob/2515a3ea0afec1dd22d0c7794b3a91557804d195/config.json", "type": "weights" }, { "label": "MiniMax-M2.5 reference implementation", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5/blob/2515a3ea0afec1dd22d0c7794b3a91557804d195/modeling_minimax_m2.py", "type": "weights" }, { "label": "MiniMax Model License for MiniMax-M2.5", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5/blob/2515a3ea0afec1dd22d0c7794b3a91557804d195/LICENSE-MODEL", "type": "weights" }, { "label": "MiniMax staff statement on M2-series size", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5/discussions/15", "type": "primary" }, { "label": "MiniMax M2.5 vLLM deployment guide", "url": "https://github.com/MiniMax-AI/MiniMax-M2.5/blob/main/docs/vllm_deploy_guide.md", "type": "primary" }, { "label": "MiniMax-M2.5 Tool Calling Guide", "url": "https://github.com/MiniMax-AI/MiniMax-M2.5/blob/main/docs/tool_calling_guide.md", "type": "primary" }, { "label": "The MiniMax-M2 Series: Mini Activations Unleashing Max Real-World Intelligence", "url": "https://arxiv.org/html/2605.26494v2", "type": "primary" }, { "label": "MiniMax-M2.5 checkpoint index and bounded Safetensors headers", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5/blob/2515a3ea0afec1dd22d0c7794b3a91557804d195/model.safetensors.index.json", "type": "weights" }, { "label": "MiniMax-M2 official base configuration", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2/raw/main/config.json", "type": "primary" }, { "label": "Forge: Scalable Agent RL Framework and Algorithm", "url": "https://www.minimax.io/news/forge-scalable-agent-rl-framework-and-algorithm", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/minimax/MiniMax-M2.5.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "230B is the developer’s nominal total. The 9.8B active count is a nominal inference from the documented M2 post-training lineage and matching backbone dimensions/routing, following the report’s main-model convention. Context 196608 is the released checkpoint limit; historical 204800-token native API context was not independently recovered. Current license is the custom MiniMax Model License, despite the model card metadata label modified-mit.", "field_status": {}, "audited_at": "2026-10-27T10:54:20.476345Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/303.json" }, { "id": "openai/gpt-5.3-codex-spark", "slug": "openai--gpt-5.3-codex-spark", "name": "GPT-5.3-Codex-Spark", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5.3", "description": "Low-latency GPT-5.3 coding model for interactive edits in Codex.", "release_date": "2026-02-12", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://openai.com/index/introducing-gpt-5-3-codex-spark/", "source_type": "primary", "sources": [ { "label": "Introducing GPT-5.3-Codex-Spark", "url": "https://openai.com/index/introducing-gpt-5-3-codex-spark/", "type": "primary" }, { "label": "Introducing OpenAI GPT-5.3-Codex-Spark Powered by Cerebras", "url": "https://www.cerebras.ai/blog/openai-codexspark", "type": "primary" }, { "label": "Codex changelog: Spark launch and retirement", "url": "https://developers.openai.com/codex/changelog/#codex-2026-09-14-codex-spark-deprecation", "type": "primary" }, { "label": "Models: deprecated Codex models", "url": "https://learn.chatgpt.com/docs/models", "type": "primary" }, { "label": "OpenAI Models at Ultrafast Speed", "url": "https://www.cerebras.ai/customer-spotlights/openai", "type": "primary" }, { "label": "Speeding up agentic workflows with WebSockets in the Responses API", "url": "https://openai.com/index/speeding-up-agentic-workflows-with-websockets/", "type": "primary" }, { "label": "All models", "url": "https://developers.openai.com/api/docs/models/all.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "Stop Shipping AI Slop: How Codex Spark Changes The Way You Code", "url": "https://www.cerebras.ai/blog/codex-spark-best-practices", "type": "primary" }, { "label": "GPT-5.3-Codex System Card", "url": "https://deploymentsafety.openai.com/gpt-5-3-codex/gpt-5-3-codex.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5.3-codex-spark.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Research preview released February 12, 2026; retired from Codex desktop, CLI and IDE on September 14, 2026. Last documented specifications found are text-only and 128K context. Total and active parameter counts remain undisclosed; the smaller-model relationship to GPT-5.3-Codex supplies no numerical size. Neural architecture remains undisclosed. Dedicated reasoning support remains unverified; this does not establish absence of reasoning. Exact API model documentation returned 404. Private design-partner API access at launch does not establish general API availability.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed", "reasoning": "unverified" }, "audited_at": "2026-10-27T16:54:10Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/429.json" }, { "id": "zhipuai/glm-5", "slug": "zhipuai--glm-5", "name": "GLM-5", "developer": "Z.ai", "developer_id": "zhipuai", "family": "glm", "description": "Text MoE model with sparse attention, switchable reasoning and function calling.", "release_date": "2026-02-06", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 744, "active_parameters_b": 40, "parameters_source": "primary_source", "architecture": "moe", "context_window": 202752, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://arxiv.org/html/2602.15763v1", "source_type": "primary", "sources": [ { "label": "GLM-5 technical report, architecture, Table 10 and anonymous release account", "url": "https://arxiv.org/html/2602.15763v1", "type": "primary" }, { "label": "GLM-5 original checkpoint card at pinned revision", "url": "https://huggingface.co/zai-org/GLM-5/blob/c183ef8c61faee82855eca1ed9bb3a9a7ce3b0b2/README.md", "type": "weights" }, { "label": "GLM-5 exact checkpoint configuration", "url": "https://huggingface.co/zai-org/GLM-5/blob/c183ef8c61faee82855eca1ed9bb3a9a7ce3b0b2/config.json", "type": "weights" }, { "label": "GLM-5 exact tokenizer configuration", "url": "https://huggingface.co/zai-org/GLM-5/blob/c183ef8c61faee82855eca1ed9bb3a9a7ce3b0b2/tokenizer_config.json", "type": "weights" }, { "label": "GLM-5 exact chat template", "url": "https://huggingface.co/zai-org/GLM-5/blob/c183ef8c61faee82855eca1ed9bb3a9a7ce3b0b2/chat_template.jinja", "type": "weights" }, { "label": "GLM-5 native API guide", "url": "https://docs.z.ai/guides/llm/glm-5", "type": "primary" }, { "label": "Pony Alpha historical release page", "url": "https://openrouter.ai/openrouter/pony-alpha/activity", "type": "primary" }, { "label": "Z.ai release notes for the named GLM-5 launch", "url": "https://docs.z.ai/release-notes/new-released.md", "type": "primary" }, { "label": "Pinned GLM-5 repository metadata", "url": "https://huggingface.co/api/models/zai-org/GLM-5/revision/c183ef8c61faee82855eca1ed9bb3a9a7ce3b0b2", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/zhipuai/glm-5.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/GLM-5", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "First public preview: Pony Alpha on February 6, 2026, directly identified as GLM-5 by its authors and original host. Z.ai release notes date the named launch February 12; final checkpoint specifications are not claims about identical preview weights. The published 744B total and 40B active figures are nominal and include MTP while excluding word embeddings and output. Automated repository storage totals differ and were not reconciled as learned parameters. GLM-5 uses MoE with MLA and DeepSeek Sparse Attention: 3 dense and 75 MoE layers, 8 of 256 routed experts plus a shared expert, and one MTP module reused across three training prediction steps. Exact checkpoint/tokenizer context is 202752 tokens; the native guide rounds this to 200K and gives a separate 128K output limit.", "parameters_scope": "nonembedding_with_mtp", "parameter_components_b": { "nonembedding_with_mtp": 744 }, "active_parameters_scope": "nonembedding_with_mtp", "active_parameters_by_phase_b": { "nonembedding_with_mtp": 40 }, "field_status": {}, "audited_at": "2026-10-27T06:17:02.002499Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/557.json" }, { "id": "anthropic/claude-opus-4-6", "slug": "anthropic--claude-opus-4-6", "name": "Claude Opus 4.6", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude-opus", "description": "Claude Opus model for coding, agentic tasks and knowledge work, with adaptive reasoning and a 1M-token context window.", "release_date": "2026-02-05", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": "undisclosed", "architecture": "transformer", "context_window": 1000000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.anthropic.com/news/claude-opus-4-6", "source_type": "primary", "sources": [ { "label": "Introducing Claude Opus 4.6", "url": "https://www.anthropic.com/news/claude-opus-4-6", "type": "primary" }, { "label": "Claude Opus 4.6 - Claude Platform Docs", "url": "https://platform.claude.com/docs/en/models/opus-4-6/overview", "type": "primary" }, { "label": "Verbalizable Representations Form a Global Workspace in Language Models", "url": "https://transformer-circuits.pub/2026/workspace/index.html", "type": "primary" }, { "label": "GLM-5.3 and the spread of advanced cyber capabilities", "url": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities", "type": "primary" }, { "label": "Commercial Terms of Service", "url": "https://www.anthropic.com/legal/commercial-terms", "type": "primary" }, { "label": "PDF support", "url": "https://platform.claude.com/docs/en/build-with-claude/pdf-support", "type": "primary" }, { "label": "Tool use with Claude", "url": "https://platform.claude.com/docs/en/agents-and-tools/tool-use/overview", "type": "primary" }, { "label": "Claude Opus 4.6 System Card", "url": "https://www.anthropic.com/claude-opus-4-6-system-card", "type": "primary" }, { "label": "Claude Opus 5.5 System Card", "url": "https://www.anthropic.com/claude-opus-5-5-system-card", "type": "primary" }, { "label": "Models overview", "url": "https://platform.claude.com/docs/en/models/overview", "type": "primary" }, { "label": "Claude Opus 4.7 System Card", "url": "https://www-cdn.anthropic.com/037f06850df7fbe871e206dad004c3db5fd50340/Claude%20Opus%204.7%20System%20Card.pdf", "type": "primary" }, { "label": "Claude Opus 4.8 System Card", "url": "https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea.pdf", "type": "primary" }, { "label": "Claude Opus 5 System Card", "url": "https://anthropic.com/claude-opus-5-system-card", "type": "primary" }, { "label": "Claude Sonnet 5.5 System Card", "url": "https://www.anthropic.com/claude-sonnet-5-5-system-card", "type": "primary" }, { "label": "Claude Haiku 5.5 System Card", "url": "https://www.anthropic.com/claude-haiku-5-5-system-card", "type": "primary" }, { "label": "Anthropic Transparency Hub — Claude Opus 4.6 summary", "url": "https://www.anthropic.com/transparency", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/anthropic/claude-opus-4-6.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Transformer architecture is supported by exact-model interpretability experiments. Anthropic has not disclosed nominal total or active parameter counts in reviewed specifications and cards; published third-party size estimates are unverified. Dense/MoE subtype remains undisclosed. PDF input denotes document processing through text and vision.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T10:38:03.109725Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/107.json" }, { "id": "openai/gpt-5.3-codex", "slug": "openai--gpt-5.3-codex", "name": "GPT-5.3-Codex", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5.3", "description": "OpenAI reasoning model optimized for agentic coding, with text and image input and tool calling.", "release_date": "2026-02-05", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 400000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-gpt-5-3-codex/", "source_type": "primary", "sources": [ { "label": "Introducing GPT-5.3-Codex", "url": "https://openai.com/index/introducing-gpt-5-3-codex/", "type": "primary" }, { "label": "GPT-5.3-Codex System Card", "url": "https://deploymentsafety.openai.com/gpt-5-3-codex/gpt-5-3-codex.pdf", "type": "primary" }, { "label": "GPT-5.3-Codex API model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.3-codex.md", "type": "primary" }, { "label": "OpenAI API changelog", "url": "https://developers.openai.com/api/docs/changelog.md", "type": "primary" }, { "label": "OpenAI API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "Introducing GPT-5.4", "url": "https://openai.com/index/introducing-gpt-5-4/", "type": "primary" }, { "label": "GPT-5.4 Thinking System Card", "url": "https://deploymentsafety.openai.com/gpt-5-4-thinking/gpt-5-4-thinking.pdf", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201.md", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "OpenAI model inventory", "url": "https://developers.openai.com/api/docs/models/all.md", "type": "primary" }, { "label": "GPT-5.5 System Card", "url": "https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf", "type": "primary" }, { "label": "Introducing GPT-5.3-Codex-Spark", "url": "https://openai.com/index/introducing-gpt-5-3-codex-spark/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5.3-codex.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released in Codex on February 5, 2026; the Responses API launch followed February 24. The API identifier is undated, with no separate dated snapshot documented. The 400,000-token context has a 272,000-token maximum input and 128,000-token maximum output. PDF denotes API extraction into text and page images. Transformer is a qualified broad inference from prompt caching and OpenAI pre-GPT-5.6 KV-cache documentation. Detailed topology and total/active parameter counts remain undisclosed in the checked OpenAI sources; Spark is a distinct smaller model. Deprecated October 1, 2026, with API shutdown scheduled April 1, 2027; live endpoint operation was not tested.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T16:53:53.542639Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/428.json" }, { "id": "stepfun/step-3.5-flash", "slug": "stepfun--step-3.5-flash", "name": "Step 3.5 Flash", "developer": "StepFun", "developer_id": "stepfun", "family": "step", "description": "Text reasoning model with sparse MoE routing, hybrid sliding-window attention, tool calling, and three multi-token prediction heads.", "release_date": "2026-02-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 199.38428928, "active_parameters_b": 11, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/stepfun-ai/Step-3.5-Flash/blob/ab446a3de5e171ea341227e24bb1f090e1b771f7/README.md", "source_type": "primary", "sources": [ { "label": "Original Step 3.5 Flash model card", "url": "https://huggingface.co/stepfun-ai/Step-3.5-Flash/blob/ab446a3de5e171ea341227e24bb1f090e1b771f7/README.md", "type": "primary" }, { "label": "Exact Step 3.5 Flash checkpoint configuration", "url": "https://huggingface.co/stepfun-ai/Step-3.5-Flash/blob/ab446a3de5e171ea341227e24bb1f090e1b771f7/config.json", "type": "primary" }, { "label": "Complete checkpoint index and bounded headers from all 44 shards", "url": "https://huggingface.co/stepfun-ai/Step-3.5-Flash/blob/ab446a3de5e171ea341227e24bb1f090e1b771f7/model.safetensors.index.json", "type": "primary" }, { "label": "Original Hugging Face backbone implementation", "url": "https://huggingface.co/stepfun-ai/Step-3.5-Flash/blob/ab446a3de5e171ea341227e24bb1f090e1b771f7/modeling_step3p5.py", "type": "primary" }, { "label": "Pinned vLLM implementation and loading rules for released Step 3.5 MTP tensors", "url": "https://github.com/vllm-project/vllm/blob/6fe3d447ac6429ef13aff117744d0298e28879ed/vllm/model_executor/models/step3p5_mtp.py", "type": "primary" }, { "label": "Pinned vLLM Step 3.5 decoder implementation", "url": "https://github.com/vllm-project/vllm/blob/6fe3d447ac6429ef13aff117744d0298e28879ed/vllm/model_executor/models/step3p5.py", "type": "primary" }, { "label": "Step 3.5 Flash technical report, architecture section", "url": "https://arxiv.org/html/2602.10604v1", "type": "primary" }, { "label": "Original checkpoint chat template", "url": "https://huggingface.co/stepfun-ai/Step-3.5-Flash/blob/ab446a3de5e171ea341227e24bb1f090e1b771f7/chat_template.jinja", "type": "primary" }, { "label": "Official Step 3.5 API model guide and variant distinction", "url": "https://platform.stepfun.ai/docs/en/guides/models/step-3.5-flash", "type": "primary" }, { "label": "Dated official Step 3.5 Flash launch announcement", "url": "https://www.stepfun.com/news/step-3-5-flash-launch", "type": "primary" }, { "label": "Original developer checkpoint history", "url": "https://huggingface.co/api/models/stepfun-ai/Step-3.5-Flash/commits/main", "type": "primary" }, { "label": "Exact hosted original Step 3.5 endpoint metadata", "url": "https://openrouter.ai/api/v1/models/stepfun/step-3.5-flash/endpoints", "type": "catalog" }, { "label": "Apache 2.0 license in official release repository", "url": "https://github.com/stepfun-ai/Step-3.5-Flash/blob/main/LICENSE", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/stepfun/step-3.5-flash.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/stepfun-ai/Step-3.5-Flash", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete original checkpoint: 199.38428928B learned parameters, comprising 196.956118272B backbone plus 2.428171008B across three MTP heads; excludes 12096 auxiliary router-bias values. The card’s 196B backbone and 196.81B one-head table are nominal and do not equal the complete three-head count. The 11B active figure is nominal with unstated vocabulary/MTP accounting. Configured context is 262144. Official public launch is documented on 2026-02-02; the earlier 2026-01-29 provider listing does not establish first serving, and first-public visibility before the announcement remains unresolved. This is the original text-only model, separate from 2603 and 3.7 revisions.", "parameter_components_b": { "backbone": 196.956118272, "mtp": 2.428171008 }, "field_status": {}, "audited_at": "2026-10-27T04:35:55.058599Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/494.json" }, { "id": "alibaba/qwen3-coder-next", "slug": "alibaba--qwen3-coder-next", "name": "Qwen3 Coder Next", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Open-weight Qwen coding model for agents, repository edits, and multi-turn tool use", "release_date": "2026-02", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 80, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-coder-next.toml", "source_type": "catalog", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-Coder-Next", "type": "weights" }, { "label": "Qwen3-Coder-Next configuration", "url": "https://huggingface.co/Qwen/Qwen3-Coder-Next/raw/main/config.json", "type": "primary" }, { "label": "Qwen3-Coder-Next announcement", "url": "https://qwen.ai/blog?id=qwen3-coder-next", "type": "primary" }, { "label": "Qwen3-Coder-Next commit history", "url": "https://huggingface.co/Qwen/Qwen3-Coder-Next/commits/main", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-coder-next.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Counts are nominal publisher figures for the exact Qwen/Qwen3-Coder-Next instruct checkpoint. Architecture combines Gated DeltaNet and gated attention with MoE. Hosted aliases may have distinct deployment revisions; no byte-identical hosted checkpoint mapping is asserted. Announcement indexing says February 2, while model-card release material is dated February 3; only month precision is asserted.", "field_status": {}, "audited_at": "2026-10-26T14:29:21.830785Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/034.json" }, { "id": "arcee-ai/trinity-large-preview", "slug": "arcee-ai--trinity-large-preview", "name": "Trinity Large Preview", "developer": "Arcee AI", "developer_id": "arcee-ai", "family": "trinity", "description": "Open-weight Trinity instruct model with sparse MoE routing, text generation, and tool calling.", "release_date": "2026-01-27", "date_precision": "day", "open_weights": true, "license": "OpenMDW-1.1", "parameters_b": 398.635286016, "active_parameters_b": 13, "parameters_source": "primary_source", "architecture": "moe", "context_window": 512000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://www.arcee.ai/blog/trinity-large", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Preview", "type": "weights" }, { "label": "Announcement", "url": "https://www.arcee.ai/blog/trinity-large", "type": "announcement" }, { "label": "Trinity-Large-Preview configuration", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Preview/raw/main/config.json", "type": "primary" }, { "label": "Trinity-Large-Preview OpenMDW-1.1 license", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Preview/raw/main/LICENSE", "type": "primary" }, { "label": "Arcee Trinity Large Technical Report", "url": "https://raw.githubusercontent.com/arcee-ai/trinity-large-tech-report/main/Arcee%20Trinity%20Large.pdf", "type": "primary" }, { "label": "Trinity-Large-Preview chat and tool template", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Preview/raw/main/chat_template.jinja", "type": "primary" }, { "label": "Trinity-Large-Preview pinned checkpoint index", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Preview/raw/d8e126719c8bf2b3ac5a3551bd1f07fcee767a90/model.safetensors.index.json", "type": "primary" }, { "label": "Trinity-Large-Preview pinned checkpoint inventory", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Preview/tree/d8e126719c8bf2b3ac5a3551bd1f07fcee767a90", "type": "primary" }, { "label": "Trinity-Large-Preview release history", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Preview/commits/main", "type": "primary" }, { "label": "Trinity-Large-Thinking release comparison", "url": "https://www.arcee.ai/blog/trinity-large-thinking", "type": "primary" }, { "label": "Arcee open-model catalog and exact Preview comparison", "url": "https://www.arcee.ai/open-source-catalog", "type": "primary" }, { "label": "Arcee Afmoe implementation for the explicitly shared architecture", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Thinking/raw/44bc97dcbb24d4c0e43a97d6abd7b0ff25f0621a/modeling_afmoe.py", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/arcee-ai/trinity-large-preview.toml", "type": "catalog" }, { "label": "License", "url": "https://huggingface.co/arcee-ai/Trinity-Large-Preview/blob/main/LICENSE", "type": "license" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete checkpoint count is 398.635286016B, including frozen router biases; developer labels 398B/400B are nominal. Active 13B is an approximate developer per-token count with 4 routed plus 1 shared expert. Context 512000 is decimal normalization of the advertised 512K extended inference capacity, not a source-verified exact boundary; the released config specifies 262144 positions. Preview is non-reasoning. OpenMDW-1.1 is the current license adopted May 28, 2026.", "parameter_components_b": { "routed_experts": 391.378894848, "shared_experts": 1.528823808, "dense_mlp": 0.679477248, "attention": 3.77488896, "routers": 0.042467328, "input_embeddings": 0.614989824, "output_head": 0.614989824, "normalization": 0.000740352, "frozen_routing_bias": 0.000013824 }, "field_status": {}, "audited_at": "2026-10-26T17:21:26.569139+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/118.json" }, { "id": "deepseek/deepseek-ocr-2", "slug": "deepseek--deepseek-ocr-2", "name": "DeepSeek-OCR 2", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-ocr", "description": "Open-weight OCR vision-language model that combines DeepEncoder V2 with a mixture-of-experts language decoder for text extraction and grounded document-to-Markdown conversion.", "release_date": "2026-01-27", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 3.38911936, "active_parameters_b": 0.5, "parameters_source": "primary_source", "architecture": "moe", "context_window": 8192, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-OCR-2", "source_type": "primary", "sources": [ { "label": "Official DeepSeek-OCR release history", "url": "https://github.com/deepseek-ai/DeepSeek-OCR#release", "type": "primary" }, { "label": "DeepSeek-OCR 2: Visual Causal Flow", "url": "https://arxiv.org/html/2601.20552v1", "type": "primary" }, { "label": "Official DeepSeek-OCR-2 model card", "url": "https://huggingface.co/deepseek-ai/DeepSeek-OCR-2", "type": "primary" }, { "label": "Official checkpoint configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-OCR-2/blob/aaa02f3811945a91062062994c5c4a3f4c0af2b0/config.json", "type": "primary" }, { "label": "Official weight repository Apache 2.0 license", "url": "https://huggingface.co/deepseek-ai/DeepSeek-OCR-2/blob/aaa02f3811945a91062062994c5c4a3f4c0af2b0/LICENSE.txt", "type": "primary" }, { "label": "Official checkpoint safetensors header", "url": "https://huggingface.co/deepseek-ai/DeepSeek-OCR-2/resolve/aaa02f3811945a91062062994c5c4a3f4c0af2b0/model-00001-of-000001.safetensors", "type": "primary" }, { "label": "Official DeepEncoder V2 implementation", "url": "https://huggingface.co/deepseek-ai/DeepSeek-OCR-2/blob/aaa02f3811945a91062062994c5c4a3f4c0af2b0/deepencoderv2.py", "type": "primary" }, { "label": "Official OCR2 multimodal implementation", "url": "https://huggingface.co/deepseek-ai/DeepSeek-OCR-2/blob/aaa02f3811945a91062062994c5c4a3f4c0af2b0/modeling_deepseekocr2.py", "type": "primary" }, { "label": "Official OCR2 language decoder implementation", "url": "https://huggingface.co/deepseek-ai/DeepSeek-OCR-2/blob/aaa02f3811945a91062062994c5c4a3f4c0af2b0/modeling_deepseekv2.py", "type": "primary" }, { "label": "Official OCR2 tokenizer configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-OCR-2/blob/aaa02f3811945a91062062994c5c4a3f4c0af2b0/tokenizer_config.json", "type": "primary" }, { "label": "Official repository metadata and checkpoint summary", "url": "https://huggingface.co/api/models/deepseek-ai/DeepSeek-OCR-2", "type": "primary" }, { "label": "Official checkpoint index", "url": "https://huggingface.co/deepseek-ai/DeepSeek-OCR-2/blob/aaa02f3811945a91062062994c5c4a3f4c0af2b0/model.safetensors.index.json", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepseek/deepseek-ocr-2.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete released checkpoint: 3.38911936B learned parameters. Approximately 0.5B active parameters refers only to the language decoder, per the developer report; image encoding is separate. Decoder context is 8192; the 256–1120 tokens per image are a separate visual budget. Native thinking mode and tool calling are not established by the developer materials.", "parameter_components_b": { "language_decoder": 2.93473408, "vision_tokenizer": 0.094979328, "vision_causal_encoder": 0.358256512, "projector": 0.00114816, "view_separator": 0.00000128 }, "active_parameters_by_phase_b": { "decoder": 0.5 }, "field_status": { "reasoning": "undisclosed", "tool_call": "undisclosed" }, "audited_at": "2026-10-27T12:37:43.440555+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/167.json", "active_parameters_scope": "language_decoder" }, { "id": "moonshotai/kimi-k2.5", "slug": "moonshotai--kimi-k2.5", "name": "Kimi K2.5", "developer": "Moonshot AI", "developer_id": "moonshotai", "family": "kimi-k2", "description": "Open-weight multimodal MoE model built on Kimi K2, with thinking and instant modes, 256K context, and support for visual understanding, coding, and tool use.", "release_date": "2026-01-27", "date_precision": "day", "open_weights": true, "license": "Modified MIT", "parameters_b": 1026.879376368, "active_parameters_b": 32, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.kimi.com/en/help/agent/agent-swarm", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/moonshotai/Kimi-K2.5", "type": "weights" }, { "label": "Kimi-K2.5 released checkpoint configuration", "url": "https://huggingface.co/moonshotai/Kimi-K2.5/blob/4d01dfe0332d63057c186e0b262165819efb6611/config.json", "type": "weights" }, { "label": "Kimi K2.5 Modified MIT License", "url": "https://huggingface.co/moonshotai/Kimi-K2.5/blob/4d01dfe0332d63057c186e0b262165819efb6611/LICENSE", "type": "primary" }, { "label": "Agent Swarm Capabilities — release history", "url": "https://www.kimi.com/en/help/agent/agent-swarm", "type": "primary" }, { "label": "Kimi K2.5: Visual Agentic Intelligence — technical report", "url": "https://arxiv.org/html/2602.02276v1", "type": "primary" }, { "label": "Kimi-K2.5 complete safetensors index and referenced shard headers", "url": "https://huggingface.co/moonshotai/Kimi-K2.5/blob/4d01dfe0332d63057c186e0b262165819efb6611/model.safetensors.index.json", "type": "weights" }, { "label": "Released Kimi K2.5 language-model implementation", "url": "https://huggingface.co/moonshotai/Kimi-K2.5/blob/4d01dfe0332d63057c186e0b262165819efb6611/modeling_deepseek.py", "type": "weights" }, { "label": "Released Kimi K2.5 vision and projector implementation", "url": "https://huggingface.co/moonshotai/Kimi-K2.5/blob/4d01dfe0332d63057c186e0b262165819efb6611/modeling_kimi_k25.py", "type": "weights" }, { "label": "Kimi K2.5: Visual Agentic Intelligence — launch blog", "url": "https://www.kimi.com/en/blog/kimi-k2-5", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/moonshotai/kimi-k2.5.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Exact complete checkpoint total is 1,026,879,376,368 logical parameters (1,026,408,232,448 language; 416,866,032 vision; 54,277,888 projector), reconstructed from all 64 shard headers and independently checked against configuration and implementation. Packed INT4 storage is unpacked; quantization scales, shape metadata, and nonparameter buffers are excluded. The model card’s 1T total, 32B activated, and 400M vision figures are nominal. The published active count is retained without asserting an exact text/vision phase convention.", "parameter_components_b": { "language_backbone": 1026.408232448, "vision_encoder": 0.416866032, "vision_projector": 0.054277888 }, "field_status": {}, "audited_at": "2026-10-26T23:56:31.007834Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/359.json" }, { "id": "upstage/solar-pro3", "slug": "upstage--solar-pro3", "name": "Solar Pro 3", "developer": "Upstage", "developer_id": "upstage", "family": "solar-pro", "description": "Upstage text-only mixture-of-experts model with reasoning and tool calling, supporting Korean, English and Japanese.", "release_date": "2026-01-26", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 102.6, "active_parameters_b": 12, "parameters_source": "primary_source", "architecture": "moe", "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.upstage.ai/blog/en/solar-pro-3-0127", "source_type": "primary", "sources": [ { "label": "Solar Pro 3 model specifications and version history", "url": "https://console.upstage.ai/docs/models/solar-pro-3", "type": "primary" }, { "label": "Solar Pro 3: Better reasoning at production scale, January 26, 2026", "url": "https://www.upstage.ai/blog/en/solar-pro-3-0127", "type": "primary" }, { "label": "Solar Pro 3: 2x Agentic Performance — What Changed, March 24, 2026", "url": "https://www.upstage.ai/blog/en/solar-pro-3-0323", "type": "primary" }, { "label": "Upstage AI Champion/Rookie competition presentation, slide 9", "url": "https://file.aichallenge4all.or.kr/notices/1775173912481_%EC%97%85%EC%8A%A4%ED%85%8C%EC%9D%B4%EC%A7%80-AI_%EB%AA%A8%EB%8D%B8_%EB%B0%8F_API_%ED%99%9C%EC%9A%A9%EB%B2%95.pdf", "type": "primary" }, { "label": "Solar Pro 3 endpoint metadata", "url": "https://openrouter.ai/api/v1/models/upstage/solar-pro-3/endpoints", "type": "primary" }, { "label": "Terms of Service, Articles 7, 8 and 23", "url": "https://www.upstage.ai/terms-of-service", "type": "primary" }, { "label": "Solar Pro 3 tokenizer card and file listing", "url": "https://huggingface.co/upstage/solar-pro3-tokenizer", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/upstage/solar-pro3.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released January 26, 2026; the solar-pro3 alias now points to the March 23 update. Upstage reports 102B nominally, while its Solar Pro 3 presentation specifies 102.6B total and 12B active; these are author-reported counts, not an independently counted checkpoint. Native Upstage documentation specifies 128,000 context tokens; its OpenRouter route advertises 131,072. Public tokenizer assets do not constitute a weight release. Upstage schedules all Solar Pro 3 versions for retirement on October 30, 2026 KST.", "field_status": {}, "audited_at": "2026-10-27T05:14:31.695957Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/517.json" }, { "id": "minimax/MiniMax-M2-Her", "slug": "minimax--minimax-m2-her", "name": "MiniMax-M2-her", "developer": "MiniMax", "developer_id": "minimax", "family": "minimax", "description": "MiniMax dialogue model optimized for role-playing and multi-turn conversations, with character and user role settings and example-dialogue prompting.", "release_date": "2026-01-23", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 65536, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": false, "source_url": "https://openrouter.ai/minimax/minimax-m2-her-20260123", "source_type": "primary", "sources": [ { "label": "MiniMax M2-her technical introduction", "url": "https://www.minimax.io/blog/minimax-m2-her", "type": "primary" }, { "label": "MiniMax M2-her API guide", "url": "https://platform.minimax.io/docs/guides/text-chat", "type": "primary" }, { "label": "MiniMax M2-her dedicated OpenAPI schema", "url": "https://platform.minimax.io/docs/api-reference/text/api/openapi-chat.json", "type": "primary" }, { "label": "MiniMax official reply about Her weights", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5/discussions/2", "type": "primary" }, { "label": "MiniMax FY2025 business review", "url": "https://www.minimax.io/news/minimax-global-announces-full-year-2025-financial-results", "type": "primary" }, { "label": "MiniMax retired model list", "url": "https://platform.minimax.io/docs/faq/models", "type": "primary" }, { "label": "OpenRouter MiniMax M2-her endpoint metadata", "url": "https://openrouter.ai/api/v1/models/minimax/minimax-m2-her/endpoints", "type": "catalog" }, { "label": "MiniMax-M2-her technical deep dive, Chinese edition", "url": "https://www.minimax.cn/blog/minimax-m2-her", "type": "primary" }, { "label": "MiniMax FY2025 annual results announcement, Chinese", "url": "https://www.hkexnews.hk/listedco/listconews/sehk/2026/0302/2026030202838_c.pdf", "type": "primary" }, { "label": "MiniMax Role-play Benchmark dataset card and comparison table", "url": "https://huggingface.co/datasets/MiniMaxAI/role-play-bench/raw/main/README.md", "type": "primary" }, { "label": "MiniMax-M3.1-Flash official model card", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M3.1-Flash/raw/main/README.md", "type": "primary" }, { "label": "MiniMax native model release chronology", "url": "https://platform.minimax.io/docs/release-notes/models.md", "type": "primary" }, { "label": "MiniMax native API release chronology", "url": "https://platform.minimax.io/docs/release-notes/apis.md", "type": "primary" }, { "label": "Original December M2.1 announcement checked for Her", "url": "https://www.minimax.io/news/minimax-m21", "type": "primary" }, { "label": "OpenRouter original MiniMax M2-her release and provider page", "url": "https://openrouter.ai/minimax/minimax-m2-her-20260123", "type": "primary" }, { "label": "MiniMax direct Her API guide revisited", "url": "https://platform.minimax.io/docs/guides/text-chat.md", "type": "primary" }, { "label": "Puter original MiniMax M2-her integration announcement", "url": "https://developer.puter.com/blog/minimax-m2-her-in-puter-js/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/minimax/MiniMax-M2-Her.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "MiniMax-M2-her uses the direct API ID M2-her. The public API product launched on OpenRouter on January 23, 2026 with MiniMax serving it; earlier Q4 2025 consumer-product use is distinct from this API launch. MiniMax published its technical introduction on January 27. The official model list now marks M2-her retired. Context 65536 is the last-documented MiniMax-provider limit; the native guide says 64 K. The Her API has no documented native reasoning output or tool calls. Parameter counts and architecture remain undisclosed; no developer source establishes Her’s named base. Ordinary M2 specifications and licenses are not inherited.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-28T08:01:37.174468Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/301.json" }, { "id": "liquid/lfm2.5-1.2b-thinking", "slug": "liquid--lfm2.5-1.2b-thinking", "name": "LFM2.5-1.2B-Thinking", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm", "description": null, "release_date": "2026-01-20", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 1.17, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "hybrid", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/LiquidAI/LFM2.5-1.2B-Thinking/blob/f313478934a7612d22991f752959d7a1a8756fec/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/LiquidAI/LFM2.5-1.2B-Thinking/blob/f313478934a7612d22991f752959d7a1a8756fec/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:23.459Z", "last_seen": "2026-10-29T10:17:23.459Z", "upstream_present": false, "availability": "public", "notes": "Distinct reasoning post-training of Liquid’s own 1.2B lineage; published model-card count and supported context." }, { "id": "zhipuai/glm-4.7-flash", "slug": "zhipuai--glm-4.7-flash", "name": "GLM-4.7-Flash", "developer": "Z.ai", "developer_id": "zhipuai", "family": "glm-flash", "description": "Open-weight text MoE model with MLA attention, reasoning and tool calling; published as 30B total and 3B active parameters.", "release_date": "2026-01-19", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 30, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 202752, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://docs.bigmodel.cn/cn/update/new-releases", "source_type": "primary", "sources": [ { "label": "GLM-4.7-Flash official model card", "url": "https://huggingface.co/zai-org/GLM-4.7-Flash/blob/7dd20894a642a0aa287e9827cb1a1f7f91386b67/README.md", "type": "weights" }, { "label": "GLM-4.7-Flash checkpoint configuration", "url": "https://huggingface.co/zai-org/GLM-4.7-Flash/blob/7dd20894a642a0aa287e9827cb1a1f7f91386b67/config.json", "type": "weights" }, { "label": "BigModel model and product release log", "url": "https://docs.bigmodel.cn/cn/update/new-releases", "type": "primary" }, { "label": "GLM-4.7-Flash model guide", "url": "https://docs.bigmodel.cn/cn/guide/models/free/glm-4.7-flash", "type": "primary" }, { "label": "GLM-4.7-Flash tensor index and all 48 bounded shard headers", "url": "https://huggingface.co/zai-org/GLM-4.7-Flash/blob/7dd20894a642a0aa287e9827cb1a1f7f91386b67/model.safetensors.index.json", "type": "weights" }, { "label": "Transformers GLM-4 MoE Lite implementation", "url": "https://github.com/huggingface/transformers/blob/c0d024fbc873ab0325e1b40fc1bd69069609de65/src/transformers/models/glm4_moe_lite/modeling_glm4_moe_lite.py", "type": "primary" }, { "label": "vLLM GLM-4 MoE Lite multi-token predictor", "url": "https://github.com/vllm-project/vllm/blob/5d35ce3934a7804b298559d2d99eec33792e1eb9/vllm/model_executor/models/glm4_moe_lite_mtp.py", "type": "primary" }, { "label": "SGLang GLM-4 MoE Lite loader", "url": "https://github.com/sgl-project/sglang/blob/db90f584d2f4ab15f0823f82614ec7dce23e5dcd/python/sglang/srt/models/glm4_moe_lite.py", "type": "primary" }, { "label": "Z.ai GLM-4.7 series API guide", "url": "https://docs.z.ai/guides/llm/glm-4.7", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/zhipuai/glm-4.7-flash.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/GLM-4.7-Flash", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "30B total and 3B active are Z.ai’s nominal labels. The checkpoint includes one MTP layer and stored embedding/output copies; supported runtimes handle sharing differently, so its 31.221485568B BF16 storage-element sum is not treated as an exact unique model size. Native context is 202752 tokens; API documentation rounds this to 200K. These specifications describe GLM-4.7-Flash, not FlashX or the older GLM-4.5-Flash endpoint.", "field_status": {}, "audited_at": "2026-10-27T06:16:31.482284Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/555.json" }, { "id": "zhipuai/glm-4.7-flashx", "slug": "zhipuai--glm-4.7-flashx", "name": "GLM-4.7-FlashX", "developer": "Z.ai", "developer_id": "zhipuai", "family": "glm-flash", "description": "Hosted GLM text model with optional reasoning and function calling.", "release_date": "2026-01-19", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 200000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://docs.z.ai/guides/llm/glm-4.7", "source_type": "primary", "sources": [ { "label": "Official GLM-4.7 family API guide", "url": "https://docs.z.ai/guides/llm/glm-4.7", "type": "primary" }, { "label": "Official Chinese GLM-4.7 API guide", "url": "https://docs.bigmodel.cn/cn/guide/models/text/glm-4.7", "type": "primary" }, { "label": "Native BigModel Chat Completion API specification", "url": "https://docs.bigmodel.cn/api-reference/模型-api/对话补全", "type": "primary" }, { "label": "Official API prices", "url": "https://docs.z.ai/guides/overview/pricing", "type": "primary" }, { "label": "Original January 19 GLM-4.7-Flash launch post", "url": "https://x.com/Zai_org/status/2013261304060866758", "type": "primary" }, { "label": "Full text mirror of original Z.ai launch post", "url": "https://api.fxtwitter.com/Zai_org/status/2013261304060866758", "type": "docs" }, { "label": "Official Z.ai public checkpoint inventory", "url": "https://huggingface.co/api/models?author=zai-org&limit=1000", "type": "primary" }, { "label": "Related GLM-4.7-Flash model card", "url": "https://huggingface.co/zai-org/GLM-4.7-Flash/blob/main/README.md", "type": "primary" }, { "label": "Later GLM-5 technical report", "url": "https://arxiv.org/html/2602.15763v2", "type": "primary" }, { "label": "Original developer model download comparison", "url": "https://github.com/zai-org/GLM-4.5/blob/main/README.md", "type": "primary" }, { "label": "Z.ai API Terms of Use", "url": "https://docs.z.ai/legal-agreement/terms-of-use", "type": "primary" }, { "label": "Original Vercel gateway FlashX service page", "url": "https://vercel.com/ai-gateway/models/glm-4.7-flashx", "type": "docs" }, { "label": "PoliLegalLM original comparison and local-evaluation protocol", "url": "https://arxiv.org/html/2604.17543v1", "type": "primary" }, { "label": "Official GLM-4.7-Flash open-source launch", "url": "https://www.zhipuai.cn/zh/news/148", "type": "primary" }, { "label": "Official GLM-5 card used to check comparison-table accuracy", "url": "https://huggingface.co/zai-org/GLM-5/raw/main/README.md", "type": "primary" }, { "label": "Z.ai GLM 4.7 FlashX API guide", "url": "https://docs.z.ai/guides/llm/glm-4.7.md", "type": "primary" }, { "label": "Public GLM 4.7 Flash model card", "url": "https://huggingface.co/zai-org/GLM-4.7-Flash/raw/main/README.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/zhipuai/glm-4.7-flashx.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "GLM-4.7-FlashX is a paid high-speed text API with a 200,000-token context. Total and active parameters remain unverified: PoliLegalLM lists 30B/3B, but its local-model description leaves possible confusion with the separate Flash checkpoint unresolved. No developer statement mapping FlashX to that checkpoint was established. No exact public FlashX checkpoint was found; proprietary describes the hosted API terms. Classified closed for FlashX because no matching public weight release was found.", "field_status": { "parameters_b": "unverified", "active_parameters_b": "unverified", "architecture": "undisclosed" }, "audited_at": "2026-10-28T08:06:39.122919Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/556.json" }, { "id": "ai21/jamba-mini", "slug": "ai21--jamba-mini", "name": "Jamba Mini", "developer": "AI21 Labs", "developer_id": "ai21", "family": "jamba", "description": "AI21 Jamba2 Mini, a text-generation model with hybrid Mamba-Transformer layers and mixture-of-experts routing, released for enterprise question answering and instruction following.", "release_date": "2026-01-08", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 52, "active_parameters_b": 12, "parameters_source": "primary_source", "architecture": "moe", "context_window": 256000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/ai21/jamba-mini.toml", "source_type": "catalog", "sources": [ { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/ai21/jamba-mini.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/ai21labs/AI21-Jamba2-Mini", "type": "weights" }, { "label": "Official Jamba model details and alias mapping", "url": "https://docs.ai21.com/docs/jamba-foundation-models", "type": "primary" }, { "label": "Official Jamba2 launch", "url": "https://www.ai21.com/blog/introducing-jamba2/", "type": "primary" }, { "label": "Official Jamba2 Mini configuration", "url": "https://huggingface.co/ai21labs/AI21-Jamba2-Mini/blob/main/config.json", "type": "primary" }, { "label": "Official function calling documentation", "url": "https://docs.ai21.com/docs/function-calling", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Current documented jamba-mini alias maps to jamba-mini-2-2026-01 (Jamba2 Mini). Parameter counts are nominal developer figures: 52B total, 12B active. The MoE routing sits within a hybrid Mamba-Transformer backbone. API context is published as 256K; the downloadable checkpoint config specifies 262144, while 256000 is only the catalog decimal normalization. API callability was not tested.", "field_status": {}, "audited_at": "2026-10-26T13:49:46.567737Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/005.json" }, { "id": "liquid/lfm2.5-1.2b-base", "slug": "liquid--lfm2.5-1.2b-base", "name": "LFM2.5-1.2B-Base", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm", "description": null, "release_date": "2026-01-05", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 1.17, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "hybrid", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.liquid.ai/blog/introducing-lfm2-5-the-next-generation-of-on-device-ai", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/introducing-lfm2-5-the-next-generation-of-on-device-ai", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/LiquidAI/LFM2.5-1.2B-Base/blob/7453bca97ca1e67754c4035a4b4c584e1c9dd725/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/LiquidAI/LFM2.5-1.2B-Base/blob/7453bca97ca1e67754c4035a4b4c584e1c9dd725/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:23.459Z", "last_seen": "2026-10-29T10:17:23.459Z", "upstream_present": false, "availability": "public", "notes": "Original January 5 release, built on Liquid’s independently pretrained LFM lineage. Published model-card parameter count." }, { "id": "liquid/lfm2.5-1.2b-instruct", "slug": "liquid--lfm2.5-1.2b-instruct", "name": "LFM2.5-1.2B-Instruct", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm", "description": null, "release_date": "2026-01-05", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 1.17, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "hybrid", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://www.liquid.ai/blog/introducing-lfm2-5-the-next-generation-of-on-device-ai", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/introducing-lfm2-5-the-next-generation-of-on-device-ai", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/LiquidAI/LFM2.5-1.2B-Instruct/blob/0f604ada3f766f9f257460c4c9f0b5d6f69d431b/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/LiquidAI/LFM2.5-1.2B-Instruct/blob/0f604ada3f766f9f257460c4c9f0b5d6f69d431b/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:23.459Z", "last_seen": "2026-10-29T10:17:23.459Z", "upstream_present": false, "availability": "public", "notes": "Original January 5 release, built on Liquid’s independently pretrained LFM lineage. Published model-card parameter count." }, { "id": "liquid/lfm2.5-1.2b-jp", "slug": "liquid--lfm2.5-1.2b-jp", "name": "LFM2.5-1.2B-JP", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm", "description": null, "release_date": "2026-01-05", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 1.2, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "hybrid", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.liquid.ai/blog/introducing-lfm2-5-the-next-generation-of-on-device-ai", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/introducing-lfm2-5-the-next-generation-of-on-device-ai", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/LiquidAI/LFM2.5-1.2B-JP/blob/58ef35ca21bb2d3ef6e320a938765a9727aefc5d/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/LiquidAI/LFM2.5-1.2B-JP/blob/58ef35ca21bb2d3ef6e320a938765a9727aefc5d/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:23.459Z", "last_seen": "2026-10-29T10:17:23.459Z", "upstream_present": false, "availability": "public", "notes": "Original Japanese-tuned January release, distinct from the June 2026 refresh. Nominal size from the release announcement; a supported context limit for this exact revision was not established." }, { "id": "liquid/lfm2.5-audio-1.5b", "slug": "liquid--lfm2.5-audio-1.5b", "name": "LFM2.5-Audio-1.5B", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm", "description": null, "release_date": "2026-01-05", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 1.5, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "hybrid", "context_window": 32768, "modalities": { "input": [ "text", "audio" ], "output": [ "text", "audio" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.liquid.ai/blog/introducing-lfm2-5-the-next-generation-of-on-device-ai", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/introducing-lfm2-5-the-next-generation-of-on-device-ai", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/LiquidAI/LFM2.5-Audio-1.5B/blob/c362a0625dfe45aa588dce5f0ada28a7e5707628/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/LiquidAI/LFM2.5-Audio-1.5B/blob/c362a0625dfe45aa588dce5f0ada28a7e5707628/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:23.459Z", "last_seen": "2026-10-29T10:17:23.459Z", "upstream_present": false, "availability": "public", "notes": "Native conversational audio-language model with text and speech input/output. The 1.5B nominal total includes the audio components; the language backbone is 1.2B. Liquid trains the model on its own language backbone, with an NVIDIA FastConformer audio encoder component." }, { "id": "liquid/lfm2.5-vl-1.6b", "slug": "liquid--lfm2.5-vl-1.6b", "name": "LFM2.5-VL-1.6B", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm", "description": null, "release_date": "2026-01-05", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 1.6, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "hybrid", "context_window": 32768, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.liquid.ai/blog/introducing-lfm2-5-the-next-generation-of-on-device-ai", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/introducing-lfm2-5-the-next-generation-of-on-device-ai", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/LiquidAI/LFM2.5-VL-1.6B/blob/919fde3d022e3f90a4716006f993938ee8c2eb97/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/LiquidAI/LFM2.5-VL-1.6B/blob/919fde3d022e3f90a4716006f993938ee8c2eb97/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:23.459Z", "last_seen": "2026-10-29T10:17:23.459Z", "upstream_present": false, "availability": "public", "notes": "Vision-language model on Liquid’s own pretrained language backbone with a SigLIP2 vision encoder. Published nominal total includes both components." }, { "id": "minimax/MiniMax-M2.1", "slug": "minimax--minimax-m2.1", "name": "MiniMax-M2.1", "developer": "MiniMax", "developer_id": "minimax", "family": "minimax", "description": "Open-weight MiniMax M2-series text model for multilingual coding, tool use, and office workflows, with interleaved reasoning.", "release_date": "2025-12-22", "date_precision": "day", "open_weights": true, "license": "Modified MIT (MiniMax M2.1)", "parameters_b": 230, "active_parameters_b": 10, "parameters_source": "primary_source", "architecture": "moe", "context_window": 196608, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://platform.minimax.io/docs/release-notes/models", "source_type": "primary", "sources": [ { "label": "MiniMax M2.1: Post-Training Experience and Insights for Agent Models", "url": "https://www.minimax.io/news/post-training-experience-and-insights-for-agent-models", "type": "primary" }, { "label": "MiniMax M2.1 launch announcement", "url": "https://www.minimax.io/news/minimax-m21", "type": "primary" }, { "label": "MiniMax API model release notes", "url": "https://platform.minimax.io/docs/release-notes/models", "type": "primary" }, { "label": "MiniMax-M2.1 checkpoint configuration at 917dbb7", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.1/raw/917dbb7d0761e24d8623c0a5d97624550181c8e6/config.json", "type": "weights" }, { "label": "MiniMax-M2.1 tensor index and all shard headers at 917dbb7", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.1/resolve/917dbb7d0761e24d8623c0a5d97624550181c8e6/model.safetensors.index.json", "type": "weights" }, { "label": "MiniMax-M2.1 reference model implementation at 917dbb7", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.1/raw/917dbb7d0761e24d8623c0a5d97624550181c8e6/modeling_minimax_m2.py", "type": "weights" }, { "label": "Hugging Face", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.1", "type": "weights" }, { "label": "MiniMax-M2.1 Modified MIT license", "url": "https://raw.githubusercontent.com/MiniMax-AI/MiniMax-M2.1/main/LICENSE", "type": "primary" }, { "label": "MiniMax-M2.1 official vLLM deployment guide at 917dbb7", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.1/raw/917dbb7d0761e24d8623c0a5d97624550181c8e6/docs/vllm_deploy_guide.md", "type": "weights" }, { "label": "MiniMax-M2.1 official tool-calling guide at 917dbb7", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.1/raw/917dbb7d0761e24d8623c0a5d97624550181c8e6/docs/tool_calling_guide.md", "type": "weights" }, { "label": "MiniMax-M2.1 chat template at 917dbb7", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.1/raw/917dbb7d0761e24d8623c0a5d97624550181c8e6/chat_template.jinja", "type": "weights" }, { "label": "The MiniMax-M2 Series: Mini Activations Unleashing Max Real-World Intelligence", "url": "https://arxiv.org/html/2605.26494v1", "type": "primary" }, { "label": "MiniMax-M2.1 official checkpoint commit history", "url": "https://huggingface.co/api/models/MiniMaxAI/MiniMax-M2.1/commits/main", "type": "weights" }, { "label": "Current MiniMax model invocation documentation", "url": "https://platform.minimax.io/docs/guides/text-generation.md", "type": "primary" }, { "label": "MiniMax OpenAI-compatible API documentation", "url": "https://platform.minimax.io/docs/api-reference/text-openai-api", "type": "primary" }, { "label": "MiniMax-M2.1 repository metadata", "url": "https://huggingface.co/api/models/MiniMaxAI/MiniMax-M2.1", "type": "weights" }, { "label": "MiniMax-M2.1 official repository file tree", "url": "https://api.github.com/repos/MiniMax-AI/MiniMax-M2.1/git/trees/main?recursive=1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/minimax/MiniMax-M2.1.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "MiniMax explicitly describes M2.1 nominally as ~230B total/~10B active. Header audit finds 228.689764864B non-FP8-scale elements, including 15872 routing-bias values whose original training status is unresolved for M2.1; do not treat inference buffer registration as proof they were not learned. The later report does not expressly identify M2.1, so its M2-specific finer counts are not transferred. Native context is 196608 tokens; current primary API tables no longer specify M2.1. Earliest explicit API release note is 2025-12-22 versus a 2025-12-23 launch article. Modified MIT requires prominent MiniMax M2.1 UI attribution for commercial uses.", "field_status": {}, "audited_at": "2026-10-26T22:24:13.376175Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/302.json" }, { "id": "zhipuai/glm-4.7", "slug": "zhipuai--glm-4.7", "name": "GLM-4.7", "developer": "Z.ai", "developer_id": "zhipuai", "family": "glm", "description": "Text MoE model for coding, reasoning and tool use, with interleaved and preserved thinking.", "release_date": "2025-12-22", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 355, "active_parameters_b": 32, "parameters_source": "primary_source", "architecture": "moe", "context_window": 202752, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.prnewswire.com/news-releases/zai-releases-glm-4-7-designed-for-real-world-development-environments-cementing-itself-as-chinas-openai-302649821.html", "source_type": "primary", "sources": [ { "label": "Z.ai GLM-4.7 release statement", "url": "https://www.prnewswire.com/news-releases/zai-releases-glm-4-7-designed-for-real-world-development-environments-cementing-itself-as-chinas-openai-302649821.html", "type": "primary" }, { "label": "GLM-4.7 exact checkpoint card", "url": "https://huggingface.co/zai-org/GLM-4.7/blob/602d01efcdd332c5238ca4bcede555defbe83eb7/README.md", "type": "weights" }, { "label": "Official model download table at pinned revision", "url": "https://github.com/zai-org/GLM-4.5/blob/170f20b2c10659008fdbc909d478bc2a75bc3627/README.md", "type": "primary" }, { "label": "GLM-4.7 exact checkpoint configuration", "url": "https://huggingface.co/zai-org/GLM-4.7/blob/602d01efcdd332c5238ca4bcede555defbe83eb7/config.json", "type": "weights" }, { "label": "GLM-4.7 native API documentation", "url": "https://docs.z.ai/guides/llm/glm-4.7", "type": "primary" }, { "label": "GLM-4.7 provider endpoint metadata", "url": "https://openrouter.ai/api/v1/models/z-ai/glm-4.7/endpoints", "type": "primary" }, { "label": "Public GLM-4.7 checkpoint metadata and chat template", "url": "https://huggingface.co/api/models/zai-org/GLM-4.7", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/zhipuai/glm-4.7.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/GLM-4.7", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Z.ai lists GLM-4.7 as nominal 355B total and 32B active in its exact download row. These are not exact tensor counts; the table does not resolve embeddings, output-head or MTP inclusion, and the explicit GLM-4.5 counting convention is not transferred. The released checkpoint and Z.AI provider route specify 202752 tokens, rounded to 200K in native docs; Novita advertises 204800 for its route. The 131072 output limit is separate. Native modalities are text input and text output.", "field_status": {}, "audited_at": "2026-10-27T06:12:30.673036Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/554.json" }, { "id": "aion-labs/aion-2.0", "slug": "aion-labs--aion-2.0", "name": "Aion-2.0", "developer": "Aion Labs", "developer_id": "aion-labs", "family": "deepseek", "description": "Aion Labs variant of DeepSeek V3.2 optimized for roleplay and storytelling, with reasoning and tool calling.", "release_date": "2025-12-21", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": 671, "active_parameters_b": 37, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.aionlabs.ai/docs/models/", "source_type": "primary", "sources": [ { "label": "Aion Labs model documentation", "url": "https://www.aionlabs.ai/docs/models/", "type": "primary" }, { "label": "Aion Labs models API", "url": "https://api.aionlabs.ai/v1/models", "type": "primary" }, { "label": "Aion Labs API reference", "url": "https://www.aionlabs.ai/docs/api-reference/", "type": "primary" }, { "label": "Aion Labs official model organization", "url": "https://huggingface.co/aion-labs", "type": "primary" }, { "label": "Aion Labs API service terms", "url": "https://www.aionlabs.ai/terms/", "type": "primary" }, { "label": "OpenRouter listing (different provider date)", "url": "https://openrouter.ai/aion-labs/aion-2.0", "type": "catalog" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/aion-labs/aion-2.0.toml", "type": "catalog" }, { "label": "DeepSeek-V3.2 official model card", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/raw/main/README.md", "type": "primary" }, { "label": "DeepSeek nominal V3.2 specifications (Table 1)", "url": "https://arxiv.org/html/2606.19348v1#S4.T1", "type": "primary" }, { "label": "DeepSeek-V3.2 configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/raw/main/config.json", "type": "primary" }, { "label": "DeepSeek-V3.2 base configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2-Exp-Base/raw/main/config.json", "type": "primary" }, { "label": "DeepSeek-V3.2 technical report", "url": "https://arxiv.org/html/2512.02556v1#S2", "type": "primary" }, { "label": "OminiGate fine-tune description (corroboration)", "url": "https://ominigate.ai/en/models/aion-labs/aion-2.0", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Nominal 671B main-model parameters, 37B active and MoE are inferred from Aion Labs explicit DeepSeek V3.2 variant lineage and DeepSeek specifications; Aion has not published separate counts or structural changes. These are inherited nominal specifications, not an exact complete checkpoint count. Developer release date: 2025-12-21; OpenRouter lists 2026-02-23.", "field_status": { "license": "undisclosed" }, "audited_at": "2026-10-27T07:24:48.078Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/006.json", "parameters_scope": "main_model", "active_parameters_scope": "main_model", "parameter_components_b": { "main_model": 671 } }, { "id": "openai/gpt-5.2-codex", "slug": "openai--gpt-5.2-codex", "name": "GPT-5.2-Codex", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5.2", "description": "GPT-5.2 coding variant with reasoning, text and image input, function calling and context compaction.", "release_date": "2025-12-18", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 400000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-gpt-5-2-codex/", "source_type": "primary", "sources": [ { "label": "Introducing GPT-5.2-Codex", "url": "https://openai.com/index/introducing-gpt-5-2-codex/", "type": "primary" }, { "label": "GPT-5.2-Codex model reference", "url": "https://developers.openai.com/api/docs/models/gpt-5.2-codex.md", "type": "primary" }, { "label": "OpenAI API changelog", "url": "https://developers.openai.com/api/docs/changelog.md", "type": "primary" }, { "label": "Addendum to GPT-5.2 System Card: GPT-5.2-Codex", "url": "https://cdn.openai.com/pdf/ac7c37ae-7f4c-4442-b741-2eabdeaf77e0/oai_5_2_Codex.pdf", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201.md", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "OpenAI API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "Introducing GPT-5.3-Codex", "url": "https://openai.com/index/introducing-gpt-5-3-codex/", "type": "primary" }, { "label": "GPT-5.3-Codex System Card", "url": "https://deploymentsafety.openai.com/gpt-5-3-codex/gpt-5-3-codex.pdf", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "OpenAI model catalog", "url": "https://developers.openai.com/api/docs/models/all.md", "type": "primary" }, { "label": "Introducing GPT-5.2", "url": "https://openai.com/index/introducing-gpt-5-2/", "type": "primary" }, { "label": "GPT-5.2 System Card", "url": "https://cdn.openai.com/pdf/3a4153c8-c748-4b71-8e31-aecbde944f8d/oai_5_2_system-card.pdf", "type": "primary" }, { "label": "Step 3.5 Flash: Open Frontier-Level Intelligence with 11B Active Parameters", "url": "https://arxiv.org/html/2602.10604v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5.2-codex.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released in Codex on 2025-12-18; Responses API release was 2026-01-14. The undated gpt-5.2-codex identifier is not an immutable dated checkpoint; the reviewed reference gives no separately dated target or complete alias history. Last-documented limits are 400,000 context, 272,000 maximum input and 128,000 maximum output tokens. Compaction does not enlarge that context window. PDF is an API input format processed into text and page images. Transformer is a qualified inference from exact-model prompt-caching support and the applicable first-party caching mechanism; dense/MoE structure and full block inventory remain undisclosed. Total and active parameter counts remain undisclosed: OpenAI explicitly identifies this as a GPT-5.2 coding variant, but neither the variant nor its named base has a supported nominal count in the reviewed sources. The API was deprecated on 2026-04-22 and shut down on 2026-07-23; the current model reference says Shut down. No live endpoint request was made, and this API retirement date is not asserted as a separate Codex product retirement date.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T16:50:27Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/425.json" }, { "id": "google/gemini-3-flash-preview", "slug": "google--gemini-3-flash-preview", "name": "Gemini 3 Flash Preview", "developer": "Google", "developer_id": "google", "family": "gemini-flash", "description": "Gemini 3 Flash preview is a multimodal reasoning model with adjustable thinking levels and function calling, accepting text, images, audio, video and PDFs and generating text.", "release_date": "2025-12-17", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ai.google.dev/gemini-api/docs/models/gemini-3-flash-preview", "source_type": "primary", "sources": [ { "label": "Gemini 3 Flash preview - Gemini API", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3-flash-preview", "type": "primary" }, { "label": "Gemini API release notes", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "primary" }, { "label": "Gemini 3 Flash Model Card - December 2025", "url": "https://deepmind.google/models/model-cards/gemini-3-flash/", "type": "primary" }, { "label": "Gemini 3 Pro Model Card - updated May 2026", "url": "https://deepmind.google/models/model-cards/gemini-3-pro/", "type": "primary" }, { "label": "Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemini-3-flash-preview.toml", "type": "catalog" }, { "label": "Gemini 3 Flash: frontier intelligence built for speed", "url": "https://blog.google/products-and-platforms/products/gemini/gemini-3-flash/", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Sparse MoE Transformer through the Flash card’s named Gemini 3 Pro dependency. Neither model has a published total or active parameter count in the reviewed sources. Context is the 1,048,576-token input limit; maximum output is 65,536 tokens. Specifications refer to gemini-3-flash-preview.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T12:55:58.340614Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/200.json" }, { "id": "xiaomi/mimo-v2-flash", "slug": "xiaomi--mimo-v2-flash", "name": "MiMo-V2-Flash", "developer": "Xiaomi", "developer_id": "xiaomi", "family": "mimo", "description": "Open-weight text model with hybrid-attention MoE architecture, switchable reasoning and tool calling.", "release_date": "2025-12-16", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 309, "active_parameters_b": 15, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://mimo.xiaomi.com/mimo-v2-flash", "source_type": "primary", "sources": [ { "label": "Original release announcement", "url": "https://mimo.xiaomi.com/mimo-v2-flash", "type": "primary" }, { "label": "Technical report and model configuration", "url": "https://arxiv.org/html/2601.02780v2", "type": "primary" }, { "label": "Released model card and weights license", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2-Flash/blob/1afd314a2406c282e0956375c34a676501c78649/README.md", "type": "weights" }, { "label": "Checkpoint configuration", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2-Flash/blob/1afd314a2406c282e0956375c34a676501c78649/config.json", "type": "weights" }, { "label": "Official model update history", "url": "https://mimo.mi.com/docs/en-US/updates/model", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/xiaomi/mimo-v2-flash.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/XiaomiMiMo/MiMo-V2-Flash", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "309B total and 15B active per token are nominal author-reported figures, not a tensor inventory. The model uses a Transformer MoE with 256 routed experts, 8 selected per token, no shared experts, and hybrid sliding-window/global attention; its MTP draft blocks are dense. The released checkpoint is text-only and supports 262144 tokens after context extension. Weights are marked MIT; GitHub source code carries Apache-2.0. December 16, 2025 is the initial release; January/February 2026 API updates kept the same model name.", "field_status": {}, "audited_at": "2026-10-27T05:42:18.132839Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/537.json" }, { "id": "nvidia/nemotron-3-nano-30b-a3b", "slug": "nvidia--nemotron-3-nano-30b-a3b", "name": "Nemotron 3 Nano 30B A3B", "developer": "NVIDIA", "developer_id": "nvidia", "family": "nemotron", "description": "NVIDIA text model with a Mamba-2, attention and sparse MoE architecture, trained for reasoning, chat and tool use with configurable thinking and a 1M-token context.", "release_date": "2025-12-15", "date_precision": "day", "open_weights": true, "license": "NVIDIA Nemotron Open Model License", "parameters_b": 31.577937344, "active_parameters_b": 3.6, "parameters_source": "primary_source", "architecture": "hybrid-ssm-transformer-moe", "context_window": 1048576, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16/blob/bf77c3174f68ad409e1c2aa60daeb46e32d1c606/README.md", "source_type": "primary", "sources": [ { "label": "Official Nemotron 3 Nano BF16 model card", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16/blob/bf77c3174f68ad409e1c2aa60daeb46e32d1c606/README.md", "type": "primary" }, { "label": "Pinned checkpoint configuration", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16/blob/bf77c3174f68ad409e1c2aa60daeb46e32d1c606/config.json", "type": "weights" }, { "label": "Pinned model implementation", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16/blob/bf77c3174f68ad409e1c2aa60daeb46e32d1c606/modeling_nemotron_h.py", "type": "weights" }, { "label": "Checkpoint index", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16/blob/bf77c3174f68ad409e1c2aa60daeb46e32d1c606/model.safetensors.index.json", "type": "weights" }, { "label": "Official BF16 shard headers, first of 13", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16/resolve/bf77c3174f68ad409e1c2aa60daeb46e32d1c606/model-00001-of-00013.safetensors", "type": "weights" }, { "label": "Nemotron 3 Nano technical report", "url": "https://arxiv.org/html/2512.20848v1", "type": "primary" }, { "label": "December 15 Nemotron 3 launch", "url": "https://nvidianews.nvidia.com/news/nvidia-debuts-nemotron-3-family-of-open-models", "type": "primary" }, { "label": "NVIDIA Nemotron Open Model License", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-nemotron-open-model-license/", "type": "primary" }, { "label": "Pinned reasoning and tool-call template", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16/blob/bf77c3174f68ad409e1c2aa60daeb46e32d1c606/chat_template.jinja", "type": "weights" }, { "label": "vLLM CLI numeric suffix definition", "url": "https://docs.vllm.ai/en/v0.32.0/cli/", "type": "primary" }, { "label": "NVIDIA model-family technical overview", "url": "https://www.nvidia.com/en-us/ai-data-science/foundation-models/nemotron/llm-info/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/nvidia/nemotron-3-nano-30b-a3b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Exact total is 31.577937344B learned parameters from the original post-trained BF16 checkpoint, excluding 2944 router-buffer elements. The report rounds this to 31.6B; 30B is the nominal model tier. Active count is the report’s nominal 3.6B including embeddings (3.2B excluding embeddings); the model card instead says 3.5B. Architecture is 23 Mamba-2, 23 MoE and six attention layers. The shared MLP has twice the routed-expert width, reconciling one shared module in code with two expert-width units in the report. Supported context is 1048576 tokens; 262144 is the lower-memory HF default.", "field_status": {}, "audited_at": "2026-10-27T00:38:31.157637+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/381.json" }, { "id": "openai/gpt-audio-mini-2025-12-15", "slug": "openai--gpt-audio-mini-2025-12-15", "name": "GPT-audio-mini-2025-12-15", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-audio", "description": null, "release_date": "2025-12-15", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 128000, "modalities": { "input": [ "text", "audio" ], "output": [ "text", "audio" ] }, "reasoning": null, "tool_call": null, "source_url": "https://developers.openai.com/api/docs/changelog", "source_type": "primary", "sources": [ { "label": "Dated publisher API releases", "url": "https://developers.openai.com/api/docs/changelog", "type": "documentation" }, { "label": "Publisher model documentation and snapshot mapping", "url": "https://developers.openai.com/api/docs/models/gpt-audio-mini", "type": "documentation" } ], "aliases": [ "openai/gpt-audio-mini" ], "first_seen": "2026-10-29T10:31:17.277Z", "last_seen": "2026-10-29T10:31:17.277Z", "upstream_present": false, "availability": "public", "notes": "Original publisher conversational audio model. " }, { "id": "openai/gpt-realtime-mini-2025-12-15", "slug": "openai--gpt-realtime-mini-2025-12-15", "name": "GPT-realtime-mini-2025-12-15", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-realtime", "description": null, "release_date": "2025-12-15", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 32000, "modalities": { "input": [ "text", "audio", "image" ], "output": [ "text", "audio" ] }, "reasoning": null, "tool_call": null, "source_url": "https://developers.openai.com/api/docs/changelog", "source_type": "primary", "sources": [ { "label": "Dated publisher API releases", "url": "https://developers.openai.com/api/docs/changelog", "type": "documentation" }, { "label": "Publisher model documentation and snapshot mapping", "url": "https://developers.openai.com/api/docs/models/gpt-realtime-mini", "type": "documentation" } ], "aliases": [ "openai/gpt-realtime-mini" ], "first_seen": "2026-10-29T10:31:17.277Z", "last_seen": "2026-10-29T10:31:17.277Z", "upstream_present": false, "availability": "public", "notes": "Original publisher conversational audio model. " }, { "id": "allenai/olmo-3.1-32b-instruct", "slug": "allenai--olmo-3.1-32b-instruct", "name": "Olmo-3.1-32B-Instruct", "developer": "Ai2", "developer_id": "allenai", "family": "olmo", "description": null, "release_date": "2025-12-12", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 32, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 65536, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://allenai.org/blog/olmo3", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://allenai.org/blog/olmo3", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/allenai/Olmo-3.1-32B-Instruct/blob/ac0587e4a7744a551c059d8cd17ba220bc940dae/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/allenai/Olmo-3.1-32B-Instruct/blob/ac0587e4a7744a551c059d8cd17ba220bc940dae/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "The developer’s December 12 update explicitly releases the Olmo 3.1 checkpoints. Distinct publisher post-training on the existing Olmo lineage; nominal parameter count." }, { "id": "allenai/olmo-3.1-32b-think", "slug": "allenai--olmo-3.1-32b-think", "name": "Olmo-3.1-32B-Think", "developer": "Ai2", "developer_id": "allenai", "family": "olmo", "description": null, "release_date": "2025-12-12", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 32, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 65536, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://allenai.org/blog/olmo3", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://allenai.org/blog/olmo3", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/allenai/Olmo-3.1-32B-Think/blob/97ade57a9cc55ed48bcd76be4b7a3b6fee61b80d/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/allenai/Olmo-3.1-32B-Think/blob/97ade57a9cc55ed48bcd76be4b7a3b6fee61b80d/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "The developer’s December 12 update explicitly releases the Olmo 3.1 checkpoints. Distinct publisher post-training on the existing Olmo lineage; nominal parameter count." }, { "id": "allenai/olmo-3.1-7b-rl-zero-code", "slug": "allenai--olmo-3.1-7b-rl-zero-code", "name": "Olmo-3.1-7B-RL-Zero-Code", "developer": "Ai2", "developer_id": "allenai", "family": "olmo", "description": null, "release_date": "2025-12-12", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 65536, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://allenai.org/blog/olmo3", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://allenai.org/blog/olmo3", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/allenai/Olmo-3.1-7B-RL-Zero-Code/blob/1ca3bb30f0e51b7ef05f7d0fefcf0e6b14eefc90/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/allenai/Olmo-3.1-7B-RL-Zero-Code/blob/1ca3bb30f0e51b7ef05f7d0fefcf0e6b14eefc90/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "The developer’s December 12 update explicitly releases the Olmo 3.1 checkpoints. Distinct publisher post-training on the existing Olmo lineage; nominal parameter count." }, { "id": "allenai/olmo-3.1-7b-rl-zero-math", "slug": "allenai--olmo-3.1-7b-rl-zero-math", "name": "Olmo-3.1-7B-RL-Zero-Math", "developer": "Ai2", "developer_id": "allenai", "family": "olmo", "description": null, "release_date": "2025-12-12", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 65536, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://allenai.org/blog/olmo3", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://allenai.org/blog/olmo3", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/allenai/Olmo-3.1-7B-RL-Zero-Math/blob/364820b002fc8e81d11a81c7fc9dd6b49c83cd05/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/allenai/Olmo-3.1-7B-RL-Zero-Math/blob/364820b002fc8e81d11a81c7fc9dd6b49c83cd05/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "The developer’s December 12 update explicitly releases the Olmo 3.1 checkpoints. Distinct publisher post-training on the existing Olmo lineage; nominal parameter count." }, { "id": "google/gemini-2.5-flash-native-audio-preview-12-2025", "slug": "google--gemini-2.5-flash-native-audio-preview-12-2025", "name": "Gemini 2.5 flash native audio preview 12 2025", "developer": "Google", "developer_id": "google", "family": "gemini", "description": null, "release_date": "2025-12-12", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 131072, "modalities": { "input": [ "text", "image", "audio", "video" ], "output": [ "text", "audio" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ai.google.dev/gemini-api/docs/changelog", "source_type": "primary", "sources": [ { "label": "Dated publisher API releases", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "documentation" }, { "label": "Publisher model capability table", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-flash-native-audio-preview-12-2025", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:31:17.277Z", "last_seen": "2026-10-29T10:31:17.277Z", "upstream_present": false, "availability": "public", "notes": "Conversational audio model. Text can be returned as the generated speech transcript; text-only response mode is not implied." }, { "id": "openai/gpt-5.2", "slug": "openai--gpt-5.2", "name": "GPT-5.2", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5.2", "description": "GPT-5.2 reasoning model with text and image input, function calling and a 400,000-token context window.", "release_date": "2025-12-11", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 400000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-gpt-5-2/", "source_type": "primary", "sources": [ { "label": "Introducing GPT-5.2", "url": "https://openai.com/index/introducing-gpt-5-2/", "type": "primary" }, { "label": "GPT-5.2 API model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.2.md", "type": "primary" }, { "label": "Update to GPT-5 System Card: GPT-5.2", "url": "https://cdn.openai.com/pdf/3a4153c8-c748-4b71-8e31-aecbde944f8d/oai_5_2_system-card.pdf", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201.md", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "Introducing GPT-5.4", "url": "https://openai.com/index/introducing-gpt-5-4/", "type": "primary" }, { "label": "GPT-5.4 Thinking System Card", "url": "https://deploymentsafety.openai.com/gpt-5-4-thinking/gpt-5-4-thinking.pdf", "type": "primary" }, { "label": "Reasoning Models Struggle to Control their Chains of Thought", "url": "https://cdn.openai.com/pdf/a21c39c1-fa07-41db-9078-973a12620117/cot_controllability.pdf", "type": "primary" }, { "label": "Update to GPT-5 System Card: GPT-5.2 — updated online card", "url": "https://deploymentsafety.openai.com/gpt-5-2", "type": "primary" }, { "label": "API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "All OpenAI API models", "url": "https://developers.openai.com/api/docs/models/all.md", "type": "primary" }, { "label": "Investigating the consequences of accidentally grading CoT during RL", "url": "https://alignment.openai.com/accidental-cot-grading/", "type": "primary" }, { "label": "Step 3.5 Flash: Open Frontier-Level Intelligence with 11B Active Parameters — Table 5", "url": "https://arxiv.org/html/2602.10604v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5.2.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The gpt-5.2 API alias maps to GPT-5.2 Thinking and documents default snapshot gpt-5.2-2025-12-11, distinct from Instant/chat-latest, Pro and Codex. Context is 400,000 tokens with a separate 128,000 maximum output; no maximum input is stated on the exact model page. Reasoning effort supports none (default), low, medium, high and xhigh. PDF denotes API extraction into text and page images. Transformer is a qualified operational inference from exact prompt-caching support and the applicable Transformer attention KV mechanism; complete block design and dense/MoE routing remain undisclosed. No published nominal total or active parameter count was found for GPT-5.2 Thinking; StepFun's comparison table leaves both GPT-5.2 xHigh counts blank. As of October 27, 2026, the exact API model page has no deprecated/shut-down designation and the reviewed deprecation list has no shutdown notice for this alias or dated snapshot; Chat and Codex retirement notices are separate. No endpoint test was made.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T16:45:32.542880Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/423.json" }, { "id": "openai/gpt-5.2-chat-latest", "slug": "openai--gpt-5.2-chat-latest", "name": "GPT-5.2 Chat", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5.2", "description": "GPT-5.2 Instant chat model with text and image input, text output and function calling in the API.", "release_date": "2025-12-11", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 128000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-gpt-5-2/", "source_type": "primary", "sources": [ { "label": "Introducing GPT-5.2", "url": "https://openai.com/index/introducing-gpt-5-2/", "type": "primary" }, { "label": "GPT-5.2 Chat API documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.2-chat-latest.md", "type": "primary" }, { "label": "Update to GPT-5 System Card: GPT-5.2", "url": "https://cdn.openai.com/pdf/3a4153c8-c748-4b71-8e31-aecbde944f8d/oai_5_2_system-card.pdf", "type": "primary" }, { "label": "GPT-5.5 Instant System Card", "url": "https://deploymentsafety.openai.com/gpt-5-5-instant/gpt-5-5-instant.pdf", "type": "primary" }, { "label": "Investigating the consequences of accidentally grading CoT during RL", "url": "https://alignment.openai.com/accidental-cot-grading/", "type": "primary" }, { "label": "Prompt caching", "url": "https://developers.openai.com/api/docs/guides/prompt-caching.md", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201.md", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "API changelog", "url": "https://developers.openai.com/api/docs/changelog.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "All OpenAI API models", "url": "https://developers.openai.com/api/docs/models/all.md", "type": "primary" }, { "label": "Azure models sold directly by Azure", "url": "https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure", "type": "primary" }, { "label": "GPT-5.3 Instant System Card", "url": "https://deploymentsafety.openai.com/gpt-5-3-instant/gpt-5-3-instant.pdf", "type": "primary" }, { "label": "Step 3.5 Flash: Open Frontier-Level Intelligence with 11B Active Parameters", "url": "https://arxiv.org/html/2602.10604v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5.2-chat-latest.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The original 2025-12-11 launch maps gpt-5.2-chat-latest to GPT-5.2 Instant, separately from Thinking and Pro. This was a mutable alias, explicitly updated on 2026-02-10; the current OpenAI model page lists no dated immutable snapshot. Exact OpenAI documentation specifies 128,000 context and 16,384 maximum output, but its Markdown also gives an inconsistent 272,000 maximum input; Azure exact-version rows corroborate 128,000 context. OpenAI training research explicitly establishes limited CoT for GPT-5.2 Instant; reasoning=true does not imply documented effort controls, adaptive reasoning or the ChatGPT router. PDF denotes API conversion to extracted text and page images. Total and active parameter counts are undisclosed for this Instant alias; no disclosed-size base lineage or exact active-count comparison supports a nominal inference. Architecture remains unverified. OpenAI’s deprecation notice gives 2026-08-10 as this alias’s shutdown date, consistent with its current model page. Specifications are historical documentation, not a successful endpoint test.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "unverified" }, "audited_at": "2026-10-27T16:49:57.172930Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/424.json" }, { "id": "openai/gpt-5.2-pro", "slug": "openai--gpt-5.2-pro", "name": "GPT-5.2 Pro", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5.2", "description": "GPT-5.2 Pro reasoning model with text and image input, function calling, and a 400,000-token context window.", "release_date": "2025-12-11", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 400000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-gpt-5-2/", "source_type": "primary", "sources": [ { "label": "OpenAI GPT-5.2 release", "url": "https://openai.com/index/introducing-gpt-5-2/", "type": "primary" }, { "label": "OpenAI GPT-5.2 Pro model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.2-pro", "type": "primary" }, { "label": "OpenAI file inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "OpenAI ChatGPT retirement notice", "url": "https://help.openai.com/en/articles/6825453-chatgpt-release-notes", "type": "primary" }, { "label": "OpenAI API deprecations", "url": "https://developers.openai.com/api/docs/deprecations", "type": "primary" }, { "label": "OpenAI GPT-5.2 system card", "url": "https://cdn.openai.com/pdf/3a4153c8-c748-4b71-8e31-aecbde944f8d/oai_5_2_system-card.pdf", "type": "primary" }, { "label": "OpenAI GeneBench report", "url": "https://cdn.openai.com/pdf/6dc7175d-d9e7-4b8d-96b8-48fe5798cd5b/oai_genebench_benchmark.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5.2-pro.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "API snapshot: gpt-5.2-pro-2025-12-11. Documented context window: 400,000 tokens; maximum output: 128,000 tokens. Supports reasoning effort medium, high and xhigh through Responses. PDF input is processed into text and page images. Total and active parameter counts remain undisclosed; no supported nominal count or count-bearing base lineage was found. Pro architecture remains unverified. ChatGPT retired GPT-5.2 Pro on June 12, 2026. No exact Pro API shutdown notice was found in the current deprecation list; endpoint operation was not tested.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "unverified" }, "audited_at": "2026-10-27T16:49:55Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/426.json" }, { "id": "mistral/devstral-2512", "slug": "mistral--devstral-2512", "name": "Devstral 2", "developer": "Mistral", "developer_id": "mistral", "family": "devstral", "description": "Mistral's coding-agent model for repository work, terminal tasks, and software fixes", "release_date": "2025-12-09", "date_precision": "day", "open_weights": true, "license": "Modified MIT", "parameters_b": 123, "active_parameters_b": 123, "parameters_source": "primary_source", "architecture": "dense", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://mistral.ai/news/devstral-2-vibe-cli/", "source_type": "primary", "sources": [ { "label": "Introducing: Devstral 2 and Mistral Vibe CLI", "url": "https://mistral.ai/news/devstral-2-vibe-cli/", "type": "primary" }, { "label": "Devstral 2 — Mistral model documentation", "url": "https://docs.mistral.ai/models/devstral-2-25-12", "type": "primary" }, { "label": "Devstral 2 123B Instruct 2512 model card — pinned revision", "url": "https://huggingface.co/mistralai/Devstral-2-123B-Instruct-2512/blob/1613bf01adb5e1c6fdc196b46e6b173eae75eb4a/README.md", "type": "weights" }, { "label": "Devstral 2 123B Instruct 2512 configuration — pinned revision", "url": "https://huggingface.co/mistralai/Devstral-2-123B-Instruct-2512/blob/1613bf01adb5e1c6fdc196b46e6b173eae75eb4a/config.json", "type": "weights" }, { "label": "Devstral 2 Modified MIT License — pinned revision", "url": "https://huggingface.co/mistralai/Devstral-2-123B-Instruct-2512/blob/1613bf01adb5e1c6fdc196b46e6b173eae75eb4a/LICENSE", "type": "weights" }, { "label": "Replace stray reasoning tokens with reserved slots — merged official checkpoint change", "url": "https://huggingface.co/mistralai/Devstral-2-123B-Instruct-2512/discussions/24", "type": "primary" }, { "label": "Official Devstral 2 checkpoint repository metadata", "url": "https://huggingface.co/api/models/mistralai/Devstral-2-123B-Instruct-2512", "type": "weights" }, { "label": "Devstral 2 chat template — pinned revision", "url": "https://huggingface.co/mistralai/Devstral-2-123B-Instruct-2512/blob/1613bf01adb5e1c6fdc196b46e6b173eae75eb4a/chat_template.jinja", "type": "weights" }, { "label": "Devstral 2 FP8 tensor index — pinned revision", "url": "https://huggingface.co/mistralai/Devstral-2-123B-Instruct-2512/blob/1613bf01adb5e1c6fdc196b46e6b173eae75eb4a/model.safetensors.index.json", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/devstral-2512.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Devstral-2-123B-Instruct-2512", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Mistral explicitly reports nominal total and active counts of 123B for Devstral 2. The quantized checkpoint storage aggregate is not an exact learned-parameter count. Modified MIT excludes companies with preceding-month global consolidated revenue above US$20 million unless separately licensed. The exact checkpoint is text-only and explicitly non-reasoning, with native tool calls. Mistral documents API deprecation on May 22, 2026; public weights remain available.", "field_status": {}, "audited_at": "2026-10-26T22:47:08Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/314.json" }, { "id": "mistral/devstral-small-2", "slug": "mistral--devstral-small-2", "name": "Devstral Small 2", "developer": "Mistral", "developer_id": "mistral", "family": "devstral", "description": "Compact multimodal coding model for repository exploration, file editing, and software agents", "release_date": "2025-12-09", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 24, "active_parameters_b": 24, "parameters_source": "primary_source", "architecture": "dense", "context_window": 262144, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://mistral.ai/news/devstral-2-vibe-cli/", "source_type": "primary", "sources": [ { "label": "Mistral release announcement", "url": "https://mistral.ai/news/devstral-2-vibe-cli/", "type": "primary" }, { "label": "Official model reference and Weights table", "url": "https://docs.mistral.ai/models/devstral-small-2-25-12", "type": "primary" }, { "label": "Official checkpoint card", "url": "https://huggingface.co/mistralai/Devstral-Small-2-24B-Instruct-2512/blob/55c5b41e98c2dbd21b0c8afffc540dcfc9eb5128/README.md", "type": "weights" }, { "label": "Official checkpoint configuration", "url": "https://huggingface.co/mistralai/Devstral-Small-2-24B-Instruct-2512/blob/55c5b41e98c2dbd21b0c8afffc540dcfc9eb5128/config.json", "type": "weights" }, { "label": "Mistral dense architecture documentation", "url": "https://docs.mistral.ai/vibe/code/cli/offline-models", "type": "primary" }, { "label": "Mistral maintainer reasoning correction", "url": "https://huggingface.co/mistralai/Devstral-Small-2-24B-Instruct-2512/commit/b396b4d51e58a72dd77e3402994af04021b01e5b", "type": "primary" }, { "label": "Mistral maintainer context clarification", "url": "https://huggingface.co/mistralai/Devstral-Small-2-24B-Instruct-2512/discussions/20", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/devstral-small-2.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Devstral-Small-2-24B-Instruct-2512", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Mistral explicitly reports nominal 24B total and 24B active parameters for Devstral Small 2. The FP8 checkpoint storage aggregate is not an exact learned-parameter count. Native text/image input and text output; explicitly non-reasoning with native tool calls. Use the documented 262144-token context: a maintainer explicitly recommends that serving limit despite the 393216 YaRN ceiling in configuration. Official model docs mark API deprecation on February 27, 2026; public weights remain downloadable.", "field_status": {}, "audited_at": "2026-10-26T22:50:42.817601Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/316.json" }, { "id": "zhipuai/autoglm-phone-9b-multilingual", "slug": "zhipuai--autoglm-phone-9b-multilingual", "name": "AutoGLM Phone 9B Multilingual", "developer": "Z.ai", "developer_id": "zhipuai", "family": "glm", "description": "Multilingual dense vision-language model that reasons over phone screens and generates text actions for a mobile-control framework.", "release_date": "2025-12-09", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 10.292777472, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 65536, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://github.com/zai-org/Open-AutoGLM/commit/d4fbb4649c8ca49f8467d95b2fb33f53616c60c8", "source_type": "primary", "sources": [ { "label": "Exact multilingual model card", "url": "https://huggingface.co/zai-org/AutoGLM-Phone-9B-Multilingual/blob/832ab5e014b965268dafd944e64cbc8e4bade892/README.md", "type": "weights" }, { "label": "Exact multilingual checkpoint configuration", "url": "https://huggingface.co/zai-org/AutoGLM-Phone-9B-Multilingual/blob/832ab5e014b965268dafd944e64cbc8e4bade892/config.json", "type": "weights" }, { "label": "Complete multilingual tensor index", "url": "https://huggingface.co/zai-org/AutoGLM-Phone-9B-Multilingual/blob/832ab5e014b965268dafd944e64cbc8e4bade892/model.safetensors.index.json", "type": "weights" }, { "label": "Checkpoint shard1 bounded tensor header", "url": "https://huggingface.co/zai-org/AutoGLM-Phone-9B-Multilingual/resolve/832ab5e014b965268dafd944e64cbc8e4bade892/model-00001-of-00005.safetensors", "type": "weights" }, { "label": "Checkpoint shard2 bounded tensor header", "url": "https://huggingface.co/zai-org/AutoGLM-Phone-9B-Multilingual/resolve/832ab5e014b965268dafd944e64cbc8e4bade892/model-00002-of-00005.safetensors", "type": "weights" }, { "label": "Checkpoint shard3 bounded tensor header", "url": "https://huggingface.co/zai-org/AutoGLM-Phone-9B-Multilingual/resolve/832ab5e014b965268dafd944e64cbc8e4bade892/model-00003-of-00005.safetensors", "type": "weights" }, { "label": "Checkpoint shard4 bounded tensor header", "url": "https://huggingface.co/zai-org/AutoGLM-Phone-9B-Multilingual/resolve/832ab5e014b965268dafd944e64cbc8e4bade892/model-00004-of-00005.safetensors", "type": "weights" }, { "label": "Checkpoint shard5 bounded tensor header", "url": "https://huggingface.co/zai-org/AutoGLM-Phone-9B-Multilingual/resolve/832ab5e014b965268dafd944e64cbc8e4bade892/model-00005-of-00005.safetensors", "type": "weights" }, { "label": "GLM4V reference implementation matching checkpoint configuration", "url": "https://raw.githubusercontent.com/huggingface/transformers/v5.0.0rc0/src/transformers/models/glm4v/modeling_glm4v.py", "type": "primary" }, { "label": "Exact multilingual chat template", "url": "https://huggingface.co/zai-org/AutoGLM-Phone-9B-Multilingual/blob/832ab5e014b965268dafd944e64cbc8e4bade892/chat_template.jinja", "type": "weights" }, { "label": "Exact multilingual video processor configuration", "url": "https://huggingface.co/zai-org/AutoGLM-Phone-9B-Multilingual/blob/832ab5e014b965268dafd944e64cbc8e4bade892/video_preprocessor_config.json", "type": "weights" }, { "label": "Official multilingual agent prompt", "url": "https://github.com/zai-org/Open-AutoGLM/blob/86f55382982fb054e8fc98ca80609dff8a2cdc3c/phone_agent/config/prompts_en.py", "type": "primary" }, { "label": "Official agent model client", "url": "https://github.com/zai-org/Open-AutoGLM/blob/86f55382982fb054e8fc98ca80609dff8a2cdc3c/phone_agent/model/client.py", "type": "primary" }, { "label": "Original developer multilingual download announcement", "url": "https://github.com/zai-org/Open-AutoGLM/commit/d4fbb4649c8ca49f8467d95b2fb33f53616c60c8", "type": "primary" }, { "label": "December8 project README", "url": "https://github.com/zai-org/Open-AutoGLM/blob/cff3bf97dbb4f5b840230617e1759142b190c295/README_en.md", "type": "primary" }, { "label": "AutoGLM Goes Open Source", "url": "https://autoglm.z.ai/blog/", "type": "primary" }, { "label": "Exact checkpoint metadata", "url": "https://huggingface.co/api/models/zai-org/AutoGLM-Phone-9B-Multilingual", "type": "weights" }, { "label": "Official multilingual product documentation", "url": "https://docs.z.ai/guides/vlm/autoglm-phone-multilingual", "type": "primary" }, { "label": "Exact tokenizer configuration", "url": "https://huggingface.co/zai-org/AutoGLM-Phone-9B-Multilingual/blob/832ab5e014b965268dafd944e64cbc8e4bade892/tokenizer_config.json", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/zhipuai/autoglm-phone-9b-multilingual.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/AutoGLM-Phone-9B-Multilingual", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Exact checkpoint count is 10,292,777,472 learned parameters: 9,400,279,040 language parameters including the untied LM head, plus 892,498,432 vision/projector parameters. All 704 tensors across five bounded BF16 headers were reconciled; the published 934400 metadata total is erroneous. Dense active parameters are not separately defined. Multilingual download instructions were published December 9, 2025, after the December 8 Chinese-only release. The model position limit of 65536 differs from the tokenizer limit of 64000 and example serving cap of 25480. Native checkpoint video input is supported, while documented phone use focuses on screenshots. Reasoning and fixed GUI actions are documented; standard function-calling support remains unverified. MIT is declared for weights, alongside a research/educational-use warning; framework code uses Apache 2.0.", "field_status": { "active_parameters_b": "not_applicable", "tool_call": "unverified" }, "audited_at": "2026-10-27T06:00:27.305858Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/546.json" }, { "id": "relace/relace-search", "slug": "relace--relace-search", "name": "Relace Search", "developer": "Relace", "developer_id": "relace", "family": null, "description": "Code-search language model trained with reinforcement learning to select parallel file and grep tools, reason across search turns, and report relevant file ranges to a coding agent.", "release_date": "2025-12-08", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 256000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://relace.ai/old/blog/fast-agentic-search", "source_type": "primary", "sources": [ { "label": "Exploiting parallel tool calls to make agentic search 4x faster", "url": "https://relace.ai/old/blog/fast-agentic-search", "type": "primary" }, { "label": "Search Code", "url": "https://docs.relace.ai/api-reference/fast-agentic-search/search.md", "type": "primary" }, { "label": "Fast Agentic Search: Overview", "url": "https://docs.relace.ai/docs/fast-agentic-search/agent.md", "type": "primary" }, { "label": "Terms of Use", "url": "https://relace.ai/legal/terms-of-use", "type": "primary" }, { "label": "Relace Search: provider endpoints", "url": "https://openrouter.ai/api/v1/models/relace/relace-search/endpoints", "type": "provider" }, { "label": "Relace Blog", "url": "https://relace.ai/blog", "type": "primary" }, { "label": "Relace complete documentation", "url": "https://docs.relace.ai/llms-full.txt", "type": "primary" }, { "label": "Relace Compact: 50k TPS for 50% cost savings", "url": "https://relace.ai/old/blog/compact", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/relace/relace-search.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Relace Search is the separately callable RL-trained model behind Fast Agentic Search, distinct from embedding retrieval and the execution harness. Total and active parameter counts remain undisclosed; no named base model supports an inherited size. It emits reasoning and model-selected tool calls, without a separately documented thinking-token interface. OpenRouter advertises a 256,000-token Relace endpoint context and 128,000-token completion limit; native context and neural architecture remain undisclosed. SDK/harness and OpenAPI licenses are not model-weight licenses.", "field_status": { "family": "undisclosed", "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T17:45:09.357348Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/481.json" }, { "id": "zhipuai/glm-4.6v", "slug": "zhipuai--glm-4.6v", "name": "GLM-4.6V", "developer": "Z.ai", "developer_id": "zhipuai", "family": "glm", "description": "Vision-language MoE model with native function calling and optional reasoning", "release_date": "2025-12-08", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 107.71092736, "active_parameters_b": 12, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/zai-org/GLM-V/blob/f093b70998bec336acbc00d1a545e05893e3c9fa/README.md", "source_type": "primary", "sources": [ { "label": "GLM-4.6V official model card at pinned revision", "url": "https://huggingface.co/zai-org/GLM-4.6V/blob/4e2d47eb0b41c5280d8294b17cef9e94fdcfff46/README.md", "type": "weights" }, { "label": "GLM-4.6V checkpoint configuration", "url": "https://huggingface.co/zai-org/GLM-4.6V/blob/4e2d47eb0b41c5280d8294b17cef9e94fdcfff46/config.json", "type": "weights" }, { "label": "GLM-4.6V complete checkpoint index and all 41 indexed tensor headers", "url": "https://huggingface.co/zai-org/GLM-4.6V/blob/4e2d47eb0b41c5280d8294b17cef9e94fdcfff46/model.safetensors.index.json", "type": "weights" }, { "label": "GLM-4V-MoE implementation referenced by the developer", "url": "https://github.com/huggingface/transformers/blob/v5.0.0rc0/src/transformers/models/glm4v_moe/modeling_glm4v_moe.py", "type": "primary" }, { "label": "GLM-4.5V and GLM-4.1V-Thinking technical report, revision6 including GLM-4.6V", "url": "https://arxiv.org/html/2507.01006v6", "type": "primary" }, { "label": "Original GLM-4.6V release announcement in official README", "url": "https://github.com/zai-org/GLM-V/blob/f093b70998bec336acbc00d1a545e05893e3c9fa/README.md", "type": "primary" }, { "label": "GLM-4.6V official API documentation", "url": "https://docs.z.ai/guides/vlm/glm-4.6v", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/zhipuai/glm-4.6v.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/GLM-4.6V", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The full BF16 checkpoint contains 107.71092736B learned parameters: 106.852245504B language/output and 0.858681856B vision/adapter. The developer labels the model 106B-A12B; 12B is a nominal active count whose exact multimodal phase convention is unspecified. The total includes untied input/output embeddings and excludes router correction buffers; there are no MTP layers. Context is 131072 tokens in the checkpoint, advertised as 128K.", "field_status": {}, "audited_at": "2026-10-27T06:10:36.118481Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/552.json" }, { "id": "zhipuai/glm-4.6v-flash", "slug": "zhipuai--glm-4.6v-flash", "name": "GLM-4.6V-Flash", "developer": "Z.ai", "developer_id": "zhipuai", "family": "glm", "description": "Dense vision-language model with hybrid reasoning and native tool calling for text, images, and video.", "release_date": "2025-12-08", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 10.292777472, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://z.ai/blog/glm-4.6v", "source_type": "primary", "sources": [ { "label": "GLM-4.6V: Open Source Multimodal Models with Native Tool Use", "url": "https://z.ai/blog/glm-4.6v", "type": "primary" }, { "label": "GLM-4.6V-Flash model card, pinned revision", "url": "https://huggingface.co/zai-org/GLM-4.6V-Flash/blob/411bb4d77144a3f03accbf4b780f5acb8b7cde4e/README.md", "type": "weights" }, { "label": "GLM-4.6V-Flash checkpoint configuration", "url": "https://huggingface.co/zai-org/GLM-4.6V-Flash/blob/411bb4d77144a3f03accbf4b780f5acb8b7cde4e/config.json", "type": "weights" }, { "label": "GLM-4.6V-Flash indexed BF16 checkpoint and bounded shard headers", "url": "https://huggingface.co/zai-org/GLM-4.6V-Flash/blob/411bb4d77144a3f03accbf4b780f5acb8b7cde4e/model.safetensors.index.json", "type": "weights" }, { "label": "Transformers GLM4V implementation at the checkpoint declared version", "url": "https://github.com/huggingface/transformers/blob/v5.0.0rc0/src/transformers/models/glm4v/modeling_glm4v.py", "type": "primary" }, { "label": "GLM-4.6V-Flash chat template", "url": "https://huggingface.co/zai-org/GLM-4.6V-Flash/blob/411bb4d77144a3f03accbf4b780f5acb8b7cde4e/chat_template.jinja", "type": "weights" }, { "label": "GLM-4.6V series developer documentation", "url": "https://docs.z.ai/guides/vlm/glm-4.6v", "type": "primary" }, { "label": "GLM-4.6V-Flash tokenizer configuration", "url": "https://huggingface.co/zai-org/GLM-4.6V-Flash/blob/411bb4d77144a3f03accbf4b780f5acb8b7cde4e/tokenizer_config.json", "type": "weights" }, { "label": "Official GLM-V repository and model download table", "url": "https://github.com/zai-org/GLM-V", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/zhipuai/glm-4.6v-flash.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/GLM-4.6V-Flash", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released December 8, 2025. Advertised as 9B; the complete BF16 checkpoint contains 10.292777472B learned parameters: 9.400279040B language/output and 0.892498432B vision/projector. Dense architecture; active parameters are not separately defined. Checkpoint model context capacity is 131072 tokens; the documented API/tokenizer limit is 128000. MIT applies to weights.", "parameter_components_b": { "language_model_and_output": 9.40027904, "vision_encoder_and_projector": 0.892498432 }, "field_status": { "active_parameters_b": "not_applicable" }, "audited_at": "2026-10-27T06:12:10.159068+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/553.json" }, { "id": "essentialai/rnj-1-instruct", "slug": "essentialai--rnj-1-instruct", "name": "Rnj-1 Instruct", "developer": "Essential AI", "developer_id": "essentialai", "family": "rnj", "description": "Open-weight 8.31B dense instruction-tuned model trained by Essential AI for code, mathematics, science and tool use.", "release_date": "2025-12-05", "date_precision": "day", "open_weights": true, "license": "Apache 2.0", "parameters_b": 8.310501376, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://www.essential.ai/research/rnj-1", "source_type": "primary", "sources": [ { "label": "Official Rnj-1 announcement", "url": "https://www.essential.ai/research/rnj-1", "type": "primary" }, { "label": "Official Instruct model card", "url": "https://huggingface.co/EssentialAI/rnj-1-instruct/blob/5ef96d7a82b2bff8009192979432c9bc64e870cd/README.md", "type": "primary" }, { "label": "Official checkpoint configuration", "url": "https://huggingface.co/EssentialAI/rnj-1-instruct/blob/5ef96d7a82b2bff8009192979432c9bc64e870cd/config.json", "type": "primary" }, { "label": "Official complete weight index", "url": "https://huggingface.co/EssentialAI/rnj-1-instruct/blob/5ef96d7a82b2bff8009192979432c9bc64e870cd/model.safetensors.index.json", "type": "weights" }, { "label": "Official release commit history", "url": "https://huggingface.co/api/models/EssentialAI/rnj-1-instruct/commits/main", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/essentialai/rnj-1-instruct.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/EssentialAI/rnj-1-instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete checkpoint count: 8.310501376B learned parameters, with tied embeddings counted once; the publisher rounds to 8.3B. Active parameters are not separately disclosed. December 5 is the public announcement date; December 4 only created the repository. Context is 32,768 by default; optional 128K YaRN extrapolation requires configuration changes. No dedicated reasoning flag is established by primary evidence.", "field_status": { "active_parameters_b": "undisclosed", "reasoning": "unverified" }, "audited_at": "2026-10-27T12:40:28.716685+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/185.json" }, { "id": "amazon/nova-2-lite-v1", "slug": "amazon--nova-2-lite-v1", "name": "Nova 2 Lite", "developer": "Amazon", "developer_id": "amazon", "family": "nova", "description": "Multimodal reasoning model for visual analysis, planning, and tool use", "release_date": "2025-12-02", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 1000000, "modalities": { "input": [ "text", "image", "video", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/amazon/nova-2-lite-v1.toml", "source_type": "catalog", "sources": [ { "label": "Amazon Nova 2 Lite service card", "url": "https://docs.aws.amazon.com/ai/responsible-ai/nova-2-lite/overview.html", "type": "primary" }, { "label": "Amazon Nova 2 technical report", "url": "https://cdn.amazon.science/c5/3d/84514a224666b5be6de4b43ef4aa/nova-2-0-technical-report2.pdf", "type": "primary" }, { "label": "Nova 2 Lite launch announcement", "url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-2-lite-a-fast-cost-effective-reasoning-model/", "type": "primary" }, { "label": "Amazon Nova 2 multimodal documentation", "url": "https://docs.aws.amazon.com/nova/latest/nova2-userguide/using-multimodal-models.html", "type": "docs" }, { "label": "Nova 2 migration comparison", "url": "https://aws.amazon.com/blogs/machine-learning/migrate-from-amazon-nova-1-to-amazon-nova-2-on-amazon-bedrock/", "type": "primary" }, { "label": "Nova customization model comparison", "url": "https://aws.amazon.com/blogs/machine-learning/customize-amazon-nova-models-with-amazon-bedrock-fine-tuning/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/amazon/nova-2-lite-v1.toml", "type": "catalog" }, { "label": "Alias documentation", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-2-lite.html", "type": "docs" } ], "aliases": [ "amazon/nova-2-lite" ], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Transformer-based architecture is documented; total and active parameter counts and dense versus mixture-of-experts classification are undisclosed. PDF input is available through the Converse API; video audio tracks are not processed.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T10:03:07Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/077.json", "alias_sources": { "amazon/nova-2-lite": { "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-2-lite.html", "note": "Nova 2 Lite display-name record and the dated Bedrock ID describe the same release." } } }, { "id": "amazon/nova-2-sonic", "slug": "amazon--nova-2-sonic", "name": "Nova 2 Sonic", "developer": "Amazon", "developer_id": "amazon", "family": "nova-sonic", "description": "Conversational speech model with text and audio responses and tool use.", "release_date": "2025-12-02", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 1000000, "modalities": { "input": [ "text", "audio" ], "output": [ "text", "audio" ] }, "reasoning": null, "tool_call": true, "source_url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-2-sonic-next-generation-speech-to-speech-model-for-conversational-ai/", "source_type": "primary", "sources": [ { "label": "Dated public launch", "url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-2-sonic-next-generation-speech-to-speech-model-for-conversational-ai/", "type": "primary" }, { "label": "Publisher service card and architecture", "url": "https://docs.aws.amazon.com/ai/responsible-ai/nova-2-sonic/overview.html", "type": "primary" }, { "label": "Bedrock model identity and limits", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-2-sonic.html", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:13:05.399Z", "last_seen": "2026-10-30T07:13:05.399Z", "upstream_present": false, "availability": "public", "notes": "Conversational speech model with a core transformer language model, speech encoder and renderer. The publisher does not disclose total or active parameters or dense/MoE subtype. Available through Bedrock; no downloadable exactcheckpoint was verified. Toolcalling is explicitly documented; a separate extended-thinking mode is unverified. Context1000000 is the current Bedrockcard nominal1M. March/May2026 in-place refreshes remain a separate revision-identity followup." }, { "id": "mistral/ministral-14b", "slug": "mistral--ministral-14b", "name": "Ministral 3 14B Instruct", "developer": "Mistral", "developer_id": "mistral", "family": "ministral", "description": "Multimodal instruction model for local assistants and edge agents, with native function calling.", "release_date": "2025-12-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 14, "active_parameters_b": 14, "parameters_source": "primary_source", "architecture": "dense", "context_window": 262144, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://mistral.ai/news/mistral-3/", "source_type": "primary", "sources": [ { "label": "Models.dev undated Ministral 14B record", "url": "https://raw.githubusercontent.com/anomalyco/models.dev/dev/models/mistral/ministral-14b.toml", "type": "catalog" }, { "label": "Introducing Mistral 3", "url": "https://mistral.ai/news/mistral-3/", "type": "primary" }, { "label": "Ministral 3 14B v25.12 — model reference and Weights table", "url": "https://docs.mistral.ai/models/ministral-3-14b-25-12", "type": "primary" }, { "label": "Ministral 3 14B — official Markdown model reference", "url": "https://docs.mistral.ai/models/ministral-3-14b-25-12.md", "type": "primary" }, { "label": "Ministral 3 14B Instruct 2512 — pinned official checkpoint card", "url": "https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512/blob/29439f81c2be264d8d393273f99e7db9c0961120/README.md", "type": "weights" }, { "label": "Ministral 3 14B Instruct — pinned Transformers configuration", "url": "https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512/blob/29439f81c2be264d8d393273f99e7db9c0961120/config.json", "type": "weights" }, { "label": "Ministral 3 technical report", "url": "https://arxiv.org/html/2601.08584v1", "type": "primary" }, { "label": "Mistral maintainer correction of unused reasoning tokens", "url": "https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512/discussions/25", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/ministral-14b.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "This undated catalog entry explicitly links Ministral-3-14B-Instruct-2512; official API names are ministral-14b-2512 and ministral-14b-latest. Specifications here describe that Instruct checkpoint, not Base or Reasoning. Mistral reports nominal 14B total and 14B active parameters; rounded component figures and FP8 storage totals are not exact whole-model counts. The checkpoint context is 262,144 tokens. A Mistral maintainer explicitly confirms it is non-reasoning.", "field_status": {}, "audited_at": "2026-10-26T22:54:58.852658Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/320.json" }, { "id": "mistral/ministral-14b-2512", "slug": "mistral--ministral-14b-2512", "name": "Ministral 3 14B 2512", "developer": "Mistral", "developer_id": "mistral", "family": "ministral", "description": "Instruction-tuned text and image model with native function calling, designed for local and edge deployment.", "release_date": "2025-12-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 14, "active_parameters_b": 14, "parameters_source": "primary_source", "architecture": "dense", "context_window": 262144, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://mistral.ai/news/mistral-3/", "source_type": "primary", "sources": [ { "label": "Introducing Mistral 3", "url": "https://mistral.ai/news/mistral-3/", "type": "primary" }, { "label": "Ministral 3 14B — exact API model reference and Weights table", "url": "https://docs.mistral.ai/models/ministral-3-14b-25-12", "type": "primary" }, { "label": "Mistral API changelog", "url": "https://docs.mistral.ai/resources/changelogs", "type": "primary" }, { "label": "Ministral 3 14B Instruct 2512 — pinned official model card", "url": "https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512/blob/29439f81c2be264d8d393273f99e7db9c0961120/README.md", "type": "weights" }, { "label": "Ministral 3 14B Instruct — pinned Transformers configuration", "url": "https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512/blob/29439f81c2be264d8d393273f99e7db9c0961120/config.json", "type": "weights" }, { "label": "Ministral 3 14B Instruct — pinned native configuration", "url": "https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512/blob/29439f81c2be264d8d393273f99e7db9c0961120/params.json", "type": "weights" }, { "label": "Ministral 3 — technical report", "url": "https://arxiv.org/html/2601.08584v1", "type": "primary" }, { "label": "Maintainer correction of unused reasoning tokens", "url": "https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512/discussions/25", "type": "primary" }, { "label": "Ministral 3 14B — official Markdown model reference", "url": "https://docs.mistral.ai/models/ministral-3-14b-25-12.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/ministral-14b-2512.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The dated ministral-14b-2512 entry corresponds to the documented Instruct model; the separately released Reasoning checkpoint is a different variant. Mistral directly reports nominal 14B total and 14B active parameters. The card gives rounded components of 13.5B for the language model and 0.4B for vision; these are not an exact whole-checkpoint census. FP8 storage and scale tensors are not learned-parameter counts. Both configurations specify 262,144 positions. The Instruct checkpoint is explicitly non-reasoning, with native text/image input and function calling.", "field_status": {}, "audited_at": "2026-10-26T22:57:46.781429Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/321.json" }, { "id": "mistral/ministral-3-14b-instruct-2512", "slug": "mistral--ministral-3-14b-instruct-2512", "name": "Ministral 3 14B Instruct 2512", "developer": "Mistral", "developer_id": "mistral", "family": "ministral", "description": "Instruction-tuned model with text and image input, text output and native function calling.", "release_date": "2025-12-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 14, "active_parameters_b": 14, "parameters_source": "primary_source", "architecture": "dense", "context_window": 262144, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://mistral.ai/news/mistral-3/", "source_type": "primary", "sources": [ { "label": "Introducing Mistral 3", "url": "https://mistral.ai/news/mistral-3/", "type": "primary" }, { "label": "Ministral 3 14B Instruct 2512 — pinned model card", "url": "https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512/blob/29439f81c2be264d8d393273f99e7db9c0961120/README.md", "type": "weights" }, { "label": "Ministral 3 14B — model reference and exact Instruct Weights row", "url": "https://docs.mistral.ai/models/ministral-3-14b-25-12", "type": "primary" }, { "label": "Ministral 3 14B Instruct 2512 — pinned Transformers configuration", "url": "https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512/blob/29439f81c2be264d8d393273f99e7db9c0961120/config.json", "type": "weights" }, { "label": "Ministral 3 14B Instruct 2512 — pinned native configuration", "url": "https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512/blob/29439f81c2be264d8d393273f99e7db9c0961120/params.json", "type": "weights" }, { "label": "Ministral 3 — technical report", "url": "https://arxiv.org/html/2601.08584v1", "type": "primary" }, { "label": "Mistral maintainer clarification of non-reasoning status", "url": "https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512/discussions/25", "type": "primary" }, { "label": "Ministral 3 14B — official Markdown reference", "url": "https://docs.mistral.ai/models/ministral-3-14b-25-12.md", "type": "primary" }, { "label": "Ministral 3 14B Instruct 2512 — pinned chat template", "url": "https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512/blob/29439f81c2be264d8d393273f99e7db9c0961120/chat_template.jinja", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/ministral-3-14b-instruct-2512.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Ministral-3-14B-Instruct-2512", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Exact Ministral 3 14B Instruct 2512 checkpoint. Mistral explicitly reports nominal 14B total and 14B active parameters; the card rounds the language model to 13.5B and vision encoder to 0.4B. These are not exact tensor counts, and FP8 storage totals include quantization metadata. The checkpoint supports 262,144 positions, text/image input, text output and native tools. A Mistral maintainer explicitly confirms non-reasoning status; Base and Reasoning variants are separate releases.", "field_status": {}, "audited_at": "2026-10-26T22:57:22.443628Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/322.json" }, { "id": "mistral/ministral-3-3b-instruct-2512", "slug": "mistral--ministral-3-3b-instruct-2512", "name": "Ministral 3 3B Instruct 2512", "developer": "Mistral", "developer_id": "mistral", "family": "ministral", "description": "Compact open vision-language model for edge deployment, instruction following, and tool use", "release_date": "2025-12-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 3.849090048, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "dense", "context_window": 262144, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://mistral.ai/news/mistral-3/", "source_type": "primary", "sources": [ { "label": "Introducing Mistral 3 — release announcement", "url": "https://mistral.ai/news/mistral-3/", "type": "primary" }, { "label": "Ministral 3 3B Instruct 2512 — pinned model card", "url": "https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512/blob/b35d4dfe56c142746f54dbd64f579faab2744308/README.md", "type": "weights" }, { "label": "Ministral 3 3B v25.12 — official model reference and Weights table", "url": "https://docs.mistral.ai/models/ministral-3-3b-25-12", "type": "primary" }, { "label": "Ministral 3 3B Instruct — pinned Transformers configuration", "url": "https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512/blob/b35d4dfe56c142746f54dbd64f579faab2744308/config.json", "type": "weights" }, { "label": "Ministral 3 3B Instruct — pinned native configuration", "url": "https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512/blob/b35d4dfe56c142746f54dbd64f579faab2744308/params.json", "type": "weights" }, { "label": "Ministral 3 — technical report", "url": "https://arxiv.org/html/2601.08584v1", "type": "primary" }, { "label": "Maintainer clarification: Ministral 3 3B Instruct is non-reasoning", "url": "https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512/discussions/17", "type": "primary" }, { "label": "Official complete checkpoint tensor header — bounded byte-range inspection", "url": "https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512/resolve/b35d4dfe56c142746f54dbd64f579faab2744308/model.safetensors", "type": "weights" }, { "label": "Ministral 3 3B Instruct — pinned native chat template", "url": "https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512/blob/b35d4dfe56c142746f54dbd64f579faab2744308/chat_template.jinja", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/ministral-3-3b-instruct-2512.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete checkpoint: 3,849,090,048 learned parameters (3.429006336B language, 0.403305472B vision and 0.01677824B projector), counted from the full safetensors header with tied embeddings counted once and 364 quantization-scale scalars excluded. Mistral labels this size class 3B, while its card describes 3.4B language plus 0.4B vision components. The official Active (B)=3 is retained as a nominal size-class disclosure, not an exact active-weight count; the discrepancy with the complete total does not imply MoE sparsity. This is the FP8 Instruct checkpoint, explicitly non-reasoning, with native text/image input, text output, tools and 262,144-token context.", "parameter_components_b": { "language_model": 3.429006336, "vision_encoder": 0.403305472, "projector": 0.01677824 }, "field_status": {}, "audited_at": "2026-10-26T22:59:16.741911Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/323.json" }, { "id": "mistral/ministral-3-8b-instruct-2512", "slug": "mistral--ministral-3-8b-instruct-2512", "name": "Ministral 3 8B Instruct 2512", "developer": "Mistral", "developer_id": "mistral", "family": "ministral", "description": "Multimodal instruction model for local assistants and edge deployment, with native function calling.", "release_date": "2025-12-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 8.91802624, "active_parameters_b": 8, "parameters_source": "primary_source", "architecture": "dense", "context_window": 262144, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://mistral.ai/news/mistral-3/", "source_type": "primary", "sources": [ { "label": "Introducing Mistral 3", "url": "https://mistral.ai/news/mistral-3/", "type": "primary" }, { "label": "Ministral 3 8B v25.12 — model reference and Weights table", "url": "https://docs.mistral.ai/models/ministral-3-8b-25-12", "type": "primary" }, { "label": "Ministral 3 8B — official Markdown reference", "url": "https://docs.mistral.ai/models/ministral-3-8b-25-12.md", "type": "primary" }, { "label": "Ministral 3 8B Instruct 2512 — pinned model card", "url": "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512/blob/5b26027e7b19eeb4b7352e1fed3926375dd2cb4d/README.md", "type": "weights" }, { "label": "Ministral 3 8B Instruct — pinned Transformers configuration", "url": "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512/blob/5b26027e7b19eeb4b7352e1fed3926375dd2cb4d/config.json", "type": "weights" }, { "label": "Ministral 3 8B Instruct — pinned native configuration", "url": "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512/blob/5b26027e7b19eeb4b7352e1fed3926375dd2cb4d/params.json", "type": "weights" }, { "label": "Ministral 3 technical report", "url": "https://arxiv.org/html/2601.08584v1", "type": "primary" }, { "label": "Exact Instruct maintainer correction of unused reasoning tokens", "url": "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512/discussions/8", "type": "primary" }, { "label": "Pinned complete Transformers tensor index", "url": "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512/blob/5b26027e7b19eeb4b7352e1fed3926375dd2cb4d/model.safetensors.index.json", "type": "weights" }, { "label": "Pinned shard 1 of 3 — bounded tensor header", "url": "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512/resolve/5b26027e7b19eeb4b7352e1fed3926375dd2cb4d/model-00001-of-00003.safetensors", "type": "weights" }, { "label": "Pinned shard 2 of 3 — bounded tensor header", "url": "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512/resolve/5b26027e7b19eeb4b7352e1fed3926375dd2cb4d/model-00002-of-00003.safetensors", "type": "weights" }, { "label": "Pinned shard 3 of 3 — bounded tensor header", "url": "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512/resolve/5b26027e7b19eeb4b7352e1fed3926375dd2cb4d/model-00003-of-00003.safetensors", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/ministral-3-8b-instruct-2512.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The full released checkpoint contains 8,918,026,240 learned parameters: 8,489,553,920 language, 403,305,472 vision and 25,166,848 projector parameters. This count excludes 476 scalar FP8 scales. Mistral’s 8B total/active labels are nominal size classes; the 8B active value is retained as published and is not an exact per-token count. The card separately gives rounded 8.4B language and 0.4B vision figures. This is the non-reasoning Instruct variant, with a 262,144-token context and native image input and tool calls.", "parameter_components_b": { "language_model": 8.48955392, "vision_encoder": 0.403305472, "multimodal_projector": 0.025166848 }, "field_status": {}, "audited_at": "2026-10-26T23:00:26.344447Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/324.json" }, { "id": "mistral/ministral-3b-2512", "slug": "mistral--ministral-3b-2512", "name": "Ministral 3 3B 2512", "developer": "Mistral", "developer_id": "mistral", "family": "ministral", "description": "Instruction-tuned text and image model with native function calling, designed for local and edge deployment.", "release_date": "2025-12-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 3.849090048, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "dense", "context_window": 262144, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://mistral.ai/news/mistral-3/", "source_type": "primary", "sources": [ { "label": "Introducing Mistral 3 — dated release", "url": "https://mistral.ai/news/mistral-3/", "type": "primary" }, { "label": "Ministral 3 3B — API model reference and Instruct Weights row", "url": "https://docs.mistral.ai/models/ministral-3-3b-25-12", "type": "primary" }, { "label": "Ministral 3 3B — official Markdown reference", "url": "https://docs.mistral.ai/models/ministral-3-3b-25-12.md", "type": "primary" }, { "label": "Mistral API changelog", "url": "https://docs.mistral.ai/resources/changelogs", "type": "primary" }, { "label": "Ministral 3 3B Instruct 2512 — pinned model card", "url": "https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512/blob/b35d4dfe56c142746f54dbd64f579faab2744308/README.md", "type": "weights" }, { "label": "Ministral 3 3B Instruct — pinned Transformers configuration", "url": "https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512/blob/b35d4dfe56c142746f54dbd64f579faab2744308/config.json", "type": "weights" }, { "label": "Ministral 3 3B Instruct — pinned native configuration", "url": "https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512/blob/b35d4dfe56c142746f54dbd64f579faab2744308/params.json", "type": "weights" }, { "label": "Ministral 3 — technical report", "url": "https://arxiv.org/html/2601.08584v1", "type": "primary" }, { "label": "Mistral maintainer clarification of non-reasoning status", "url": "https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512/discussions/17", "type": "primary" }, { "label": "Complete official checkpoint header — bounded tensor census", "url": "https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512/resolve/b35d4dfe56c142746f54dbd64f579faab2744308/model.safetensors", "type": "weights" }, { "label": "Ministral 3 3B Instruct — pinned chat template", "url": "https://huggingface.co/mistralai/Ministral-3-3B-Instruct-2512/blob/b35d4dfe56c142746f54dbd64f579faab2744308/chat_template.jinja", "type": "weights" }, { "label": "OpenRouter exact dated-model identity", "url": "https://openrouter.ai/api/v1/models", "type": "catalog" }, { "label": "OpenRouter exact-model Mistral route limits", "url": "https://openrouter.ai/api/v1/models/mistralai/ministral-3b-2512/endpoints", "type": "catalog" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/ministral-3b-2512.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The dated ministral-3b-2512 entry maps to Ministral-3-3B-Instruct-2512. Complete checkpoint: 3,849,090,048 learned parameters (3.429006336B language, 0.403305472B vision and 0.01677824B projector), with tied embeddings counted once and 364 FP8 quantization-scale scalars excluded. Mistral’s 3B total/active labels are nominal size-class figures; Active (B)=3 is retained as directly disclosed, not an exact active-weight count. Native docs and checkpoint support 262,144 tokens, while OpenRouter’s Mistral routes currently advertise 131,072. The Instruct checkpoint is explicitly non-reasoning, with native vision and tool calling; Base and Reasoning variants are separate.", "parameter_components_b": { "language_model": 3.429006336, "vision_encoder": 0.403305472, "projector": 0.01677824 }, "field_status": {}, "audited_at": "2026-10-26T23:02:40.126313Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/326.json" }, { "id": "mistral/ministral-8b-2512", "slug": "mistral--ministral-8b-2512", "name": "Ministral 3 8B 2512", "developer": "Mistral", "developer_id": "mistral", "family": "ministral", "description": "Instruction-tuned text and image model for local and edge deployment, with native function calling.", "release_date": "2025-12-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 8.91802624, "active_parameters_b": 8, "parameters_source": "primary_source", "architecture": "dense", "context_window": 262144, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://mistral.ai/news/mistral-3/", "source_type": "primary", "sources": [ { "label": "Introducing Mistral 3", "url": "https://mistral.ai/news/mistral-3/", "type": "primary" }, { "label": "Ministral 3 8B v25.12 — exact model reference and Weights table", "url": "https://docs.mistral.ai/models/ministral-3-8b-25-12", "type": "primary" }, { "label": "Mistral API changelog", "url": "https://docs.mistral.ai/resources/changelogs", "type": "primary" }, { "label": "Ministral 3 8B — official Markdown reference", "url": "https://docs.mistral.ai/models/ministral-3-8b-25-12.md", "type": "primary" }, { "label": "OpenRouter exact dated model identity", "url": "https://openrouter.ai/api/v1/models", "type": "catalog" }, { "label": "Ministral 3 8B Instruct 2512 — pinned model card", "url": "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512/blob/5b26027e7b19eeb4b7352e1fed3926375dd2cb4d/README.md", "type": "weights" }, { "label": "Ministral 3 8B Instruct — pinned Transformers configuration", "url": "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512/blob/5b26027e7b19eeb4b7352e1fed3926375dd2cb4d/config.json", "type": "weights" }, { "label": "Ministral 3 8B Instruct — pinned native configuration", "url": "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512/blob/5b26027e7b19eeb4b7352e1fed3926375dd2cb4d/params.json", "type": "weights" }, { "label": "Ministral 3 — technical report", "url": "https://arxiv.org/html/2601.08584v1", "type": "primary" }, { "label": "Mistral maintainer clarification of non-reasoning status", "url": "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512/discussions/8", "type": "primary" }, { "label": "Pinned complete Transformers tensor index", "url": "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512/blob/5b26027e7b19eeb4b7352e1fed3926375dd2cb4d/model.safetensors.index.json", "type": "weights" }, { "label": "Pinned checkpoint shard 1 of 3 — complete bounded header", "url": "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512/resolve/5b26027e7b19eeb4b7352e1fed3926375dd2cb4d/model-00001-of-00003.safetensors", "type": "weights" }, { "label": "Pinned checkpoint shard 2 of 3 — complete bounded header", "url": "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512/resolve/5b26027e7b19eeb4b7352e1fed3926375dd2cb4d/model-00002-of-00003.safetensors", "type": "weights" }, { "label": "Pinned checkpoint shard 3 of 3 — complete bounded header", "url": "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512/resolve/5b26027e7b19eeb4b7352e1fed3926375dd2cb4d/model-00003-of-00003.safetensors", "type": "weights" }, { "label": "Ministral 3 8B Instruct — pinned chat template", "url": "https://huggingface.co/mistralai/Ministral-3-8B-Instruct-2512/blob/5b26027e7b19eeb4b7352e1fed3926375dd2cb4d/chat_template.jinja", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/ministral-8b-2512.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The dated ministral-8b-2512 API entry maps to Ministral-3-8B-Instruct-2512. The complete released checkpoint contains 8,918,026,240 learned parameters: 8,489,553,920 language, 403,305,472 vision and 25,166,848 projector parameters, excluding 476 scalar FP8 scales. Mistral’s 8B total/active labels are nominal size classes; the active value is retained as published and is not an exact per-token count or evidence of MoE sparsity. The card separately gives rounded 8.4B language and 0.4B vision components. This Instruct model is explicitly non-reasoning, supports 262,144 positions and native text/image input with tools; it is distinct from the original 2024 Ministral 8B.", "parameter_components_b": { "language_model": 8.48955392, "multimodal_projector": 0.025166848, "vision_encoder": 0.403305472 }, "field_status": {}, "audited_at": "2026-10-26T23:03:08.517559Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/327.json" }, { "id": "mistral/mistral-large-2512", "slug": "mistral--mistral-large-2512", "name": "Mistral Large 3", "developer": "Mistral", "developer_id": "mistral", "family": "mistral-large", "description": "Mistral Large 3 is an Apache-2.0 multimodal mixture-of-experts instruction model for text and image understanding, multilingual chat, coding, and tool use.", "release_date": "2025-12-02", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 675, "active_parameters_b": 41, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://docs.mistral.ai/models/mistral-large-3-25-12", "source_type": "primary", "sources": [ { "label": "Mistral Large 3 model documentation", "url": "https://docs.mistral.ai/models/mistral-large-3-25-12", "type": "primary" }, { "label": "Mistral 3 release announcement", "url": "https://mistral.ai/news/mistral-3/", "type": "primary" }, { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Mistral-Large-3-675B-Instruct-2512", "type": "weights" }, { "label": "Official Large 3 architecture configuration", "url": "https://huggingface.co/mistralai/Mistral-Large-3-675B-Instruct-2512/blob/383ffea2c7d60dfd44ca960e8e691709d4fdb9cd/params.json", "type": "weights" }, { "label": "Mistral context limits", "url": "https://docs.mistral.ai/resources/known-limitations", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/mistral-large-2512.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "675B total and 41B active are Mistral's nominal full-model disclosures, not exact tensor counts. Its card separately gives a rounded 673B language model with 39B active and a 2.5B vision encoder; rounded components are not an exact additive census, and vision execution is input-dependent. The checkpoint also contains vision merger/adapter weights. Supported context is 262144 tokens, while max_position_embeddings is 294912. This December 2025 Instruct release is explicitly non-reasoning.", "parameter_components_b": { "language_model": 673, "vision_encoder": 2.5 }, "field_status": {}, "audited_at": "2026-10-26T23:16:22Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/335.json" }, { "id": "arcee-ai/trinity-mini", "slug": "arcee-ai--trinity-mini", "name": "Trinity Mini", "developer": "Arcee AI", "developer_id": "arcee-ai", "family": "trinity", "description": "Open-weight Trinity MoE language model with dedicated thinking and structured tool calling.", "release_date": "2025-12-01", "date_precision": "day", "open_weights": true, "license": "OpenMDW-1.1", "parameters_b": 26.1239744, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.arcee.ai/blog/the-trinity-manifesto", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/arcee-ai/Trinity-Mini", "type": "weights" }, { "label": "Announcement", "url": "https://www.arcee.ai/blog/the-trinity-manifesto", "type": "announcement" }, { "label": "Trinity-Mini pinned configuration", "url": "https://huggingface.co/arcee-ai/Trinity-Mini/raw/cdb81e8130815fec299cccc8f8700004b8366e1e/config.json", "type": "primary" }, { "label": "Trinity-Mini OpenMDW-1.1 license", "url": "https://huggingface.co/arcee-ai/Trinity-Mini/raw/cdb81e8130815fec299cccc8f8700004b8366e1e/LICENSE", "type": "primary" }, { "label": "Arcee Trinity Large Technical Report", "url": "https://arxiv.org/pdf/2602.17004v1", "type": "primary" }, { "label": "Trinity-Mini pinned implementation", "url": "https://huggingface.co/arcee-ai/Trinity-Mini/raw/cdb81e8130815fec299cccc8f8700004b8366e1e/modeling_afmoe.py", "type": "primary" }, { "label": "Trinity-Mini pinned checkpoint index", "url": "https://huggingface.co/arcee-ai/Trinity-Mini/raw/cdb81e8130815fec299cccc8f8700004b8366e1e/model.safetensors.index.json", "type": "primary" }, { "label": "Trinity-Mini complete pinned safetensors headers", "url": "https://huggingface.co/arcee-ai/Trinity-Mini/tree/cdb81e8130815fec299cccc8f8700004b8366e1e", "type": "primary" }, { "label": "Trinity is moving to OpenMDW-1.1", "url": "https://www.arcee.ai/blog/trinity-is-moving-to-openmdw-1-1", "type": "primary" }, { "label": "Trinity-Mini release history", "url": "https://huggingface.co/api/models/arcee-ai/Trinity-Mini/commits/main", "type": "primary" }, { "label": "Trinity-Mini pinned thinking and tool template", "url": "https://huggingface.co/arcee-ai/Trinity-Mini/raw/cdb81e8130815fec299cccc8f8700004b8366e1e/chat_template.jinja", "type": "primary" }, { "label": "Teaching an Open Model to Do Science", "url": "https://www.arcee.ai/blog/teaching-an-open-model-to-do-science", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/arcee-ai/trinity-mini.toml", "type": "catalog" }, { "label": "License", "url": "https://huggingface.co/arcee-ai/Trinity-Mini/blob/main/LICENSE", "type": "license" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete checkpoint contains 26.1239744B parameters, including 3840 learned non-gradient routing biases. Approximately 3B active is the developer nominal per-token count, with eight routed experts plus one shared expert in each of 30 MoE layers; embedding-accounting conventions differ. Exact configuration context is 131072. OpenMDW-1.1 replaced launch Apache-2.0 licensing in May 2026.", "parameter_components_b": { "output_head": 0.409993216, "input_embeddings": 0.409993216, "other_norms": 0.000264192, "dense_mlp": 0.075497472, "attention_including_qk_norms": 0.872423424, "expert_balancing_bias": 0.00000384, "routed_experts": 24.15919104, "router_weights": 0.00786432, "shared_experts": 0.18874368 }, "field_status": {}, "audited_at": "2026-10-26T17:26:02.039376+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/120.json" }, { "id": "arcee-ai/trinity-nano-preview", "slug": "arcee-ai--trinity-nano-preview", "name": "Trinity Nano Preview", "developer": "Arcee AI", "developer_id": "arcee-ai", "family": "trinity", "description": "Experimental open-weight Trinity MoE chat model with structured tool calling.", "release_date": "2025-12-01", "date_precision": "day", "open_weights": true, "license": "OpenMDW-1.1", "parameters_b": 6.120003328, "active_parameters_b": 1, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://www.arcee.ai/blog/the-trinity-manifesto", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/arcee-ai/Trinity-Nano-Preview", "type": "weights" }, { "label": "Announcement", "url": "https://www.arcee.ai/blog/the-trinity-manifesto", "type": "announcement" }, { "label": "Trinity-Nano-Preview pinned configuration", "url": "https://huggingface.co/arcee-ai/Trinity-Nano-Preview/raw/b576b32b8bba7f5218312c9509e494537f8d282d/config.json", "type": "primary" }, { "label": "Trinity-Nano-Preview OpenMDW-1.1 license", "url": "https://huggingface.co/arcee-ai/Trinity-Nano-Preview/raw/b576b32b8bba7f5218312c9509e494537f8d282d/LICENSE", "type": "primary" }, { "label": "Arcee Trinity Large Technical Report", "url": "https://arxiv.org/pdf/2602.17004v1", "type": "primary" }, { "label": "Trinity-Nano-Preview pinned model implementation", "url": "https://huggingface.co/arcee-ai/Trinity-Nano-Preview/raw/b576b32b8bba7f5218312c9509e494537f8d282d/modeling_afmoe.py", "type": "primary" }, { "label": "Trinity-Nano-Preview pinned checkpoint index", "url": "https://huggingface.co/arcee-ai/Trinity-Nano-Preview/raw/b576b32b8bba7f5218312c9509e494537f8d282d/model.safetensors.index.json", "type": "primary" }, { "label": "Trinity-Nano-Preview complete pinned checkpoint headers", "url": "https://huggingface.co/arcee-ai/Trinity-Nano-Preview/tree/b576b32b8bba7f5218312c9509e494537f8d282d", "type": "primary" }, { "label": "Trinity-Nano-Preview repository history", "url": "https://huggingface.co/api/models/arcee-ai/Trinity-Nano-Preview/commits/main", "type": "primary" }, { "label": "Trinity is moving to OpenMDW-1.1", "url": "https://www.arcee.ai/blog/trinity-is-moving-to-openmdw-1-1", "type": "primary" }, { "label": "Trinity-Nano-Preview pinned chat template", "url": "https://huggingface.co/arcee-ai/Trinity-Nano-Preview/raw/b576b32b8bba7f5218312c9509e494537f8d282d/chat_template.jinja", "type": "primary" }, { "label": "Trinity-Nano-Preview launch-era model card", "url": "https://huggingface.co/arcee-ai/Trinity-Nano-Preview/raw/889c8c480f558d1410179f6c9880b148172a900d/README.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/arcee-ai/trinity-nano-preview.toml", "type": "catalog" }, { "label": "License", "url": "https://huggingface.co/arcee-ai/Trinity-Nano-Preview/blob/main/LICENSE", "type": "license" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Full checkpoint contains 6.120003328B learned parameters, including 6912 expert-balancing biases updated outside backpropagation. Approximately 1B active is the developer nominal per-token figure; exact full-module accounting is 1.023923968B, with eight routed experts plus the shared/dense/attention components. Configured context is 131072 tokens. Nano Preview is explicitly non-thinking despite a generic reasoning-parser flag in its card. Current OpenMDW-1.1 superseded launch Apache-2.0 in May 2026.", "parameter_components_b": { "expert_balancing_bias": 0.000006912, "input_embeddings": 0.204996608, "other_norms": 0.0002304, "dense_mlp": 0.018874368, "attention_including_qk_norms": 0.205535232, "routed_experts": 5.435817984, "router_weights": 0.007077888, "shared_experts": 0.042467328, "output_head": 0.204996608 }, "field_status": {}, "audited_at": "2026-10-26T17:26:41.880134+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/121.json" }, { "id": "deepseek/deepseek-v3.2", "slug": "deepseek--deepseek-v3.2", "name": "DeepSeek V3.2", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek", "description": "Hybrid-reasoning DeepSeek model with thinking and non-thinking modes, sparse attention, and tool-use", "release_date": "2025-12-01", "date_precision": "day", "open_weights": true, "license": "MIT License", "parameters_b": 683.501971712, "active_parameters_b": 37, "parameters_source": "primary_source", "architecture": "moe", "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2", "source_type": "primary", "sources": [ { "label": "DeepSeek-V3.2 official model card", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/blob/a7e62ac04ecb2c0a54d736dc46601c5606cf10a6/README.md", "type": "primary" }, { "label": "DeepSeek-V3.2 release announcement, 2025-12-01", "url": "https://api-docs.deepseek.com/news/news251201/", "type": "primary" }, { "label": "DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/a7e62ac04ecb2c0a54d736dc46601c5606cf10a6/assets/paper.pdf", "type": "primary" }, { "label": "DeepSeek-V3.2 pinned checkpoint configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/blob/a7e62ac04ecb2c0a54d736dc46601c5606cf10a6/config.json", "type": "weights" }, { "label": "DeepSeek-V3.2 tensor index and all 163 pinned safetensors headers", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/a7e62ac04ecb2c0a54d736dc46601c5606cf10a6/model.safetensors.index.json", "type": "weights" }, { "label": "DeepSeek-V3.2 official inference implementation", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/blob/a7e62ac04ecb2c0a54d736dc46601c5606cf10a6/inference/model.py", "type": "primary" }, { "label": "DeepSeek-V3 weight structure and sharing documentation", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3/raw/main/README_WEIGHTS.md", "type": "primary" }, { "label": "DeepSeek-V3.1-Terminus model structure statement", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.1-Terminus/raw/main/README.md", "type": "primary" }, { "label": "DeepSeek-V4 technical report, Table 1", "url": "https://arxiv.org/html/2606.19348v1#S4.T1", "type": "primary" }, { "label": "DeepSeek-V3.2-Exp-Base checkpoint configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2-Exp-Base/raw/main/config.json", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepseek/deepseek-v3.2.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete unique checkpoint count derived from all 163 tensor headers: 683.501971712B = 671.877944064B main + 11.624027648B unique MTP. FP8 scale tensors and duplicated MTP embedding/output-head matrices are excluded; the MTP final normalization differs from the main norm and is retained. The 685.355329792B non-scale storage count includes duplicated shared matrices. The later first-party V3.2-Base comparison reports nominal 671B main/37B active; unchanged base/instruct architecture supports nominal 37B main active, not a full MTP speculative-decoding total. Supported context is advertised 128K; configured position capacity is 163840. Historical V3.2 release, distinct from V3.2-Exp and V3.2-Speciale and from current rolling API aliases.", "active_parameters_scope": "main", "parameter_components_b": { "main": 671.877944064, "mtp_unique": 11.624027648 }, "field_status": {}, "audited_at": "2026-10-26T19:12:06.826663+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/176.json" }, { "id": "bytedance-seed/seed-1-8", "slug": "bytedance-seed--seed-1-8", "name": "Seed 1.8", "developer": "ByteDance", "developer_id": "bytedance-seed", "family": "seed", "description": "ByteDance Seed model for multimodal reasoning, long-context analysis, and agent workflows", "release_date": "2025-12", "date_precision": "month", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 256000, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://seed.bytedance.com/en/blog/official-release-of-seed1-8-a-generalized-agentic-model", "source_type": "primary", "sources": [ { "label": "Official Release of Seed1.8: A Generalized Agentic Model", "url": "https://seed.bytedance.com/en/blog/official-release-of-seed1-8-a-generalized-agentic-model", "type": "primary" }, { "label": "Seed1.8 Model Card: Towards Generalized Real-World Agency", "url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/research/Seed-1.8-Modelcard.pdf", "type": "primary" }, { "label": "ByteDance-Seed-1.8 — BytePlus ModelArk", "url": "https://docs.byteplus.com/api/docs/ModelArk/2123228", "type": "primary" }, { "label": "ByteDance-Seed/Seed-1.8 official repository", "url": "https://github.com/ByteDance-Seed/Seed-1.8", "type": "primary" }, { "label": "Seed Model Portfolio", "url": "https://seed.bytedance.com/en/seed_model_portfolio", "type": "primary" }, { "label": "Seed2.0 Model Card", "url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2/0214/Seed2.0%20Model%20Card.pdf", "type": "primary" }, { "label": "Seed2.1 Model Card", "url": "https://lf3-static.bytednsdoc.com/obj/eden-cn/lapzild-tss/ljhwZthlaukjlkulzlp/seed2.1/Seed2_1_Model_Card.pdf", "type": "primary" }, { "label": "BytePlus Service Specific Terms", "url": "https://docs.byteplus.com/en/docs/legal/docs-service-specific-terms", "type": "primary" }, { "label": "BytePlus current model list", "url": "https://docs.byteplus.com/en/docs/modelark/model-list", "type": "primary" }, { "label": "Seed2.0 Model Card: Towards Intelligence Frontier for Real-World Complexity (June 2026)", "url": "https://arxiv.org/abs/2607.00248", "type": "primary" }, { "label": "Introduction to techniques used in Seed1.6", "url": "https://seed.bytedance.com/en/blog/introduction-to-techniques-used-in-seed1-6", "type": "primary" }, { "label": "Seed2.0 current model comparison page", "url": "https://seed.bytedance.com/en/seed2", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/bytedance-seed/seed-1-8.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Seed1.8 is documented internationally as ByteDance-Seed-1.8 (seed-1-8-251228); the launch announcement maps it to Volcano Engine Doubao1.8. Public launch is December 2025: official portfolio says December 17, while the launch announcement says December 18 (midnight China / December 17 UTC); exact day is retained as unresolved. API suffix 251228 denotes a later revision. BytePlus indexed documentation includes video input and 256k context; it does not establish current endpoint operation. Nominal total and active parameter counts remain undisclosed; comparisons with Seed1.6 do not establish a size-preserving base relationship. Backbone architecture remains undisclosed. Apache-2.0 covers the official cookbook repository, not hosted weights.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T10:52:40.055402+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/131.json" }, { "id": "anthropic/claude-opus-4-5-20251101", "slug": "anthropic--claude-opus-4-5-20251101", "name": "Claude Opus 4.5", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude-opus", "description": "Claude Opus model for reasoning, coding and agentic tool use, with optional extended thinking.", "release_date": "2025-11-24", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 200000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.anthropic.com/news/claude-opus-4-5", "source_type": "primary", "sources": [ { "label": "Introducing Claude Opus 4.5", "url": "https://www.anthropic.com/news/claude-opus-4-5", "type": "primary" }, { "label": "Claude Opus 4.5 - Claude Platform Docs", "url": "https://platform.claude.com/docs/en/models/opus-4-5/overview", "type": "docs" }, { "label": "Claude Opus 4.5 System Card", "url": "https://www.anthropic.com/document/claude-opus-4-5-system-card", "type": "primary" }, { "label": "Verbalizable Representations Form a Global Workspace in Language Models", "url": "https://transformer-circuits.pub/2026/workspace/index.html", "type": "primary" }, { "label": "PDF support - Claude Platform Docs", "url": "https://platform.claude.com/docs/en/build-with-claude/pdf-support", "type": "docs" }, { "label": "GLM-5.3 and the spread of advanced cyber capabilities", "url": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities", "type": "primary" }, { "label": "Commercial Terms of Service", "url": "https://www.anthropic.com/legal/commercial-terms", "type": "primary" }, { "label": "Claude Opus 4.6 System Card", "url": "https://anthropic.com/claude-opus-4-6-system-card", "type": "primary" }, { "label": "Claude Opus 5.5 System Card", "url": "https://www.anthropic.com/claude-opus-5-5-system-card", "type": "primary" }, { "label": "Claude Haiku 5.5 System Card", "url": "https://www.anthropic.com/claude-haiku-5-5-system-card", "type": "primary" }, { "label": "Models overview - Claude Platform Docs", "url": "https://platform.claude.com/docs/en/models/overview", "type": "docs" }, { "label": "Anthropic Transparency Hub - Claude Opus 4.5 Summary Table", "url": "https://www.anthropic.com/transparency", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/anthropic/claude-opus-4-5-20251101.toml", "type": "catalog" }, { "label": "Alias documentation", "url": "https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions", "type": "docs" } ], "aliases": [ "anthropic/claude-opus-4-5" ], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Transformer architecture is supported by Anthropic interpretability research that explicitly studies Opus 4.5. Anthropic does not disclose a nominal total or active parameter count in the checked sources or identify a sized base from which either can be inferred; dense/MoE topology is also undisclosed. PDF denotes an API document input format processed through text and vision.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T10:35:29.074965Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/106.json", "alias_sources": { "anthropic/claude-opus-4-5": { "url": "https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions", "note": "Alias or alternate catalog name for the same released model; only the retained ID counts as a release." } } }, { "id": "allenai/olmo-3-1025-7b", "slug": "allenai--olmo-3-1025-7b", "name": "Olmo-3-1025-7B", "developer": "Ai2", "developer_id": "allenai", "family": "olmo", "description": null, "release_date": "2025-11-20", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 65536, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://allenai.org/blog/olmo3", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://allenai.org/blog/olmo3", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/allenai/Olmo-3-1025-7B/blob/a81bae42db3975be1671e27b9c9a56da1a9f980f/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/allenai/Olmo-3-1025-7B/blob/a81bae42db3975be1671e27b9c9a56da1a9f980f/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Publisher-trained Olmo 3 release from Ai2’s own pretrained base. The date is the public family release, not the 1025/1125 checkpoint identifier. Nominal parameter count; configuration supports 65,536 tokens." }, { "id": "allenai/olmo-3-1125-32b", "slug": "allenai--olmo-3-1125-32b", "name": "Olmo-3-1125-32B", "developer": "Ai2", "developer_id": "allenai", "family": "olmo", "description": null, "release_date": "2025-11-20", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 32, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 65536, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://allenai.org/blog/olmo3", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://allenai.org/blog/olmo3", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/allenai/Olmo-3-1125-32B/blob/c2b61dae89a1ad10e4ad5653d0e46b590902607b/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/allenai/Olmo-3-1125-32B/blob/c2b61dae89a1ad10e4ad5653d0e46b590902607b/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Publisher-trained Olmo 3 release from Ai2’s own pretrained base. The date is the public family release, not the 1025/1125 checkpoint identifier. Nominal parameter count; configuration supports 65,536 tokens." }, { "id": "allenai/olmo-3-32b-think", "slug": "allenai--olmo-3-32b-think", "name": "Olmo-3-32B-Think", "developer": "Ai2", "developer_id": "allenai", "family": "olmo", "description": null, "release_date": "2025-11-20", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 32, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 65536, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://allenai.org/blog/olmo3", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://allenai.org/blog/olmo3", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/allenai/Olmo-3-32B-Think/blob/f2edda15216e738ef2bb73771e11890e152b2112/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/allenai/Olmo-3-32B-Think/blob/f2edda15216e738ef2bb73771e11890e152b2112/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Publisher-trained Olmo 3 release from Ai2’s own pretrained base. This is the final released post-trained checkpoint, not an intermediate training step. Nominal parameter count; configuration supports 65,536 tokens." }, { "id": "allenai/olmo-3-7b-instruct", "slug": "allenai--olmo-3-7b-instruct", "name": "Olmo-3-7B-Instruct", "developer": "Ai2", "developer_id": "allenai", "family": "olmo", "description": null, "release_date": "2025-11-20", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 65536, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://allenai.org/blog/olmo3", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://allenai.org/blog/olmo3", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/allenai/Olmo-3-7B-Instruct/blob/6e5971d9eba42665f5bd5a0fcf047f299ce1dccc/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/allenai/Olmo-3-7B-Instruct/blob/6e5971d9eba42665f5bd5a0fcf047f299ce1dccc/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Publisher-trained Olmo 3 release from Ai2’s own pretrained base. This is the final released post-trained checkpoint, not an intermediate training step. Nominal parameter count; configuration supports 65,536 tokens." }, { "id": "allenai/olmo-3-7b-rl-zero-code", "slug": "allenai--olmo-3-7b-rl-zero-code", "name": "Olmo-3-7B-RL-Zero-Code", "developer": "Ai2", "developer_id": "allenai", "family": "olmo", "description": null, "release_date": "2025-11-20", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 65536, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://allenai.org/blog/olmo3", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://allenai.org/blog/olmo3", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/allenai/Olmo-3-7B-RL-Zero-Code/blob/a961af69d75f9c52b3ae4db7b1664d890229b7ca/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/allenai/Olmo-3-7B-RL-Zero-Code/blob/a961af69d75f9c52b3ae4db7b1664d890229b7ca/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Publisher-trained Olmo 3 release from Ai2’s own pretrained base. This is the final released domain-specific reinforcement-learning checkpoint, not an intermediate training step. Nominal parameter count; configuration supports 65,536 tokens." }, { "id": "allenai/olmo-3-7b-rl-zero-general", "slug": "allenai--olmo-3-7b-rl-zero-general", "name": "Olmo-3-7B-RL-Zero-General", "developer": "Ai2", "developer_id": "allenai", "family": "olmo", "description": null, "release_date": "2025-11-20", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 65536, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://allenai.org/blog/olmo3", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://allenai.org/blog/olmo3", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/allenai/Olmo-3-7B-RL-Zero-General/blob/24e5814cb36944d3eb36a49679d02c2b3bb46195/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/allenai/Olmo-3-7B-RL-Zero-General/blob/24e5814cb36944d3eb36a49679d02c2b3bb46195/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Publisher-trained Olmo 3 release from Ai2’s own pretrained base. This is the final released domain-specific reinforcement-learning checkpoint, not an intermediate training step. Nominal parameter count; configuration supports 65,536 tokens." }, { "id": "allenai/olmo-3-7b-rl-zero-if", "slug": "allenai--olmo-3-7b-rl-zero-if", "name": "Olmo-3-7B-RL-Zero-IF", "developer": "Ai2", "developer_id": "allenai", "family": "olmo", "description": null, "release_date": "2025-11-20", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 65536, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://allenai.org/blog/olmo3", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://allenai.org/blog/olmo3", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/allenai/Olmo-3-7B-RL-Zero-IF/blob/3cae007c5324a9a06ee6780f03b848bc7b8fd440/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/allenai/Olmo-3-7B-RL-Zero-IF/blob/3cae007c5324a9a06ee6780f03b848bc7b8fd440/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Publisher-trained Olmo 3 release from Ai2’s own pretrained base. This is the final released domain-specific reinforcement-learning checkpoint, not an intermediate training step. Nominal parameter count; configuration supports 65,536 tokens." }, { "id": "allenai/olmo-3-7b-rl-zero-math", "slug": "allenai--olmo-3-7b-rl-zero-math", "name": "Olmo-3-7B-RL-Zero-Math", "developer": "Ai2", "developer_id": "allenai", "family": "olmo", "description": null, "release_date": "2025-11-20", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 65536, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://allenai.org/blog/olmo3", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://allenai.org/blog/olmo3", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/allenai/Olmo-3-7B-RL-Zero-Math/blob/dee26a4074b88215cd1326472e1ab14d0259d892/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/allenai/Olmo-3-7B-RL-Zero-Math/blob/dee26a4074b88215cd1326472e1ab14d0259d892/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Publisher-trained Olmo 3 release from Ai2’s own pretrained base. This is the final released domain-specific reinforcement-learning checkpoint, not an intermediate training step. Nominal parameter count; configuration supports 65,536 tokens." }, { "id": "allenai/olmo-3-7b-think", "slug": "allenai--olmo-3-7b-think", "name": "Olmo-3-7B-Think", "developer": "Ai2", "developer_id": "allenai", "family": "olmo", "description": null, "release_date": "2025-11-20", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 65536, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://allenai.org/blog/olmo3", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://allenai.org/blog/olmo3", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/allenai/Olmo-3-7B-Think/blob/d97e442d7cc678210054dbcc9b440894d62c89a4/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/allenai/Olmo-3-7B-Think/blob/d97e442d7cc678210054dbcc9b440894d62c89a4/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Publisher-trained Olmo 3 release from Ai2’s own pretrained base. This is the final released post-trained checkpoint, not an intermediate training step. Nominal parameter count; configuration supports 65,536 tokens." }, { "id": "google/gemini-3-pro-image", "slug": "google--gemini-3-pro-image", "name": "Gemini 3 pro image", "developer": "Google", "developer_id": "google", "family": "gemini", "description": null, "release_date": "2025-11-20", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 65536, "modalities": { "input": [ "text", "image" ], "output": [ "text", "image" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.google.dev/gemini-api/docs/changelog", "source_type": "primary", "sources": [ { "label": "Dated publisher API releases", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "documentation" }, { "label": "Publisher interleaved text and image capability", "url": "https://ai.google.dev/gemini-api/docs/image-generation", "type": "documentation" }, { "label": "Publisher model specifications", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3-pro-image", "type": "documentation" } ], "aliases": [ "google/gemini-3-pro-image-preview" ], "first_seen": "2026-10-29T10:31:17.277Z", "last_seen": "2026-10-29T10:31:17.277Z", "upstream_present": false, "availability": "public", "notes": "Generates interleaved text and images. Date records first preview availability; general availability followed 2026-05-28. Preview and stable serving names are retained as aliases, without evidence of a separate trained checkpoint. Context and capabilities describe current published model documentation; preview limits may have differed." }, { "id": "openai/gpt-5.1-codex-max", "slug": "openai--gpt-5.1-codex-max", "name": "GPT-5.1-Codex-Max", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5.1", "description": "GPT-5.1 coding model trained for work across successive context windows, with reasoning, text and image input, and function calling.", "release_date": "2025-11-19", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 400000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/gpt-5-1-codex-max/", "source_type": "primary", "sources": [ { "label": "Building more with GPT-5.1-Codex-Max", "url": "https://openai.com/index/gpt-5-1-codex-max/", "type": "primary" }, { "label": "API changelog", "url": "https://developers.openai.com/api/docs/changelog.md", "type": "primary" }, { "label": "GPT-5.1-Codex-Max API model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.1-codex-max.md", "type": "primary" }, { "label": "GPT-5.1-Codex-Max System Card", "url": "https://cdn.openai.com/pdf/2a7d98b1-57e5-4147-8d0e-683894d782ae/5p1_codex_max_card_03.pdf", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201.md", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "Addendum to GPT-5.2 System Card: GPT-5.2-Codex", "url": "https://cdn.openai.com/pdf/ac7c37ae-7f4c-4442-b741-2eabdeaf77e0/oai_5_2_Codex.pdf", "type": "primary" }, { "label": "A Practical Approach to Verifying Code at Scale", "url": "https://alignment.openai.com/scaling-code-verification/", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "All OpenAI API models", "url": "https://developers.openai.com/api/docs/models/all.md", "type": "primary" }, { "label": "Details about METR's evaluation of OpenAI GPT-5.1-Codex-Max", "url": "https://metr.org/evaluations/gpt-5-1-codex-max-report/", "type": "primary" }, { "label": "Claude Opus 4.5 System Card", "url": "https://www-cdn.anthropic.com/bf10f64990cfda0ba858290be7b8cc6317685f47.pdf", "type": "primary" }, { "label": "IDE-Bench: Evaluating Large Language Models as IDE Agents on Real-World Software Engineering Tasks", "url": "https://arxiv.org/html/2601.20886v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5.1-codex-max.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "GPT-5.1-Codex-Max first became available in Codex on November 19, 2025; Responses API access followed December 4. The system-card cover date, November 18, does not establish public release. The model documentation lists only the undated alias, with no dated immutable snapshot. Its context window is 400,000 tokens and maximum output is 128,000; compaction across successive windows enables longer cumulative tasks without enlarging a single context window. PDF input denotes API conversion into text and page images. Transformer is a qualified inference from exact Max prompt-caching support and the applicable attention-KV mechanism; complete block inventory and dense/MoE routing remain undisclosed. No nominal total or active count is established by the reviewed sources, and the updated reasoning-base relationship supplies no counted base for a size inference. API access shut down on July 23, 2026; these are last-documented capabilities, not a claim of current endpoint operation.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T16:45:14.250159+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/421.json" }, { "id": "xai/grok-4.1-fast", "slug": "xai--grok-4.1-fast", "name": "Grok 4.1 Fast (Non-Reasoning)", "developer": "xAI", "developer_id": "xai", "family": "grok", "description": "Non-reasoning variant of Grok 4.1 Fast, with text and image input, function calling and a two-million-token context window.", "release_date": "2025-11-19", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": null, "context_window": 2000000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://x.ai/news/grok-4-1-fast", "source_type": "primary", "sources": [ { "label": "Grok 4.1 Fast and Agent Tools API, November 19, 2025", "url": "https://x.ai/news/grok-4-1-fast", "type": "primary" }, { "label": "Provider mapping for the unqualified Grok 4.1 Fast identifier", "url": "https://api.ofox.ai/v1/models/x-ai/grok-4.1-fast", "type": "provider" }, { "label": "Original upstream variant mapping", "url": "https://raw.githubusercontent.com/anomalyco/models.dev/dev/models/xai/grok-4.1-fast.toml", "type": "catalog" }, { "label": "Grok 4.1 Fast historical model documentation", "url": "https://docs.oracle.com/en-us/iaas/Content/generative-ai/xai-grok-4-1-fast.htm", "type": "provider" }, { "label": "May 15, 2026 model retirement", "url": "https://docs.x.ai/developers/migration/may-15-retirement", "type": "primary" }, { "label": "Launch-period Enterprise Terms of Service", "url": "https://x.ai/legal/terms-of-service-enterprise/previous-2025-06-27", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/xai/grok-4.1-fast.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The catalog identifier grok-4.1-fast denotes the non-reasoning variant, mapped to native grok-4-1-fast-non-reasoning. Both Fast variants launched November 19, 2025; this is separate from the November 17 standard Grok 4.1 release. Total and active parameter counts remain undisclosed, with no counted base lineage established; neural topology is also undisclosed. Historical specifications are retained: xAI redirects this slug to Grok 4.3 with none reasoning effort from May 15, 2026, and Oracle lists the original model as retired. A provider metadata listing is not proof that original-model inference remains available.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T18:23:34.272911+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/527.json" }, { "id": "xai/grok-4.1-fast-reasoning", "slug": "xai--grok-4.1-fast-reasoning", "name": "Grok 4.1 Fast (Reasoning)", "developer": "xAI", "developer_id": "xai", "family": "grok", "description": "Reasoning variant of Grok 4.1 Fast, with text and image input, tool calling and a two-million-token context window.", "release_date": "2025-11-19", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": null, "context_window": 2000000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://x.ai/news/grok-4-1-fast", "source_type": "primary", "sources": [ { "label": "Grok 4.1 Fast and Agent Tools API, November 19, 2025", "url": "https://x.ai/news/grok-4-1-fast", "type": "primary" }, { "label": "November 19, 2025 API release notes", "url": "https://docs.x.ai/developers/release-notes", "type": "primary" }, { "label": "Grok 4.1 Fast historical provider specifications", "url": "https://docs.oracle.com/en-us/iaas/Content/generative-ai/xai-grok-4-1-fast.htm", "type": "provider" }, { "label": "May 15, 2026 model retirement", "url": "https://docs.x.ai/developers/migration/may-15-retirement", "type": "primary" }, { "label": "Microsoft Foundry Grok 4.1 Fast Reasoning", "url": "https://ai.azure.com/catalog/models/grok-4-1-fast-reasoning?publisher=xAI", "type": "provider" }, { "label": "Launch-period enterprise terms", "url": "https://x.ai/legal/terms-of-service-enterprise/previous-2025-06-27", "type": "primary" }, { "label": "Grok 4 Fast launch and predecessor lineage", "url": "https://x.ai/news/grok-4-fast", "type": "primary" }, { "label": "xAI Frontier Artificial Intelligence Framework, June 30, 2026", "url": "https://media.x.ai/v1/website/xai-frontier-artificial-intelligence-framework-30-june-2026-f1f0c916.pdf", "type": "primary" }, { "label": "OVO-S-Bench: A Hierarchical Benchmark for Streaming Spatial Intelligence in Multimodal LLMs, Tables 2, 3 and 14", "url": "https://arxiv.org/html/2606.03890v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/xai/grok-4.1-fast-reasoning.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released November 19, 2025 as grok-4-1-fast-reasoning, with 2M context and text/image input; Microsoft Foundry documents a 128K route limit. Total and active parameters and neural topology remain undisclosed. Oracle names Grok 4 Fast as a training predecessor, but no counted base is established. The standard consumer Grok 4.1 card does not identify Fast. xAI retired the original endpoint May 15, 2026 and redirects its slug to Grok 4.3 with low reasoning effort; replacement specifications are not backdated.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T18:26:47.379108+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/528.json" }, { "id": "google/gemini-3-pro-preview", "slug": "google--gemini-3-pro-preview", "name": "Gemini 3 Pro Preview", "developer": "Google", "developer_id": "google", "family": "gemini-pro", "description": "Google's Gemini 3 Pro preview reasoning model for coding and multimodal understanding; its Gemini API endpoint was shut down on March 9, 2026.", "release_date": "2025-11-18", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "video", "audio", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ai.google.dev/gemini-api/docs/models/gemini-3-pro-preview", "source_type": "primary", "sources": [ { "label": "Google exact preview endpoint documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-3-pro-preview", "type": "primary" }, { "label": "Google DeepMind Gemini 3 Pro model card", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-3-Pro-Model-Card.pdf", "type": "primary" }, { "label": "Google launch announcement, November 18, 2025", "url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemini-3-developers/", "type": "primary" }, { "label": "Google exact endpoint release and shutdown dates", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "type": "primary" }, { "label": "Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Gemini 3.1 Pro model card comparisons", "url": "https://deepmind.google/models/model-cards/gemini-3-1-pro/", "type": "primary" }, { "label": "Gemini 4 Argon Safety and Security Report", "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemini-3-pro-preview.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Sparse MoE Transformer; Google identifies Gemini 3 Pro as a new model, not a fine-tune of an earlier one. Total and active parameter counts are undisclosed. Context and capabilities describe the retired gemini-3-pro-preview endpoint, shut down March 9, 2026. Distinct from Gemini 3 Pro Image and Gemini 3.1 Pro.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T12:58:32.549740+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/201.json" }, { "id": "openai/gpt-5.1-codex", "slug": "openai--gpt-5.1-codex", "name": "GPT-5.1-Codex", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5.1", "description": "GPT-5.1 variant for coding agents, with text and image input, reasoning, function calling and a 400,000-token context window.", "release_date": "2025-11-13", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 400000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/gpt-5-1-for-developers/", "source_type": "primary", "sources": [ { "label": "Introducing GPT-5.1 for developers", "url": "https://openai.com/index/gpt-5-1-for-developers/", "type": "primary" }, { "label": "Codex changelog: Introducing GPT-5.1-Codex and GPT-5.1-Codex-Mini", "url": "https://developers.openai.com/codex/changelog/", "type": "primary" }, { "label": "GPT-5.1-Codex API model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.1-codex.md", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201.md", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "Building more with GPT-5.1-Codex-Max", "url": "https://openai.com/index/gpt-5-1-codex-max/", "type": "primary" }, { "label": "GPT-5.1-Codex-Max System Card", "url": "https://cdn.openai.com/pdf/2a7d98b1-57e5-4147-8d0e-683894d782ae/5p1_codex_max_card_03.pdf", "type": "primary" }, { "label": "A Practical Approach to Verifying Code at Scale", "url": "https://alignment.openai.com/scaling-code-verification/", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "All OpenAI API models", "url": "https://developers.openai.com/api/docs/models/all.md", "type": "primary" }, { "label": "GPT-5.1 API model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.1.md", "type": "primary" }, { "label": "BioTool: A Comprehensive Tool-Calling Dataset for Enhancing Biomedical Capabilities of Large Language Models", "url": "https://arxiv.org/html/2605.05758v1", "type": "primary" }, { "label": "SWE-Refactor: A Repository-Level Benchmark for Real-World LLM-Based Code Refactoring", "url": "https://arxiv.org/html/2602.03712v1", "type": "primary" }, { "label": "Foundry Models sold by Azure: GPT-5.1 capabilities", "url": "https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5.1-codex.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Original GPT-5.1-Codex became available in Codex and the API on November 13, 2025. It is distinct from GPT-5.1-Codex-Max and mini. The API documentation describes a mutable alias and supplies no dated immutable snapshot. The documented context is 400,000 tokens, with a separate 128,000 maximum output; no maximum input is supplied. PDF input denotes API conversion into text and page images. Transformer is a qualified operational inference from exact prompt-caching support and the applicable Transformer KV-cache explanation; complete block design and dense/MoE routing remain undisclosed. Nominal total and active parameters remain undisclosed; the named GPT-5.1 base also has no published count to inherit. API access shut down on July 23, 2026; specifications are last-documented capabilities, not a claim of current endpoint operation.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T16:42:12.144705Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/420.json" }, { "id": "openai/gpt-5.1-codex-mini", "slug": "openai--gpt-5.1-codex-mini", "name": "GPT-5.1-Codex Mini", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5.1", "description": "GPT-5.1 coding variant with text and image input, reasoning, function calling and a 400,000-token context window.", "release_date": "2025-11-13", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 400000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/gpt-5-1-for-developers/", "source_type": "primary", "sources": [ { "label": "Introducing GPT-5.1 for developers", "url": "https://openai.com/index/gpt-5-1-for-developers/", "type": "primary" }, { "label": "Codex changelog", "url": "https://developers.openai.com/codex/changelog/", "type": "primary" }, { "label": "GPT-5.1-Codex Mini model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.1-codex-mini.md", "type": "primary" }, { "label": "Prompt caching", "url": "https://developers.openai.com/api/docs/guides/prompt-caching.md", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201.md", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "All OpenAI API models", "url": "https://developers.openai.com/api/docs/models/all.md", "type": "primary" }, { "label": "GPT-5.1-Codex-Max System Card", "url": "https://cdn.openai.com/pdf/2a7d98b1-57e5-4147-8d0e-683894d782ae/5p1_codex_max_card_03.pdf", "type": "primary" }, { "label": "Update to GPT-5 System Card: GPT-5.2", "url": "https://cdn.openai.com/pdf/3a4153c8-c748-4b71-8e31-aecbde944f8d/oai_5_2_system-card.pdf", "type": "primary" }, { "label": "A Practical Approach to Verifying Code at Scale", "url": "https://alignment.openai.com/scaling-code-verification/", "type": "primary" }, { "label": "GPT-5.1 Instant and GPT-5.1 Thinking System Card Addendum", "url": "https://cdn.openai.com/pdf/4173ec8d-1229-47db-96de-06d87147e07e/5_1_system_card.pdf", "type": "primary" }, { "label": "GPT-5.2-Codex System Card", "url": "https://cdn.openai.com/pdf/ac7c37ae-7f4c-4442-b741-2eabdeaf77e0/oai_5_2_Codex.pdf", "type": "primary" }, { "label": "GPT-5.3-Codex System Card", "url": "https://deploymentsafety.openai.com/gpt-5-3-codex/gpt-5-3-codex.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5.1-codex-mini.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "GPT-5.1-Codex Mini launched in Codex and the API on November 13, 2025, separately from GPT-5-Codex-Mini on November 7. Its API page lists only the undated alias, with no dated immutable snapshot. Context is 400,000 tokens and maximum output is 128,000; no maximum input is stated. PDF denotes API conversion into text and page images. Transformer is a qualified operational inference from exact caching support and the pre-GPT-5.6 attention-KV explanation; complete block design and dense/MoE routing remain undisclosed. Total and active parameter counts remain undisclosed after the October 27 source review; the documented smaller-version relationship supplies no nominal size or count-preserving base lineage. Codex access with ChatGPT sign-in ended April 14, 2026, and API access shut down July 23, 2026. These are last-documented capabilities; no inference request was made.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T16:44:45.383509Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/422.json" }, { "id": "openai/gpt-5.1", "slug": "openai--gpt-5.1", "name": "GPT-5.1", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5.1", "description": "GPT-5.1 reasoning model with text and image input and function calling.", "release_date": "2025-11-12", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 400000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/gpt-5-1/", "source_type": "primary", "sources": [ { "label": "GPT-5.1: A smarter, more conversational ChatGPT", "url": "https://openai.com/index/gpt-5-1/", "type": "primary" }, { "label": "Introducing GPT-5.1 for developers", "url": "https://openai.com/index/gpt-5-1-for-developers/", "type": "primary" }, { "label": "GPT-5.1 model documentation (Markdown)", "url": "https://developers.openai.com/api/docs/models/gpt-5.1.md", "type": "primary" }, { "label": "GPT-5.1 Instant and GPT-5.1 Thinking System Card Addendum", "url": "https://cdn.openai.com/pdf/4173ec8d-1229-47db-96de-06d87147e07e/5_1_system_card.pdf", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201.md", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "Update to GPT-5 System Card: GPT-5.2", "url": "https://cdn.openai.com/pdf/3a4153c8-c748-4b71-8e31-aecbde944f8d/oai_5_2_system-card.pdf", "type": "primary" }, { "label": "Reasoning Models Struggle to Control their Chains of Thought", "url": "https://cdn.openai.com/pdf/a21c39c1-fa07-41db-9078-973a12620117/cot_controllability.pdf", "type": "primary" }, { "label": "Sidestepping Evaluation Awareness and Anticipating Misalignment with Production Evaluations", "url": "https://alignment.openai.com/prod-evals/", "type": "primary" }, { "label": "API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "GPT-5.6 System Card", "url": "https://deploymentsafety.openai.com/gpt-5-6", "type": "primary" }, { "label": "CaveAgent: Transforming LLMs into Stateful Runtime Operators (v4, June 29, 2026)", "url": "https://arxiv.org/html/2601.01569v4", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5.1.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "GPT-5.1 Thinking first rolled out in ChatGPT on November 12, 2025; the API release and default gpt-5.1-2025-11-13 snapshot followed November 13. This record is separate from GPT-5.1 Instant (gpt-5.1-chat-latest) and Auto routing. API specifications are 400K context and 128K maximum output; reasoning effort supports none (default), low, medium and high. PDF input is processed into text and page images by the API. Transformer is a qualified inference from exact-model prompt caching and the documented pre-GPT-5.6 attention KV mechanism; the full block inventory and dense/MoE routing remain undisclosed. Total and active parameter counts remain undisclosed; the original CaveAgent comparison also marks GPT-5.1's size unavailable. The API model is deprecated with shutdown announced for April 1, 2027, still in the future at review time.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T16:40:54.989084Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/418.json" }, { "id": "openai/gpt-5.1-chat-latest", "slug": "openai--gpt-5.1-chat-latest", "name": "GPT-5.1 Chat", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5.1", "description": "GPT-5.1 Instant chat model with adaptive reasoning, text and image input, text output, and function calling in the API.", "release_date": "2025-11-12", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 128000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/gpt-5-1/", "source_type": "primary", "sources": [ { "label": "GPT-5.1: A smarter, more conversational ChatGPT", "url": "https://openai.com/index/gpt-5-1/", "type": "primary" }, { "label": "GPT-5.1 Chat API model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.1-chat-latest.md", "type": "primary" }, { "label": "OpenAI API changelog", "url": "https://developers.openai.com/api/docs/changelog.md", "type": "primary" }, { "label": "GPT-5.1 Instant and GPT-5.1 Thinking System Card Addendum", "url": "https://cdn.openai.com/pdf/4173ec8d-1229-47db-96de-06d87147e07e/5_1_system_card.pdf", "type": "primary" }, { "label": "Prompt caching", "url": "https://developers.openai.com/api/docs/guides/prompt-caching.md", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201.md", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "All OpenAI API models", "url": "https://developers.openai.com/api/docs/models/all.md", "type": "primary" }, { "label": "Update to GPT-5 System Card: GPT-5.2", "url": "https://cdn.openai.com/pdf/3a4153c8-c748-4b71-8e31-aecbde944f8d/oai_5_2_system-card.pdf", "type": "primary" }, { "label": "GPT-5.5 Instant System Card", "url": "https://deploymentsafety.openai.com/gpt-5-5-instant/gpt-5-5-instant.pdf", "type": "primary" }, { "label": "Introducing GPT-5.1 for developers", "url": "https://openai.com/index/gpt-5-1-for-developers/", "type": "primary" }, { "label": "GPT-5.3 Instant System Card", "url": "https://deploymentsafety.openai.com/gpt-5-3-instant/gpt-5-3-instant.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5.1-chat-latest.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "GPT-5.1 Instant began public ChatGPT rollout on November 12, 2025; its gpt-5.1-chat-latest API alias followed on November 13. API specifications refer to the documented alias, whose final immutable target is not published on the model page. The 128,000-token context and 16,384-token output limit are explicit in HTML and Markdown; Markdown also lists an inconsistent 272,000 maximum-input budget, left unresolved. Adaptive reasoning is explicitly documented for Instant and its API binding, although the current model page has no reasoning_tokens flag; configurable effort controls are not asserted. PDF is an API file format processed into text/page images. Transformer is a qualified operational inference from exact-alias caching support and the pre-GPT-5.6 KV-attention explanation, not a complete architecture, dense/MoE classification or parameter disclosure. No nominal total or active parameter count is published in the reviewed exact-model specifications or later GPT-5.2, GPT-5.3 Instant and GPT-5.5 Instant comparison tables; both remain undisclosed. API access shut down July 23, 2026; no endpoint test was made.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T16:41:05.650303Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/419.json" }, { "id": "moonshotai/kimi-k2-thinking", "slug": "moonshotai--kimi-k2-thinking", "name": "Kimi K2 Thinking", "developer": "Moonshot AI", "developer_id": "moonshotai", "family": "kimi-k2", "description": "Open-weight Kimi K2 reasoning model with a mixture-of-experts Transformer, MLA attention, and native tool calling for multi-step research, coding, and writing.", "release_date": "2025-11-06", "date_precision": "day", "open_weights": true, "license": "Modified MIT", "parameters_b": 1000, "active_parameters_b": 32, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.kimi.com/en/blog/kimi-k2-thinking", "source_type": "primary", "sources": [ { "label": "Kimi K2 Thinking announcement", "url": "https://www.kimi.com/en/blog/kimi-k2-thinking", "type": "primary" }, { "label": "Moonshot release-date index", "url": "https://www.kimi.com/en/blog/", "type": "primary" }, { "label": "Hugging Face", "url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "type": "weights" }, { "label": "Kimi K2 Thinking configuration", "url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking/blob/a51ccc050d73dab088bf7b0e2dd9b30ae85a4e55/config.json", "type": "primary" }, { "label": "Kimi K2 Thinking license", "url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking/blob/a51ccc050d73dab088bf7b0e2dd9b30ae85a4e55/LICENSE", "type": "primary" }, { "label": "Kimi K2 Thinking model code", "url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking/blob/a51ccc050d73dab088bf7b0e2dd9b30ae85a4e55/modeling_deepseek.py", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/moonshotai/kimi-k2-thinking.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "1T total and 32B activated are nominal developer counts, not exact counts of stored tensors. This is a decoder-only MoE Transformer with MLA and a supported 256K (262,144-token) context. Modified MIT requires prominent Kimi K2 UI attribution for commercial products/services above 100 million monthly active users or USD 20 million monthly revenue. This describes the released checkpoint and does not assert that its original API endpoint remains operational.", "field_status": {}, "audited_at": "2026-10-26T23:50:13.257Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/357.json" }, { "id": "moonshotai/kimi-k2-thinking-turbo", "slug": "moonshotai--kimi-k2-thinking-turbo", "name": "Kimi K2 Thinking Turbo", "developer": "Moonshot AI", "developer_id": "moonshotai", "family": "kimi-k2", "description": "High-speed Kimi K2 Thinking API variant for text reasoning and tool calling; the original Moonshot endpoint is deprecated.", "release_date": "2025-11-06", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": 1000, "active_parameters_b": 32, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://platform.kimi.ai/blog/posts/changelog", "source_type": "primary", "sources": [ { "label": "Moonshot official release log", "url": "https://platform.kimi.ai/blog/posts/changelog", "type": "primary" }, { "label": "Moonshot historical Turbo API specifications (archived page bundle)", "url": "https://web.archive.org/web/20251106103333if_/https://platform.moonshot.ai/_next/static/chunks/pages/docs/pricing/chat.en-US-11ca3b4cff62353d.js", "type": "primary" }, { "label": "Moonshot API model lifecycle", "url": "https://platform.kimi.ai/docs/models", "type": "primary" }, { "label": "Moonshot exact-model always-thinking classification", "url": "https://moonshotai.github.io/kimi-cli/en/guides/interaction.html", "type": "primary" }, { "label": "Moonshot exact-model tool-calling support", "url": "https://forum.moonshot.ai/t/tool-calling-specification-violation-on-moonshot-api/102", "type": "primary" }, { "label": "Hugging Face", "url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "type": "weights" }, { "label": "Moonshot exact-model high-speed generation guidance", "url": "https://platform.kimi.ai/blog/posts/coding_with_kimi_agents_setup", "type": "primary" }, { "label": "Moonshot public model-weight inventory", "url": "https://huggingface.co/api/models?author=moonshotai&limit=100", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/moonshotai/kimi-k2-thinking-turbo.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Historical Moonshot API speed variant, released November 6, 2025 and listed as deprecated since May 25, 2026. Its advertised 256K context is represented as 262144 tokens; always-thinking and tool calling are explicit. Parameter re-audit on October 27, 2026 accepts nominal 1T total/32B activated parameters through the official identification of this endpoint as the high-speed version of Kimi K2 Thinking and that named model's developer-published specification. Later first-party speed guidance corroborates this relationship; fresh searches found no Turbo-specific pruning, distillation, changed expert selection or added modalities. These are nominal model counts inferred from explicit lineage, not an exact tensor census, whole-service inventory or proof of byte-identical hosted weights. MoE architecture follows the same explicit lineage. A public standard Thinking weights link is not a verified Turbo checkpoint; the existing license assessment remains unchanged. Classified closed for the exact Turbo variant because no matching public weight release was found.", "field_status": { "license": "unverified" }, "audited_at": "2026-10-28T08:02:38.262553Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/358.json" }, { "id": "openai/gpt-oss-safeguard-120b", "slug": "openai--gpt-oss-safeguard-120b", "name": "gpt-oss-safeguard-120b", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-oss", "description": "Open-weight text model fine-tuned from gpt-oss-120b to reason over supplied policies and generate content classifications.", "release_date": "2025-10-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 116.829156672, "active_parameters_b": 5.1, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-gpt-oss-safeguard/", "source_type": "primary", "sources": [ { "label": "Introducing gpt-oss-safeguard — October 29, 2025", "url": "https://openai.com/index/introducing-gpt-oss-safeguard/", "type": "primary" }, { "label": "Pinned gpt-oss-safeguard-120b model card", "url": "https://huggingface.co/openai/gpt-oss-safeguard-120b/blob/3c7391182603991a904031244e7822488c67796d/README.md", "type": "weights" }, { "label": "Safeguard technical report — October 29, 2025", "url": "https://cdn.openai.com/pdf/08b7dee4-8bc6-4955-a219-7793fb69090c/Technical_report__Research_Preview_of_gpt_oss_safeguard.pdf", "type": "primary" }, { "label": "OpenAI open-weight models — Safeguard specifications", "url": "https://help.openai.com/en/articles/11870455-openai-open-weight-models-gpt-oss", "type": "primary" }, { "label": "Pinned exact Safeguard120b configuration", "url": "https://huggingface.co/openai/gpt-oss-safeguard-120b/blob/3c7391182603991a904031244e7822488c67796d/config.json", "type": "primary" }, { "label": "Pinned complete Safeguard120b weight index", "url": "https://huggingface.co/openai/gpt-oss-safeguard-120b/blob/3c7391182603991a904031244e7822488c67796d/model.safetensors.index.json", "type": "weights" }, { "label": "Pinned OpenAI inference modules", "url": "https://github.com/openai/gpt-oss/blob/5681a6b6552635e7b29cde94a61ca6a7aee90ebd/gpt_oss/torch/model.py", "type": "primary" }, { "label": "Transformers 4.57.1 GptOssForCausalLM implementation", "url": "https://github.com/huggingface/transformers/blob/v4.57.1/src/transformers/models/gpt_oss/modeling_gpt_oss.py", "type": "primary" }, { "label": "Transformers 4.57.1 MXFP4 implementation", "url": "https://github.com/huggingface/transformers/blob/v4.57.1/src/transformers/integrations/mxfp4.py", "type": "primary" }, { "label": "gpt-oss original architecture report", "url": "https://arxiv.org/html/2508.10925v1", "type": "primary" }, { "label": "Pinned Safeguard120b license", "url": "https://huggingface.co/openai/gpt-oss-safeguard-120b/blob/3c7391182603991a904031244e7822488c67796d/LICENSE", "type": "primary" }, { "label": "Pinned Safeguard120b Harmony chat template", "url": "https://huggingface.co/openai/gpt-oss-safeguard-120b/blob/3c7391182603991a904031244e7822488c67796d/chat_template.jinja", "type": "primary" }, { "label": "Amazon Bedrock exact Safeguard120b model card", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-openai-gpt-oss-safeguard-120b.html", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-oss-safeguard-120b.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/openai/gpt-oss-safeguard-120b", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Total is the complete logical pinned-checkpoint count, including both untied embedding matrices, norms, biases, routers and learned sinks; MXFP4 exponent scales are excluded and packed weights expanded. Exact Safeguard documentation rounds active parameters to 5.1 B; the unchanged architecture uses four experts per layer, with output head included and input embeddings excluded in OpenAI’s convention. Harmony is required. Tool calling is documented for this exact variant on AWS bedrock-mantle; tool execution depends on the hosting interface. These open weights are not served by OpenAI’s own API.", "field_status": {}, "audited_at": "2026-10-27T02:54:15.013776Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/448.json" }, { "id": "openai/gpt-oss-safeguard-20b", "slug": "openai--gpt-oss-safeguard-20b", "name": "gpt-oss-safeguard-20b", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-oss", "description": "Open-weight safety reasoning model that classifies text against developer-provided policies and generates an explanation of its decision.", "release_date": "2025-10-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 20.914757184, "active_parameters_b": 3.6, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-gpt-oss-safeguard/", "source_type": "primary", "sources": [ { "label": "Introducing gpt-oss-safeguard — October 29, 2025", "url": "https://openai.com/index/introducing-gpt-oss-safeguard/", "type": "primary" }, { "label": "Pinned gpt-oss-safeguard-20b model card", "url": "https://huggingface.co/openai/gpt-oss-safeguard-20b/resolve/8a11e17b25c973a24099d4016bf2e17dd7ec1574/README.md", "type": "weights" }, { "label": "Technical Report: Performance and baseline evaluations of gpt-oss-safeguard-120b and gpt-oss-safeguard-20b", "url": "https://cdn.openai.com/pdf/08b7dee4-8bc6-4955-a219-7793fb69090c/Technical_report__Research_Preview_of_gpt_oss_safeguard.pdf", "type": "primary" }, { "label": "OpenAI open-weight models — safeguard variants and sizing", "url": "https://help.openai.com/en/articles/11870455-openai-open-weight-models-gpt-oss", "type": "primary" }, { "label": "Pinned safeguard-20b configuration", "url": "https://huggingface.co/openai/gpt-oss-safeguard-20b/resolve/8a11e17b25c973a24099d4016bf2e17dd7ec1574/config.json", "type": "weights" }, { "label": "Pinned complete safeguard-20b checkpoint index", "url": "https://huggingface.co/openai/gpt-oss-safeguard-20b/resolve/8a11e17b25c973a24099d4016bf2e17dd7ec1574/model.safetensors.index.json", "type": "weights" }, { "label": "Pinned safeguard-20b shard1 bounded header", "url": "https://huggingface.co/openai/gpt-oss-safeguard-20b/resolve/8a11e17b25c973a24099d4016bf2e17dd7ec1574/model-00000-of-00002.safetensors", "type": "weights" }, { "label": "Pinned safeguard-20b shard2 bounded header", "url": "https://huggingface.co/openai/gpt-oss-safeguard-20b/resolve/8a11e17b25c973a24099d4016bf2e17dd7ec1574/model-00001-of-00002.safetensors", "type": "weights" }, { "label": "Pinned safeguard-20b shard3 bounded header", "url": "https://huggingface.co/openai/gpt-oss-safeguard-20b/resolve/8a11e17b25c973a24099d4016bf2e17dd7ec1574/model-00002-of-00002.safetensors", "type": "weights" }, { "label": "Pinned official gpt-oss reference implementation", "url": "https://raw.githubusercontent.com/openai/gpt-oss/5681a6b6552635e7b29cde94a61ca6a7aee90ebd/gpt_oss/torch/model.py", "type": "primary" }, { "label": "Pinned official MXFP4 checkpoint loader", "url": "https://raw.githubusercontent.com/openai/gpt-oss/5681a6b6552635e7b29cde94a61ca6a7aee90ebd/gpt_oss/torch/weights.py", "type": "primary" }, { "label": "Original gpt-oss model architecture and counting conventions", "url": "https://arxiv.org/html/2508.10925v1", "type": "primary" }, { "label": "Pinned safeguard-20b license", "url": "https://huggingface.co/openai/gpt-oss-safeguard-20b/resolve/8a11e17b25c973a24099d4016bf2e17dd7ec1574/LICENSE", "type": "weights" }, { "label": "Pinned safeguard-20b Harmony template", "url": "https://huggingface.co/openai/gpt-oss-safeguard-20b/resolve/8a11e17b25c973a24099d4016bf2e17dd7ec1574/chat_template.jinja", "type": "weights" }, { "label": "Groq tool-use support table", "url": "https://console.groq.com/docs/tool-use/overview", "type": "primary" }, { "label": "Groq safeguard-20b model reference", "url": "https://console.groq.com/docs/model/openai/gpt-oss-safeguard-20b", "type": "primary" }, { "label": "User guide for gpt-oss-safeguard", "url": "https://developers.openai.com/cookbook/articles/gpt-oss-safeguard-guide", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-oss-safeguard-20b.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/openai/gpt-oss-safeguard-20b", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete checkpoint count: 20,914,757,184 logical learned parameters, including untied embeddings/head, norms, attention sinks/biases, routers and expert biases. MXFP4 packed weights are expanded and quantization scales excluded. Official safeguard active count is nominal 3.6B per token; the unchanged base architecture uses a convention that includes unembedding and excludes input embedding. Released October 29, 2025 under Apache 2.0. Harmony is required. Tool calling is documented for this exact model by Groq; parallel/built-in tool support is not claimed.", "field_status": {}, "audited_at": "2026-10-27T02:53:45.394112Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/449.json" }, { "id": "nvidia/nemotron-nano-12b-v2-vl", "slug": "nvidia--nemotron-nano-12b-v2-vl", "name": "Nemotron Nano 12B v2 VL", "developer": "NVIDIA", "developer_id": "nvidia", "family": "nemotron", "description": "Vision-language model combining a C-RADIO encoder with a hybrid Mamba-2 and attention decoder, trained for document understanding, video comprehension and reasoning.", "release_date": "2025-10-28", "date_precision": "day", "open_weights": true, "license": "NVIDIA Open Model License", "parameters_b": 13.181860352, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "hybrid-ssm-transformer", "context_window": 131072, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16/blob/ca9543b126e8bf3176916d3d305ccc415f89fd4d/README.md", "source_type": "primary", "sources": [ { "label": "Pinned Nemotron Nano 12B v2 VL BF16 model card", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16/blob/ca9543b126e8bf3176916d3d305ccc415f89fd4d/README.md", "type": "primary" }, { "label": "Pinned VL composite model configuration", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16/blob/ca9543b126e8bf3176916d3d305ccc415f89fd4d/config.json", "type": "primary" }, { "label": "Complete seven-shard BF16 tensor index", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16/blob/ca9543b126e8bf3176916d3d305ccc415f89fd4d/model.safetensors.index.json", "type": "primary" }, { "label": "Pinned vision-language model composition", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16/blob/ca9543b126e8bf3176916d3d305ccc415f89fd4d/modeling.py", "type": "primary" }, { "label": "Pinned language module implementation", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16/blob/ca9543b126e8bf3176916d3d305ccc415f89fd4d/modeling_nemotron_h.py", "type": "primary" }, { "label": "NVIDIA Nemotron Nano V2 VL technical report, revision 2", "url": "https://arxiv.org/html/2511.03929v2", "type": "primary" }, { "label": "October 28, 2025 NVIDIA model launch article", "url": "https://developer.nvidia.com/blog/develop-specialized-ai-agents-with-new-nvidia-nemotron-vision-rag-and-guardrail-models/", "type": "primary" }, { "label": "NVIDIA Open Model License Agreement", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/", "type": "primary" }, { "label": "Current NVIDIA Nemotron Nano 12B v2 VL BF16 model card, rechecked October 27", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16/raw/main/README.md", "type": "primary" }, { "label": "NVIDIA Build exact Nemotron Nano 12B v2 VL specifications", "url": "https://build.nvidia.com/nvidia/nemotron-nano-12b-v2-vl", "type": "primary" }, { "label": "Named Nemotron Nano 12B v2 Base language-model card", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-12B-v2-Base/raw/main/README.md", "type": "primary" }, { "label": "Nemotron 3 Nano Omni report, predecessor comparison", "url": "https://arxiv.org/html/2604.24954v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/nvidia/nemotron-nano-12b-v2-vl.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete BF16 checkpoint has 13.181860352B learned parameters, including vision encoder and connector and excluding normalization buffers. NVIDIA Build rounds this to 13B; the BF16 card instead labels it 12.6B. Separate active parameters remain undisclosed; later comparisons give only a 12B size label for this model. Supported context is 131072 input-plus-output tokens; the larger 311296-token training sequence length is not advertised as the inference context. Reasoning is supported for text/image inputs; the pinned chat template disables reasoning for video.", "parameter_components_b": { "language_backbone": 12.320486912, "vision_encoder": 0.65165312, "vision_connector": 0.20972032 }, "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T16:07:26.529771Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/392.json" }, { "id": "minimax/MiniMax-M2", "slug": "minimax--minimax-m2", "name": "MiniMax-M2", "developer": "MiniMax", "developer_id": "minimax", "family": "minimax", "description": "Open-weight text MoE model for coding and agent workflows, with interleaved reasoning and native tool calling.", "release_date": "2025-10-27", "date_precision": "day", "open_weights": true, "license": "Modified MIT", "parameters_b": 228.689764864, "active_parameters_b": 9.8, "parameters_source": "primary_source", "architecture": "moe", "context_window": 196608, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/minimax/MiniMax-M2.toml", "source_type": "catalog", "sources": [ { "label": "MiniMax-M2 official model card", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2/blob/04312f0cfbdeefb912575595b7c045512a29293c/README.md", "type": "primary" }, { "label": "MiniMax-M2 series technical report v2", "url": "https://arxiv.org/html/2605.26494v2", "type": "primary" }, { "label": "MiniMax-M2 checkpoint config", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2/blob/04312f0cfbdeefb912575595b7c045512a29293c/config.json", "type": "primary" }, { "label": "MiniMax-M2 modified MIT license", "url": "https://raw.githubusercontent.com/MiniMax-AI/MiniMax-M2/main/LICENSE", "type": "primary" }, { "label": "MiniMax-M2 release announcement", "url": "https://www.minimax.io/news/minimax-m2", "type": "primary" }, { "label": "MiniMax-M2 full-attention architecture", "url": "https://platform.minimax.io/docs/guides/text-m2-full-attention", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/minimax/MiniMax-M2.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M2", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete released checkpoint has 228.689764864B learned values, including 15,872 routing biases identified as learned in the report but registered as inference buffers; excludes 13,880,064 FP8 scale entries. Later M2 report gives 229.9B total/9.8B active and original card rounds to 230B/10B. The published-total discrepancy remains unresolved; no MTP keys are released. Context 196608 is the released-weight limit; hosted listings advertise 204800. Modified MIT adds large-commercial UI attribution. Developer final-release date is October 27, 2025; earlier provider/public-history dates do not establish exact final/preview equivalence.", "field_status": {}, "audited_at": "2026-10-26T22:20:33.488894Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/300.json" }, { "id": "paddlepaddle/paddleocr-vl", "slug": "paddlepaddle--paddleocr-vl", "name": "PaddleOCR-VL", "developer": "PaddlePaddle", "developer_id": "paddlepaddle", "family": "paddleocr-vl", "description": "Vision-language model from PaddlePaddle for recognizing text, tables, formulas and charts in document images, supporting 109 languages.", "release_date": "2025-10-16", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.958588736, "active_parameters_b": 0.9, "parameters_source": "primary_source", "architecture": "dense", "context_window": 16384, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/PaddlePaddle/PaddleOCR/releases/tag/v3.3.0", "source_type": "primary", "sources": [ { "label": "PaddleOCR v3.3.0 release, 16 October 2025", "url": "https://github.com/PaddlePaddle/PaddleOCR/releases/tag/v3.3.0", "type": "primary" }, { "label": "Original PaddleOCR-VL technical report", "url": "https://arxiv.org/html/2510.14528v1", "type": "primary" }, { "label": "Original PaddleOCR-VL model card at pinned revision", "url": "https://huggingface.co/PaddlePaddle/PaddleOCR-VL/blob/7fa00a8c55b735ba51ba49a9058f3f9c57a99a11/README.md", "type": "primary" }, { "label": "Original PaddleOCR-VL checkpoint configuration", "url": "https://huggingface.co/PaddlePaddle/PaddleOCR-VL/blob/7fa00a8c55b735ba51ba49a9058f3f9c57a99a11/config.json", "type": "primary" }, { "label": "Original PaddleOCR-VL model implementation", "url": "https://huggingface.co/PaddlePaddle/PaddleOCR-VL/blob/7fa00a8c55b735ba51ba49a9058f3f9c57a99a11/modeling_paddleocr_vl.py", "type": "primary" }, { "label": "Original PaddleOCR-VL configuration implementation", "url": "https://huggingface.co/PaddlePaddle/PaddleOCR-VL/blob/7fa00a8c55b735ba51ba49a9058f3f9c57a99a11/configuration_paddleocr_vl.py", "type": "primary" }, { "label": "Complete original PaddleOCR-VL safetensors header at pinned revision", "url": "https://huggingface.co/PaddlePaddle/PaddleOCR-VL/resolve/7fa00a8c55b735ba51ba49a9058f3f9c57a99a11/model.safetensors", "type": "weights" }, { "label": "Original checkpoint Apache License 2.0", "url": "https://huggingface.co/PaddlePaddle/PaddleOCR-VL/blob/7fa00a8c55b735ba51ba49a9058f3f9c57a99a11/LICENSE", "type": "primary" }, { "label": "PaddleOCR-VL-1.5 technical report, original-model comparison", "url": "https://arxiv.org/html/2601.21957v1", "type": "primary" }, { "label": "Original PaddleOCR-VL chat template", "url": "https://huggingface.co/PaddlePaddle/PaddleOCR-VL/blob/7fa00a8c55b735ba51ba49a9058f3f9c57a99a11/chat_template.jinja", "type": "primary" }, { "label": "PaddlePaddle FastDeploy guide for PaddleOCR-VL-0.9B", "url": "https://github.com/PaddlePaddle/FastDeploy/blob/cba185f1fe5b676bdfc8dcbf5a2fb13e0ea78e88/docs/best_practices/PaddleOCR-VL-0.9B.md", "type": "primary" }, { "label": "Original PaddleOCR-VL tokenizer configuration", "url": "https://huggingface.co/PaddlePaddle/PaddleOCR-VL/blob/7fa00a8c55b735ba51ba49a9058f3f9c57a99a11/tokenizer_config.json", "type": "primary" }, { "label": "Jina OCR v1 original model-team comparison, September 17, 2026", "url": "https://www.elastic.co/search-labs/blog/ocr-model-jina-ocr-v1-document-parsing", "type": "primary" }, { "label": "PaddleOCR-VL-1.6 developer architecture and deployment documentation", "url": "https://www.paddleocr.ai/main/en/version3.x/algorithm/PaddleOCR-VL/PaddleOCR-VL-1.6.html", "type": "primary" }, { "label": "Original PaddleOCR-VL current developer configuration, October 27 recheck", "url": "https://huggingface.co/PaddlePaddle/PaddleOCR-VL/raw/main/config.json", "type": "primary" }, { "label": "PaddleOCR-VL-1.5 developer configuration, October 27 recheck", "url": "https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.5/raw/main/config.json", "type": "primary" }, { "label": "PaddleOCR-VL-1.6 developer configuration, October 27 recheck", "url": "https://huggingface.co/PaddlePaddle/PaddleOCR-VL-1.6/raw/main/config.json", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/paddlepaddle/paddleocr-vl.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/PaddlePaddle/PaddleOCR-VL", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Original PaddleOCR-VL: 0.958588736B complete checkpoint parameters, nominally 0.9B. Active 0.9B is inferred from Jina AI/Elastic's nominal 1.6 comparison convention through PaddlePaddle's documented unchanged VLM architecture across original, 1.5 and 1.6, corroborated by identical configurations. This is not a separate PaddlePaddle active disclosure or exact per-token/phase count. Total includes stored vision pooling and packing-position parameters bypassed in normal recognition; excludes the separate layout detector. Supported context is 16,384; the 131,072 config/tokenizer value is a larger ceiling. Dedicated reasoning and native tool calling remain unverified.", "parameter_components_b": { "language_model_and_head": 0.466654208, "vision_encoder": 0.465974336, "projector": 0.025960192 }, "field_status": { "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T17:31:02.367306+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/463.json" }, { "id": "anthropic/claude-haiku-4-5-20251001", "slug": "anthropic--claude-haiku-4-5-20251001", "name": "Claude Haiku 4.5", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude-haiku", "description": "Small, fast hybrid reasoning model for coding, computer use, responsive assistants, and parallel subagents.", "release_date": "2025-10-15", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 200000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://platform.claude.com/docs/en/models/haiku-4-5/overview", "source_type": "primary", "sources": [ { "label": "Claude Haiku 4.5 model overview", "url": "https://platform.claude.com/docs/en/models/haiku-4-5/overview", "type": "docs" }, { "label": "Claude Haiku 4.5 launch announcement", "url": "https://www.anthropic.com/news/claude-haiku-4-5", "type": "primary" }, { "label": "Claude Haiku 4.5 system card", "url": "https://www-cdn.anthropic.com/7aad69bf12627d42234e01ee7c36305dc2f6a970.pdf", "type": "primary" }, { "label": "Anthropic interpretability research including Haiku 4.5", "url": "https://transformer-circuits.pub/2026/workspace/index.html", "type": "paper" }, { "label": "Anthropic Commercial Terms", "url": "https://www.anthropic.com/legal/commercial-terms", "type": "docs" }, { "label": "Claude Haiku 5.5 system card", "url": "https://assets.claude.com/e1080d6bf5ae2018ea3c2f414064be03232f5be5.pdf", "type": "model_card" }, { "label": "Claude Opus 4.5 system card", "url": "https://assets.claude.com/bf10f64990cfda0ba858290be7b8cc6317685f47.pdf", "type": "model_card" }, { "label": "GLM-5.3 and the spread of advanced cyber capabilities", "url": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities", "type": "research" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/anthropic/claude-haiku-4-5-20251001.toml", "type": "catalog" }, { "label": "Alias documentation", "url": "https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions", "type": "docs" } ], "aliases": [ "anthropic/claude-haiku-4-5" ], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Transformer architecture is supported by first-party interpretability research explicitly studying Haiku 4.5. Anthropic has not disclosed nominal total or active parameter counts, dense-versus-MoE structure, or a named, sized base. The dated API suffix precedes the public October 15, 2025 launch.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T10:27:44.390Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/098.json", "alias_sources": { "anthropic/claude-haiku-4-5": { "url": "https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions", "note": "Alias or alternate catalog name for the same released model; only the retained ID counts as a release." } } }, { "id": "alibaba/qwen3-vl-flash", "slug": "alibaba--qwen3-vl-flash", "name": "Qwen3 VL Flash (Alibaba Cloud)", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen3-VL", "description": "Alibaba Cloud hosted Qwen3-VL model for image, video and document understanding, with optional thinking mode.", "release_date": "2025-10-14", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-vl-flash.toml", "source_type": "catalog", "sources": [ { "label": "Alibaba Cloud model lifecycle", "url": "https://help.aliyun.com/zh/model-studio/newly-released-models", "type": "primary" }, { "label": "Alibaba Cloud Qwen3-VL-Flash card", "url": "https://docs.modelstudio.console.alibabacloud.com/en/model-studio/qwen3-vl-flash", "type": "primary" }, { "label": "Alibaba Cloud regional capability card", "url": "https://docs.modelstudio.console.alibabacloud.com/tc/model-studio/qwen3-vl-flash", "type": "primary" }, { "label": "Alibaba Cloud Flash alias update", "url": "https://www.alibabacloud.com/en/notice/model_studio_update_notice_for_qwen3vlflash_models_6e1?_p_lc=1", "type": "primary" }, { "label": "Qwen3-VL technical report", "url": "https://arxiv.org/pdf/2511.21631", "type": "primary" }, { "label": "Qwen3.5 later comparison and distinct Flash mapping", "url": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B", "type": "primary" }, { "label": "Qwen3.5 later vision-language comparison", "url": "https://huggingface.co/Qwen/Qwen3.5-9B", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=qwen3-vl-flash&limit=100", "label": "Official Qwen inventory search for the exact hosted model", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=Qwen3-VL&limit=1000", "label": "Official Qwen Qwen3-VL weight inventory", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-vl-flash.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Hosted rolling alias; total and active parameter counts and dense/MoE architecture remain undisclosed. The official card compares performance with Qwen3-VL-30B-A3B without identifying it as the base. Function calling is documented for Beijing but unsupported in Singapore, Frankfurt and Virginia. Beijing availability began 2025-10-14; international/global availability began 2025-10-15. The documented alias maps to the 2026-01-22 snapshot. Model-card specifications were read through the search index; current endpoint availability was not tested. Classified as closed weights: no public release of this hosted variant was found in the developer documentation and weight inventory.", "field_status": { "license": "undisclosed", "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-28T08:01:10.204833+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/044.json" }, { "id": "google/gemini-2.5-computer-use-preview-10-2025", "slug": "google--gemini-2.5-computer-use-preview-10-2025", "name": "Gemini 2.5 Computer Use Preview", "developer": "Google", "developer_id": "google", "family": "gemini-pro", "description": "Specialized Gemini 2.5 model for browser-control agents that automate UI tasks", "release_date": "2025-10-07", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 128000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-computer-use-model/", "source_type": "primary", "sources": [ { "label": "Introducing the Gemini 2.5 Computer Use model", "url": "https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-computer-use-model/", "type": "primary" }, { "label": "Gemini 2.5 Computer Use Model Card", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Computer-Use-Model-Card.pdf", "type": "primary" }, { "label": "Gemini 2.5 Pro Model Card, June 27, 2025", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Pro-Model-Card.pdf", "type": "primary" }, { "label": "Gemini 2.5 computer use model", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-computer-use-preview-10-2025", "type": "primary" }, { "label": "Gemini deprecations", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "type": "primary" }, { "label": "Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Official Computer Use reference implementation: agent.py", "url": "https://github.com/google-gemini/computer-use-preview/blob/main/agent.py", "type": "primary" }, { "label": "Gemini 2.5 technical report, version 5", "url": "https://arxiv.org/html/2507.06261v5", "type": "primary" }, { "label": "Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities", "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_v2_5_report.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemini-2.5-computer-use-preview-10-2025.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Post-trained from Gemini 2.5 Pro (06-2025), a sparse MoE Transformer. Total and active parameter counts are undisclosed for both the variant and its named base. Browser UI actions require client-side execution. Google documents Gemini API shutdown on July 28, 2026; specifications describe the last documented model.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T12:47:20.896696Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/192.json" }, { "id": "liquid/lfm2-8b-a1b", "slug": "liquid--lfm2-8b-a1b", "name": "LFM2-8B-A1B", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm", "description": null, "release_date": "2025-10-07", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 8.3, "active_parameters_b": 1.5, "parameters_source": "primary_source", "architecture": "moe", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.liquid.ai/blog/lfm2-8b-a1b-an-efficient-on-device-mixture-of-experts", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/lfm2-8b-a1b-an-efficient-on-device-mixture-of-experts", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/LiquidAI/LFM2-8B-A1B/blob/c1c44ff9fc00db3ebf4516970563f5f383d23670/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/LiquidAI/LFM2-8B-A1B/blob/c1c44ff9fc00db3ebf4516970563f5f383d23670/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:23.459Z", "last_seen": "2026-10-29T10:17:23.459Z", "upstream_present": false, "availability": "public", "notes": "Original LFM2 release, distinct from the later LFM2.5 refresh. The developer labels the MoE checkpoint as an early release; weights were publicly available on the announcement date." }, { "id": "openai/gpt-audio-mini-2025-10-06", "slug": "openai--gpt-audio-mini-2025-10-06", "name": "GPT-audio-mini-2025-10-06", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-audio", "description": null, "release_date": "2025-10-06", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 128000, "modalities": { "input": [ "text", "audio" ], "output": [ "text", "audio" ] }, "reasoning": null, "tool_call": null, "source_url": "https://developers.openai.com/api/docs/changelog", "source_type": "primary", "sources": [ { "label": "Dated publisher API releases", "url": "https://developers.openai.com/api/docs/changelog", "type": "documentation" }, { "label": "Publisher model documentation and snapshot mapping", "url": "https://developers.openai.com/api/docs/models/gpt-audio-mini", "type": "documentation" }, { "label": "Publisher shutdown record", "url": "https://developers.openai.com/api/docs/deprecations", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:30:48.692Z", "last_seen": "2026-10-29T10:30:48.692Z", "upstream_present": false, "availability": "public", "notes": "Original publisher conversational audio model. API access ended July 23, 2026." }, { "id": "openai/gpt-realtime-mini-2025-10-06", "slug": "openai--gpt-realtime-mini-2025-10-06", "name": "GPT-realtime-mini-2025-10-06", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-realtime", "description": null, "release_date": "2025-10-06", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 32000, "modalities": { "input": [ "text", "audio", "image" ], "output": [ "text", "audio" ] }, "reasoning": null, "tool_call": null, "source_url": "https://developers.openai.com/api/docs/changelog", "source_type": "primary", "sources": [ { "label": "Dated publisher API releases", "url": "https://developers.openai.com/api/docs/changelog", "type": "documentation" }, { "label": "Publisher model documentation and snapshot mapping", "url": "https://developers.openai.com/api/docs/models/gpt-realtime-mini", "type": "documentation" }, { "label": "Publisher shutdown record", "url": "https://developers.openai.com/api/docs/deprecations", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:31:17.277Z", "last_seen": "2026-10-29T10:31:17.277Z", "upstream_present": false, "availability": "public", "notes": "Original publisher conversational audio model. API access ended July 23, 2026." }, { "id": "alibaba/qwen3-vl-30b-a3b-instruct", "slug": "alibaba--qwen3-vl-30b-a3b-instruct", "name": "Qwen3 VL 30B A3B Instruct", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Open vision-language MoE model for image and video understanding, documents, and visual agents", "release_date": "2025-10-04", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 31.070754032, "active_parameters_b": 3.3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-vl-30b-a3b-instruct.toml", "source_type": "catalog", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-VL-30B-A3B-Instruct", "type": "weights" }, { "label": "Official checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen3-VL-30B-A3B-Instruct", "type": "primary" }, { "label": "Official checkpoint weight index", "url": "https://huggingface.co/Qwen/Qwen3-VL-30B-A3B-Instruct/raw/main/model.safetensors.index.json", "type": "primary" }, { "label": "Official checkpoint configuration", "url": "https://huggingface.co/Qwen/Qwen3-VL-30B-A3B-Instruct/raw/main/config.json", "type": "primary" }, { "label": "Qwen3-VL release announcement", "url": "https://github.com/QwenLM/Qwen3-VL", "type": "primary" }, { "label": "Qwen3-VL technical report", "url": "https://arxiv.org/pdf/2511.21631", "type": "paper" }, { "label": "Qwen3 language-backbone parameter disclosure", "url": "https://huggingface.co/Qwen/Qwen3-30B-A3B/raw/main/README.md", "type": "primary" }, { "label": "Model Studio deployment history", "url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "type": "primary" }, { "label": "Qwen3-30B-A3B language-backbone configuration", "url": "https://huggingface.co/Qwen/Qwen3-30B-A3B/raw/main/config.json", "type": "primary" }, { "label": "Qwen3: Think Deeper, Act Faster", "url": "https://qwenlm.github.io/blog/qwen3/", "type": "primary" }, { "label": "Qwen3-VL MoE implementation, Transformers 4.57.0", "url": "https://raw.githubusercontent.com/huggingface/transformers/v4.57.0/src/transformers/models/qwen3_vl_moe/modeling_qwen3_vl_moe.py", "type": "primary" }, { "label": "Qwen3 MoE implementation, Transformers 4.57.0", "url": "https://raw.githubusercontent.com/huggingface/transformers/v4.57.0/src/transformers/models/qwen3_moe/modeling_qwen3_moe.py", "type": "primary" }, { "label": "Qwen3-30B-A3B weight index", "url": "https://huggingface.co/Qwen/Qwen3-30B-A3B/raw/main/model.safetensors.index.json", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-vl-30b-a3b-instruct.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Full checkpoint: 31,070,754,032 parameters including vision components. Active 3.3B is an approximate language-decoder-only figure inferred from the explicitly linked Qwen3 backbone after verifying complete learned-parameter and routing equivalence; it is not a full multimodal active count. Decoder total 30,532,122,624 includes embeddings/output head. A3B is the backbone announcement's nominal 3B label. October 4 is the open-weight announcement; Model Studio lists earlier region-specific API dates.", "active_parameters_scope": "language_decoder", "parameter_components_b": { "language_decoder": 30.532122624 }, "field_status": {}, "audited_at": "2026-10-26T14:45:50Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/043.json" }, { "id": "ibm/granite-4-h-micro", "slug": "ibm--granite-4-h-micro", "name": "Granite-4.0-H-Micro", "developer": "IBM", "developer_id": "ibm", "family": "granite", "description": "IBM instruction-tuned text model combining Mamba2 and attention layers with dense feedforward blocks, supporting multilingual chat, coding, retrieval-augmented generation and tool calling.", "release_date": "2025-10-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 3.191396096, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "hybrid-ssm-transformer", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://huggingface.co/ibm-granite/granite-4.0-h-micro", "source_type": "primary", "sources": [ { "label": "IBM H-Micro model card", "url": "https://huggingface.co/ibm-granite/granite-4.0-h-micro/blob/d5f01a3ea75f088947be3aae039f4ad52837dfde/README.md", "type": "primary" }, { "label": "IBM Granite 4.0 launch", "url": "https://www.ibm.com/new/announcements/ibm-granite-4-0-hyper-efficient-high-performance-hybrid-models", "type": "primary" }, { "label": "Exact H-Micro configuration", "url": "https://huggingface.co/ibm-granite/granite-4.0-h-micro/resolve/d5f01a3ea75f088947be3aae039f4ad52837dfde/config.json", "type": "primary" }, { "label": "Original checkpoint index", "url": "https://huggingface.co/ibm-granite/granite-4.0-h-micro/resolve/d5f01a3ea75f088947be3aae039f4ad52837dfde/model.safetensors.index.json", "type": "weights" }, { "label": "Checkpoint metadata", "url": "https://huggingface.co/api/models/ibm-granite/granite-4.0-h-micro?blobs=true", "type": "primary" }, { "label": "Granite hybrid implementation", "url": "https://github.com/huggingface/transformers/blob/v4.56.0/src/transformers/models/granitemoehybrid/modeling_granitemoehybrid.py", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/ibm/granite-4-h-micro.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/ibm-granite/granite-4.0-h-micro", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Total is the exact 3,191,396,096 learned-parameter count from the original BF16 checkpoint headers, including tied embeddings once. IBM labels this model 3B total and 3B active; active_parameters_b preserves that nominal published figure, not an exact per-token or phase-specific count. Dense MLPs have no routed experts. Configured and validated context is 131,072 tokens; larger training samples do not change that supported limit.", "parameter_components_b": { "tied_embeddings": 0.205520896, "dense_mlp": 2.01326592, "mamba2_mixers": 0.930500352, "attention_projections": 0.04194304, "other_norms": 0.000165888 }, "field_status": {}, "audited_at": "2026-10-26T20:47:32.388410Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/235.json" }, { "id": "ibm/granite-4-h-small", "slug": "ibm--granite-4-h-small", "name": "Granite-4.0-H-Small", "developer": "IBM", "developer_id": "ibm", "family": "granite", "description": "Open-weight hybrid model for enterprise chat, coding, retrieval-augmented generation, and tool-calling workloads", "release_date": "2025-10-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 32.207337984, "active_parameters_b": 9, "parameters_source": "primary_source", "architecture": "Hybrid Mamba-2/Transformer MoE", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://huggingface.co/ibm-granite/granite-4.0-h-small", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/ibm-granite/granite-4.0-h-small", "type": "weights" }, { "label": "IBM Granite 4.0 launch announcement", "url": "https://www.ibm.com/new/announcements/ibm-granite-4-0-hyper-efficient-high-performance-hybrid-models", "type": "primary" }, { "label": "IBM original checkpoint configuration", "url": "https://huggingface.co/ibm-granite/granite-4.0-h-small/blob/b8c0982bab7fde4eb48110f5a069527c008fab39/config.json", "type": "primary" }, { "label": "IBM BF16 checkpoint index and tensors", "url": "https://huggingface.co/ibm-granite/granite-4.0-h-small/blob/b8c0982bab7fde4eb48110f5a069527c008fab39/model.safetensors.index.json", "type": "weights" }, { "label": "GraniteMoeHybrid parameter and routing implementation", "url": "https://github.com/huggingface/transformers/blob/v4.56.0/src/transformers/models/granitemoehybrid/modeling_granitemoehybrid.py", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/ibm/granite-4-h-small.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "IBM nominally reports 32B total and 9B active. Original BF16 checkpoint has 32,207,337,984 unique learned parameters with tied embeddings counted once; header/config accounting yields 8,803,121,664 active per generated token including full output projection, all shared experts, Mamba/attention, routers, and norms. The architecture is hybrid Mamba-2/Transformer MoE (36/4 layers, 10 of 72 routed experts per layer). This is the instruct variant without a dedicated Thinking mode.", "parameter_components_b": { "routed_experts": 27.17908992, "shared_experts": 0.75497472, "mamba": 3.682331136, "attention": 0.16777216, "tied_embedding_output": 0.411041792, "routers": 0.01179648, "other_norms": 0.000331776 }, "field_status": {}, "audited_at": "2026-10-26T20:47:29.819274Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/236.json" }, { "id": "zhipuai/glm-4.6", "slug": "zhipuai--glm-4.6", "name": "GLM-4.6", "developer": "Z.ai", "developer_id": "zhipuai", "family": "glm", "description": "Text MoE model for reasoning, coding and tool use, with a 200K-token context window.", "release_date": "2025-09-30", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 355, "active_parameters_b": 32, "parameters_source": "primary_source", "architecture": "moe", "context_window": 202752, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://z.ai/blog/glm-4.6", "source_type": "primary", "sources": [ { "label": "GLM-4.6 original release announcement", "url": "https://z.ai/blog/glm-4.6", "type": "primary" }, { "label": "GLM-4.6 exact checkpoint card", "url": "https://huggingface.co/zai-org/GLM-4.6/blob/be72194883d968d7923a07e2f61681ea9a2826d1/README.md", "type": "weights" }, { "label": "Official model download table at pinned repository revision", "url": "https://github.com/zai-org/GLM-4.5/blob/170f20b2c10659008fdbc909d478bc2a75bc3627/README.md", "type": "primary" }, { "label": "GLM-4.6 exact model configuration", "url": "https://huggingface.co/zai-org/GLM-4.6/blob/be72194883d968d7923a07e2f61681ea9a2826d1/config.json", "type": "weights" }, { "label": "GLM-4.6 exact chat template", "url": "https://huggingface.co/zai-org/GLM-4.6/blob/be72194883d968d7923a07e2f61681ea9a2826d1/chat_template.jinja", "type": "weights" }, { "label": "GLM-4.6 native API overview", "url": "https://docs.z.ai/guides/llm/glm-4.6", "type": "primary" }, { "label": "GLM-4.6 provider endpoint metadata", "url": "https://openrouter.ai/api/v1/models/z-ai/glm-4.6/endpoints", "type": "primary" }, { "label": "Exact public checkpoint repository metadata", "url": "https://huggingface.co/api/models/zai-org/GLM-4.6", "type": "weights" }, { "label": "GLM-4.6 AI Gateway provider specifications", "url": "https://vercel.com/ai-gateway/models/glm-4.6", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/zhipuai/glm-4.6.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/GLM-4.6", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Z.ai reports nominal 355B total and 32B active for GLM-4.6 in its exact model download row. These are not exact tensor counts; the table does not resolve embedding, output-head or MTP inclusion, so the GLM-4.5 counting convention is not silently transferred. The released checkpoint and Z.AI provider route specify 202752 tokens, while some other routes advertise 204800; native documentation rounds this to 200K. Maximum output is a separate limit. This is the text-only GLM-4.6, not GLM-4.6V.", "field_status": {}, "audited_at": "2026-10-27T06:04:47.660678Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/551.json" }, { "id": "anthropic/claude-sonnet-4-5-20250929", "slug": "anthropic--claude-sonnet-4-5-20250929", "name": "Claude Sonnet 4.5", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude-sonnet", "description": "Claude Sonnet model for coding, agentic tasks and computer use, with optional extended thinking.", "release_date": "2025-09-29", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 200000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.anthropic.com/news/claude-sonnet-4-5", "source_type": "primary", "sources": [ { "label": "Introducing Claude Sonnet 4.5", "url": "https://www.anthropic.com/news/claude-sonnet-4-5", "type": "primary" }, { "label": "Claude Sonnet 4.5 - Claude Platform Docs", "url": "https://platform.claude.com/docs/en/models/sonnet-4-5/overview", "type": "primary" }, { "label": "Verbalizable Representations Form a Global Workspace in Language Models", "url": "https://transformer-circuits.pub/2026/workspace/index.html", "type": "primary" }, { "label": "GLM-5.3 and the spread of advanced cyber capabilities", "url": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities", "type": "primary" }, { "label": "Commercial Terms of Service", "url": "https://www.anthropic.com/legal/commercial-terms", "type": "primary" }, { "label": "PDF support", "url": "https://platform.claude.com/docs/en/build-with-claude/pdf-support", "type": "primary" }, { "label": "Tool use with Claude", "url": "https://platform.claude.com/docs/en/agents-and-tools/tool-use/overview", "type": "primary" }, { "label": "Claude Sonnet 4.5 System Card", "url": "https://www.anthropic.com/claude-sonnet-4-5-system-card", "type": "primary" }, { "label": "Claude Sonnet 4.6 System Card", "url": "https://www.anthropic.com/claude-sonnet-4-6-system-card", "type": "primary" }, { "label": "Claude Platform release notes", "url": "https://platform.claude.com/docs/en/release-notes/overview", "type": "primary" }, { "label": "Anthropic's Claude Sonnet 4.5 on Google Cloud", "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/partner-models/claude/sonnet-4-5", "type": "primary" }, { "label": "Claude Sonnet 4.5 - Amazon Bedrock", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-anthropic-claude-sonnet-4-5.html", "type": "primary" }, { "label": "Claude Sonnet 5 System Card", "url": "https://www.anthropic.com/claude-sonnet-5-system-card", "type": "primary" }, { "label": "Claude Sonnet 5.5 System Card", "url": "https://www.anthropic.com/claude-sonnet-5-5-system-card", "type": "primary" }, { "label": "Claude Opus 5.5 System Card", "url": "https://www.anthropic.com/claude-opus-5-5-system-card", "type": "primary" }, { "label": "Claude Haiku 5.5 System Card", "url": "https://www.anthropic.com/claude-haiku-5-5-system-card", "type": "primary" }, { "label": "Anthropic Transparency Hub - Claude Sonnet 4.5 summary", "url": "https://www.anthropic.com/transparency", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/anthropic/claude-sonnet-4-5-20250929.toml", "type": "catalog" }, { "label": "Alias documentation", "url": "https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions", "type": "docs" } ], "aliases": [ "anthropic/claude-sonnet-4-5" ], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Transformer architecture is supported by exact-model interpretability experiments; dense/MoE subtype and nominal total/active parameter counts remain undisclosed. Third-party size estimates do not establish specifications. Current Anthropic API context is 200K after the 1M beta retired April 30, 2026; Google Cloud still lists a 1M preview. PDF input denotes text/vision document processing. Anthropic release was September 29, 2025; Bedrock lists September 30. Claude API deprecation was September 30, 2026 with retirement scheduled November 30, 2026; provider lifecycle dates differ.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T10:47:08.453407+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/113.json", "alias_sources": { "anthropic/claude-sonnet-4-5": { "url": "https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions", "note": "Alias or alternate catalog name for the same released model; only the retained ID counts as a release." } } }, { "id": "google/gemini-2.5-flash-lite-preview-09-2025", "slug": "google--gemini-2.5-flash-lite-preview-09-2025", "name": "Gemini 2.5 Flash-Lite Preview 09-2025", "developer": "Google", "developer_id": "google", "family": "gemini-flash-lite", "description": "September 2025 preview of Gemini 2.5 Flash-Lite with multimodal input, optional thinking and function calling.", "release_date": "2025-09-25", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "audio", "video", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://developers.googleblog.com/continuing-to-bring-you-our-latest-models-with-an-improved-gemini-2-5-flash-and-flash-lite-release/", "source_type": "primary", "sources": [ { "label": "Continuing to bring you our latest models, with an improved Gemini 2.5 Flash and Flash-Lite release", "url": "https://developers.googleblog.com/continuing-to-bring-you-our-latest-models-with-an-improved-gemini-2-5-flash-and-flash-lite-release/", "type": "primary" }, { "label": "Gemini 2.5 Flash-Lite preview — Gemini API", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-flash-lite-preview-09-2025", "type": "primary" }, { "label": "Gemini 2.5 Flash-Lite Model Card — September 26, 2025", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Flash-Lite-Model-Card.pdf", "type": "primary" }, { "label": "Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Gemini deprecations — Gemini API", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "type": "primary" }, { "label": "Gemini 3.1 Flash-Lite Model Card — March 2026", "url": "https://deepmind.google/models/model-cards/gemini-3-1-flash-lite/", "type": "primary" }, { "label": "Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities", "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_v2_5_report.pdf", "type": "primary" }, { "label": "Gemini 3.5 Flash-Lite Model Card", "url": "https://deepmind.google/models/model-cards/gemini-3-5-flash-lite/", "type": "primary" }, { "label": "Gemini 3.8 Flash Model Card", "url": "https://deepmind.google/models/model-cards/gemini-3-8-flash/", "type": "primary" }, { "label": "Gemini 4 Argon Model Card", "url": "https://deepmind.google/models/model-cards/gemini-4-argon/", "type": "primary" }, { "label": "Gemini 4 Argon Safety and Security Report", "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemini-2.5-flash-lite-preview-09-2025.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Exact September 2025 preview, distinct from the June preview and July stable model. Google records its shutdown on March 31, 2026. Sparse MoE architecture is documented; total and active parameter counts remain undisclosed.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T12:53:24.407471+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/196.json" }, { "id": "relace/relace-apply-3", "slug": "relace--relace-apply-3", "name": "Relace Apply 3", "developer": "Relace", "developer_id": "relace", "family": "relace-apply", "description": "Specialized code-editing model that merges an edit snippet into an existing source file and returns the updated code.", "release_date": "2025-09-25", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 256000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": false, "source_url": "https://x.com/relace_ai/status/1971265369466974308", "source_type": "primary", "sources": [ { "label": "Apply 3 training details", "url": "https://relace.ai/old/blog/relace-apply-3", "type": "primary" }, { "label": "Apply API reference", "url": "https://docs.relace.ai/api-reference/instant-apply/apply.md", "type": "primary" }, { "label": "Agent tool integration", "url": "https://docs.relace.ai/docs/instant-apply/agent.md", "type": "primary" }, { "label": "Relace service terms", "url": "https://relace.ai/legal/terms-of-use", "type": "primary" }, { "label": "OpenRouter endpoint metadata", "url": "https://openrouter.ai/api/v1/models/relace/relace-apply-3/endpoints", "type": "catalog" }, { "label": "Relace Apply 3 original launch announcement, September 25, 2025", "url": "https://x.com/relace_ai/status/1971265369466974308", "type": "primary" }, { "label": "Relace announces subsequent OpenRouter availability", "url": "https://x.com/relace_ai/status/1971624869377999269", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/relace/relace-apply-3.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Relace Apply 3 merges code edits; OpenAI-compatible transport does not make it a general-purpose chat model. The documented total context is 256k, with 128k input and 128k output API limits. Total and active parameters remain undisclosed: Relace gives only a 3–8B range for Apply-series backbones, without naming Apply 3’s base. Architecture is also undisclosed. The developer launched Apply 3 on September 25, 2025, followed by OpenRouter availability on September 26. October 29 is the later training retrospective. Apply can be invoked by an agent’s edit tool, but its own API returns merged code rather than tool calls. A model-level reasoning declaration was not found.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed", "reasoning": "unverified" }, "audited_at": "2026-10-27T19:30:59.672914Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/480.json" }, { "id": "alibaba/qwen3-max", "slug": "alibaba--qwen3-max", "name": "Qwen3 Max", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Hosted Qwen3 text model for coding, reasoning, and tool-using agents.", "release_date": "2025-09-23", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-max.toml", "source_type": "catalog", "sources": [ { "label": "Qwen3-Max official launch", "url": "https://qwen.ai/blog?from=research.latest-advancements-list&id=241398b9cd6353de490b0f82806c7848c5d2777d", "type": "primary" }, { "label": "Alibaba Cloud release log", "url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "type": "primary" }, { "label": "Alibaba Cloud alias mapping and modes", "url": "https://www.alibabacloud.com/help/en/model-studio/model-pricing", "type": "primary" }, { "label": "Qwen3-Max-Thinking announcement", "url": "https://qwen.ai/blog?id=qwen3-max-thinking", "type": "primary" }, { "label": "Official exact model and snapshot specifications", "url": "https://docs.modelstudio.console.alibabacloud.com/zh/model-studio/model-qwen3-max", "type": "primary" }, { "label": "Alibaba Group Qwen3-Max published scale", "url": "https://home.alibabagroup.com/en-US/document-1911884625546838016", "type": "primary" }, { "label": "Alibaba Qwen3-Max-Thinking parameter lower bound", "url": "https://www.alizila.com/alibabas-latest-thinking-model-excels-at-adaptive-tool-use/", "type": "primary" }, { "label": "Qwen3.5 comparison with Qwen3-Max-Base", "url": "https://qwen.ai/blog?id=qwen3.5", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-max.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "First documented China availability is September 23, 2025; global rollout and launch post followed September 24. The alias maps to the January 23, 2026 Thinking snapshot. Alibaba discloses over 1,000B total parameters for both the original base and this Thinking model; this is a strict lower bound, not a 1,000B point estimate. Active parameters are undisclosed. Exact context is 262144 in the indexed model card.", "field_status": { "license": "not_applicable", "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T09:37:40.807853Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/036.json", "parameters_lower_bound_b": 1000, "parameters_lower_bound_exclusive": true }, { "id": "alibaba/qwen3-vl-235b-a22b-instruct", "slug": "alibaba--qwen3-vl-235b-a22b-instruct", "name": "Qwen3 VL 235B A22B Instruct", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Qwen vision-language instruct model for visual reasoning, documents, and agent tasks", "release_date": "2025-09-23", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 235.670022896, "active_parameters_b": 22, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-vl-235b-a22b-instruct.toml", "source_type": "catalog", "sources": [ { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-vl-235b-a22b-instruct.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-VL-235B-A22B-Instruct", "type": "weights" }, { "label": "Official checkpoint configuration", "url": "https://huggingface.co/Qwen/Qwen3-VL-235B-A22B-Instruct/raw/main/config.json", "type": "primary" }, { "label": "Official release announcement", "url": "https://github.com/QwenLM/Qwen3-VL", "type": "primary" }, { "label": "Official checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen3-VL-235B-A22B-Instruct", "type": "primary" }, { "label": "Qwen3-VL technical report", "url": "https://arxiv.org/pdf/2511.21631", "type": "primary" }, { "label": "Official tool-calling template", "url": "https://huggingface.co/Qwen/Qwen3-VL-235B-A22B-Instruct/raw/main/chat_template.json", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Full checkpoint metadata reports 235.670022896B stored parameters including vision and mergers. The 235B-A22B designation and reported 22B active-per-token figure describe nominal language-backbone scale; 22B does not establish a complete multimodal activation count. Instruct is the non-thinking edition.", "active_parameters_scope": "language_decoder", "field_status": {}, "audited_at": "2026-10-26T14:39:01.571680+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/041.json" }, { "id": "alibaba/qwen3-vl-235b-a22b-thinking", "slug": "alibaba--qwen3-vl-235b-a22b-thinking", "name": "Qwen3 VL 235B A22B Thinking", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Qwen vision-language thinking model for visual reasoning, documents, and agent tasks", "release_date": "2025-09-23", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 235.670022896, "active_parameters_b": 22, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-vl-235b-a22b-thinking.toml", "source_type": "catalog", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-VL-235B-A22B-Thinking", "type": "weights" }, { "label": "Qwen checkpoint configuration", "url": "https://huggingface.co/Qwen/Qwen3-VL-235B-A22B-Thinking/raw/main/config.json", "type": "primary" }, { "label": "Qwen3-VL official release announcement", "url": "https://github.com/QwenLM/Qwen3-VL", "type": "primary" }, { "label": "Qwen3-VL Technical Report", "url": "https://arxiv.org/pdf/2511.21631", "type": "primary" }, { "label": "Full checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen3-VL-235B-A22B-Thinking", "type": "primary" }, { "label": "Qwen checkpoint chat template", "url": "https://huggingface.co/Qwen/Qwen3-VL-235B-A22B-Thinking/raw/main/chat_template.json", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-vl-235b-a22b-thinking.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Full multimodal checkpoint contains 235.670022896B stored parameters according to its safetensors metadata. The nominal 235B/22B figures are presented in the technical report language-model subsection; 22B denotes published per-token language-model activation, not a separately measured full-multimodal active count. Native context is 262144 tokens; optional extension to 1M requires configuration changes.", "active_parameters_scope": "language_decoder", "field_status": {}, "audited_at": "2026-10-26T14:38:51.932351Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/042.json" }, { "id": "alibaba/qwen3-vl-plus", "slug": "alibaba--qwen3-vl-plus", "name": "Qwen3-VL Plus", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Alibaba Cloud Qwen3 vision-language API model for visual reasoning, spatial understanding, video understanding and visual agent tasks, with optional thinking.", "release_date": "2025-09-23", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 262144, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://docs.modelstudio.console.alibabacloud.com/en/model-studio/qwen3-vl-plus", "source_type": "primary", "sources": [ { "label": "Alibaba Cloud exact model card", "url": "https://docs.modelstudio.console.alibabacloud.com/en/model-studio/qwen3-vl-plus", "type": "primary" }, { "label": "Alibaba Cloud model releases", "url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "type": "primary" }, { "label": "Alibaba Cloud alias mapping", "url": "https://docs.modelstudio.console.alibabacloud.com/en/model-studio/model-pricing", "type": "primary" }, { "label": "Qwen3-VL official repository", "url": "https://github.com/QwenLM/Qwen3-VL", "type": "primary" }, { "label": "Qwen API platform", "url": "https://qwen.ai/apiplatform", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-vl-plus.toml", "type": "catalog" }, { "label": "qwen3-vl-plus Model Info, current public documentation", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen3-vl-plus", "type": "primary" }, { "label": "Qwen3-VL Technical Report, Model Architecture and comparison tables", "url": "https://arxiv.org/html/2511.21631v1", "type": "primary" }, { "label": "Qwen3.5-397B-A17B official model card and later multimodal comparison table", "url": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B", "type": "primary" }, { "label": "Alibaba Cloud Model Studio model pricing, fresh public documentation", "url": "https://www.alibabacloud.com/help/en/model-studio/model-pricing", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=qwen3-vl-plus&limit=100", "label": "Official Qwen inventory search for the exact hosted model", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=Qwen3-VL&limit=1000", "label": "Official Qwen Qwen3-VL weight inventory", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Total/active parameter counts and architecture are undisclosed; no size-specific base is identified for Plus. Published Qwen3-VL-235B-A22B counts cannot be assigned to this alias. Function calling is documented for Beijing but unsupported in the other listed regional deployments. Context follows the Alibaba Cloud card; Qwen's platform lists a conflicting 131027. No public Plus weight release is available. Classified as closed weights: no public release of this hosted variant was found in the developer documentation and weight inventory.", "field_status": { "license": "undisclosed", "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-28T08:01:10.205319+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/045.json" }, { "id": "liquid/lfm2-2.6b", "slug": "liquid--lfm2-2.6b", "name": "LFM2-2.6B", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm", "description": null, "release_date": "2025-09-23", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 2.56927232, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "hybrid", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.liquid.ai/blog/introducing-lfm2-2-6b-redefining-efficiency-in-language-models", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/introducing-lfm2-2-6b-redefining-efficiency-in-language-models", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/LiquidAI/LFM2-2.6B/blob/36ed799f7024ef169ba92ffe184821835e15cc66/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/LiquidAI/LFM2-2.6B/blob/36ed799f7024ef169ba92ffe184821835e15cc66/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:23.459Z", "last_seen": "2026-10-29T10:17:23.459Z", "upstream_present": false, "availability": "public", "notes": "Original LFM2 release, distinct from the later LFM2.5 refresh. Exact total parameters and supported context follow the model card." }, { "id": "alibaba/qwen3-omni-30b-a3b-thinking", "slug": "alibaba--qwen3-omni-30b-a3b-thinking", "name": "Qwen3 Omni 30B A3B Thinking", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Multimodal reasoning model accepting text, images, video, and audio with text output", "release_date": "2025-09-22", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 31.719205488, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 65536, "modalities": { "input": [ "text", "image", "video", "audio" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-omni-30b-a3b-thinking.toml", "source_type": "catalog", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-Omni-30B-A3B-Thinking", "type": "weights" }, { "label": "Thinking checkpoint parameter metadata", "url": "https://huggingface.co/Qwen/Qwen3-Omni-30B-A3B-Thinking/raw/main/model.safetensors.index.json", "type": "primary" }, { "label": "Thinking configuration", "url": "https://huggingface.co/Qwen/Qwen3-Omni-30B-A3B-Thinking/raw/main/config.json", "type": "primary" }, { "label": "Qwen3-Omni technical report", "url": "https://arxiv.org/html/2509.17765v1", "type": "paper" }, { "label": "Official release announcement", "url": "https://github.com/QwenLM/Qwen3-Omni", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-omni-30b-a3b-thinking.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The Thinking checkpoint has 31,719,205,488 parameters, including audio and vision encoders plus its MoE text decoder. The nominal 30B total and 3B active labels describe the Thinker decoder; 3B is not a complete multimodal active count. Thinking contains no Talker, MTP, or Code2Wav speech-generation components.", "active_parameters_scope": "thinker_decoder", "parameter_components_b": { "thinker_decoder": 30, "audio_encoder": 0.65, "vision_encoder": 0.54 }, "field_status": {}, "audited_at": "2026-10-26T14:38:15.479318Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/040.json" }, { "id": "xai/grok-4-fast-non-reasoning", "slug": "xai--grok-4-fast-non-reasoning", "name": "Grok 4 Fast (Non-Reasoning)", "developer": "xAI", "developer_id": "xai", "family": "grok", "description": "Non-reasoning mode of Grok 4 Fast, accepting text and images with function calling and a two-million-token context window.", "release_date": "2025-09-19", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": null, "context_window": 2000000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://x.ai/news/grok-4-fast", "source_type": "primary", "sources": [ { "label": "Grok 4 Fast launch, September 19, 2025", "url": "https://x.ai/news/grok-4-fast", "type": "primary" }, { "label": "Grok 4 Fast Model Card", "url": "https://data.x.ai/2025-09-19-grok-4-fast-model-card.pdf", "type": "primary" }, { "label": "xAI Grok 4 Fast provider documentation", "url": "https://docs.oracle.com/en-us/iaas/Content/generative-ai/xai-grok-4-fast.htm", "type": "provider" }, { "label": "Grok model retirement on May 15, 2026", "url": "https://docs.x.ai/developers/migration/may-15-retirement", "type": "primary" }, { "label": "Former exact model documentation URL", "url": "https://docs.x.ai/developers/models/grok-4-fast-non-reasoning", "type": "primary" }, { "label": "Enterprise Terms of Service, June 27, 2025", "url": "https://x.ai/legal/terms-of-service-enterprise/previous-2025-06-27", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/xai/grok-4-fast-non-reasoning.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released September 19, 2025. Reasoning and non-reasoning modes use the same Grok 4 Fast weights, steered by system prompts. Total and active parameter counts remain undisclosed; the stated learnings from Grok 4 do not establish unchanged model size. Neural network topology remains undisclosed. Historical API specifications support text/image input and text output. xAI retired the original endpoint on May 15, 2026 and redirects this slug to Grok 4.3 with no reasoning; replacement specifications are not assigned to the original Fast release.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T18:23:03.499091Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/525.json" }, { "id": "xai/grok-4-fast-reasoning", "slug": "xai--grok-4-fast-reasoning", "name": "Grok 4 Fast (Reasoning)", "developer": "xAI", "developer_id": "xai", "family": "grok", "description": "Reasoning mode of Grok 4 Fast, with text and image input, text output, a two-million-token context window and function calling.", "release_date": "2025-09-19", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 2000000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://x.ai/news/grok-4-fast", "source_type": "primary", "sources": [ { "label": "Grok 4 Fast launch, September 19, 2025", "url": "https://x.ai/news/grok-4-fast", "type": "primary" }, { "label": "Grok 4 Fast model card", "url": "https://data.x.ai/2025-09-19-grok-4-fast-model-card.pdf", "type": "primary" }, { "label": "xAI Grok 4 Fast archived provider specifications", "url": "https://docs.oracle.com/en-us/iaas/Content/generative-ai/xai-grok-4-fast.htm", "type": "provider" }, { "label": "May 15, 2026 model retirement", "url": "https://docs.x.ai/developers/migration/may-15-retirement", "type": "primary" }, { "label": "Enterprise terms covering the xAI API, June 27, 2025", "url": "https://x.ai/legal/terms-of-service-enterprise/previous-2025-06-27", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/xai/grok-4-fast-reasoning.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released September 19, 2025. Reasoning and non-reasoning modes share weights and differ through system prompting. Total and active parameter counts and neural network architecture remain undisclosed; xAI does not identify Fast as an unchanged Grok 4 base. Historical specifications are text/image input, text output and 2,000,000 context tokens. Since May 15, 2026, the original reasoning slug redirects to Grok 4.3 with low reasoning effort; this replacement does not revise original Fast specifications.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T18:24:17.851461+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/526.json" }, { "id": "mistral/magistral-small-2509", "slug": "mistral--magistral-small-2509", "name": "Magistral Small 1.2", "developer": "Mistral", "developer_id": "mistral", "family": "magistral", "description": "Multimodal reasoning model for text and image inputs, with native tool calling.", "release_date": "2025-09-17", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 24, "active_parameters_b": 24, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://docs.mistral.ai/resources/changelogs", "source_type": "primary", "sources": [ { "label": "Mistral official API changelog", "url": "https://docs.mistral.ai/resources/changelogs", "type": "primary" }, { "label": "Magistral Small 1.2 model reference and Weights table", "url": "https://docs.mistral.ai/models/magistral-small-1-2-25-09", "type": "primary" }, { "label": "Magistral Small 1.2 pinned model card", "url": "https://huggingface.co/mistralai/Magistral-Small-2509/blob/a31cc96ab10cf19bc42c628fedf1e359e0853c49/README.md", "type": "weights" }, { "label": "Magistral Small 2509 pinned Transformers configuration", "url": "https://huggingface.co/mistralai/Magistral-Small-2509/blob/a31cc96ab10cf19bc42c628fedf1e359e0853c49/config.json", "type": "weights" }, { "label": "Magistral Small 2509 pinned native configuration", "url": "https://huggingface.co/mistralai/Magistral-Small-2509/blob/a31cc96ab10cf19bc42c628fedf1e359e0853c49/params.json", "type": "weights" }, { "label": "Magistral Small 2509 pinned tensor index", "url": "https://huggingface.co/mistralai/Magistral-Small-2509/blob/a31cc96ab10cf19bc42c628fedf1e359e0853c49/model.safetensors.index.json", "type": "weights" }, { "label": "Magistral Small 2509 public launch-day model-card contribution", "url": "https://huggingface.co/mistralai/Magistral-Small-2509/discussions/1", "type": "primary" }, { "label": "Magistral Small 1 legal lifecycle record", "url": "https://legal.mistral.ai/ai-governance/models/magistral-small-1/", "type": "primary" }, { "label": "Magistral Small 1.2 official Markdown reference", "url": "https://docs.mistral.ai/models/magistral-small-1-2-25-09.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/magistral-small-2509.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Magistral-Small-2509", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Mistral explicitly reports nominal 24B total and 24B active parameters. Native text/image input and text output with reasoning and tool calls. The checkpoint context is 131,072 tokens; the card warns quality may degrade past 40k but recommends keeping 128k. Release date follows the September 17, 2025 API changelog and same-day public maintainer activity; model-reference and legal pages give September 18. API docs record deprecation on April 30, 2026 and retirement on July 31, 2026; public weights remain downloadable.", "field_status": {}, "audited_at": "2026-10-26T22:54:56.569092Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/319.json" }, { "id": "openai/gpt-5-codex", "slug": "openai--gpt-5-codex", "name": "GPT-5-Codex", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5", "description": "GPT-5 variant trained for software engineering, including code changes, tests and reviews, with text and image input and function calling.", "release_date": "2025-09-15", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 400000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-upgrades-to-codex/", "source_type": "primary", "sources": [ { "label": "Introducing upgrades to Codex", "url": "https://openai.com/index/introducing-upgrades-to-codex/", "type": "primary" }, { "label": "GPT-5-Codex model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5-codex.md", "type": "primary" }, { "label": "Addendum to GPT-5 system card: GPT-5-Codex", "url": "https://cdn.openai.com/pdf/97cc5669-7a25-4e63-b15f-5fd5bdc4d149/gpt-5-codex-system-card.pdf", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201", "type": "primary" }, { "label": "API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "GPT-5.1-Codex-Max System Card", "url": "https://cdn.openai.com/pdf/2a7d98b1-57e5-4147-8d0e-683894d782ae/5p1_codex_max_card_03.pdf", "type": "primary" }, { "label": "Addendum to GPT-5.2 System Card: GPT-5.2-Codex", "url": "https://cdn.openai.com/pdf/ac7c37ae-7f4c-4442-b741-2eabdeaf77e0/oai_5_2_Codex.pdf", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "GPT-5 System Card", "url": "https://cdn.openai.com/gpt-5-system-card.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5-codex.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "First released in Codex on September 15, 2025; Responses API access followed September 23. The gpt-5-codex name referred to an updateable model, not a frozen dated snapshot. Listed context and capabilities are its last-documented specifications. OpenAI lists API shutdown on July 23, 2026; no live endpoint request was made. PDF denotes API file input processed into text and page images, supported by combining the exact vision/Responses interface with the documented PDF rule. Broad Transformer attention is inferred from exact prompt-caching support and the documented mechanism. Complete structure and dense/MoE routing remain undisclosed. Nominal total and active counts are undisclosed; the explicit GPT-5 coding-variant lineage supplies no number because the named base has no published count.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T16:33:41.366449Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/414.json" }, { "id": "alibaba/qwen3-max-preview", "slug": "alibaba--qwen3-max-preview", "name": "Qwen3 Max Preview", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Hosted preview of Alibaba's Qwen3-Max model, announced with more than one trillion parameters for text generation, instruction following, and agentic tasks.", "release_date": "2025-09-05", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": false, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-max-preview.toml", "source_type": "catalog", "sources": [ { "label": "Qwen exact-preview announcement", "url": "https://www.linkedin.com/posts/qwen_big-news-introducing-qwen3-max-preview-activity-7369757496581763073-SYEF", "type": "primary" }, { "label": "Alibaba Cloud model release history", "url": "https://help.aliyun.com/en/model-studio/newly-released-models", "type": "primary" }, { "label": "Alibaba Cloud current preview mode pricing", "url": "https://www.alibabacloud.com/help/en/model-studio/model-pricing", "type": "primary" }, { "label": "Official exact model and snapshot specifications", "url": "https://docs.modelstudio.console.alibabacloud.com/zh/model-studio/model-qwen3-max", "type": "primary" }, { "label": "Qwen3-Max named preview lineage and MoE base", "url": "https://qwen.ai/blog?from=research.latest-advancements-list&id=241398b9cd6353de490b0f82806c7848c5d2777d", "type": "primary" }, { "label": "Qwen3.5 later Max parameter comparison", "url": "https://qwen.ai/blog?id=qwen3.5", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=qwen3-max-preview&limit=100", "label": "Official Qwen inventory search for the exact hosted model", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=Qwen3.8&limit=1000", "label": "Official Qwen Qwen3.8 weight inventory", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=Qwen3-Max&limit=1000", "label": "Official Qwen3-Max family weight inventory", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-max-preview.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Qwen disclosed more than 1 trillion total parameters (>1000B) for Preview Instruct; no nominal point total or active count is disclosed. MoE is inferred from the developer’s explicit Qwen3-Max-Instruct preview lineage and Max-Base architecture. The indexed preview card documents 262144 context, thinking/non-thinking support and no native Function Calling. No public Preview weight release is available; later Max generations are separate. Classified as closed weights: no public release of this hosted variant was found in the developer documentation and weight inventory.", "field_status": { "license": "undisclosed", "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-28T08:13:49.514266+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/037.json", "parameters_lower_bound_b": 1000, "parameters_lower_bound_exclusive": true }, { "id": "moonshotai/kimi-k2-0905", "slug": "moonshotai--kimi-k2-0905", "name": "Kimi K2 Instruct 0905", "developer": "Moonshot AI", "developer_id": "moonshotai", "family": "kimi-k2", "description": "September 2025 Kimi K2 instruction update with a 256K context and improved agentic coding", "release_date": "2025-09-05", "date_precision": "day", "open_weights": true, "license": "Modified MIT", "parameters_b": 1000, "active_parameters_b": 32, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct-0905", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct-0905", "type": "weights" }, { "label": "Kimi-K2-Instruct-0905 configuration", "url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct-0905/blob/ac6c49f04883bd0a0598b790693a72061c676629/config.json", "type": "primary" }, { "label": "Kimi-K2-Instruct-0905 Modified MIT License", "url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct-0905/blob/ac6c49f04883bd0a0598b790693a72061c676629/LICENSE", "type": "primary" }, { "label": "Kimi K2 model update: stronger coding and faster API", "url": "https://platform.kimi.com/blog/posts/kimi-k2-0905", "type": "primary" }, { "label": "Kimi-K2-Instruct-0905 tool-calling guide", "url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct-0905/blob/ac6c49f04883bd0a0598b790693a72061c676629/docs/tool_call_guidance.md", "type": "primary" }, { "label": "Kimi-K2-Instruct-0905 causal transformer implementation", "url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct-0905/blob/ac6c49f04883bd0a0598b790693a72061c676629/modeling_deepseek.py", "type": "primary" }, { "label": "Kimi K2: Open Agentic Intelligence, July 2025 report", "url": "https://arxiv.org/html/2507.20534v1", "type": "primary" }, { "label": "Kimi K2 Thinking model card and K2 0905 comparisons", "url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking", "type": "primary" }, { "label": "Kimi Research Blog release index", "url": "https://www.kimi.com/en/blog/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/moonshotai/kimi-k2-0905.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Parameter counts are the exact 0905 card's nominal 1T total and 32B activated. July K2 report counts of 1.04T/32.6B are not transferred. The native checkpoint is a decoder-only MoE transformer with MLA and 262,144-token context; FP8 scale/storage elements are not counted as learned parameters. Dedicated reasoning-mode classification remains unverified.", "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-26T23:48:32.732412Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/355.json" }, { "id": "swiss-ai/apertus-70b", "slug": "swiss-ai--apertus-70b", "name": "Apertus 70B Instruct 2509", "developer": "Swiss AI", "developer_id": "swiss-ai", "family": "apertus", "description": "Instruction-tuned multilingual text model from the Swiss AI Initiative, using a dense decoder-only Transformer with xIELU activations. Pretrained on 15 trillion tokens; 65,536-token context, tool-use support and Apache-2.0 weights.", "release_date": "2025-09-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 70.59986448, "active_parameters_b": 70, "parameters_source": "primary_source", "architecture": "dense", "context_window": 65536, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://huggingface.co/swiss-ai/Apertus-70B-Instruct-2509", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/swiss-ai/Apertus-70B-Instruct-2509", "type": "weights" }, { "label": "Apertus release announcement", "url": "https://ethz.ch/en/news-and-events/eth-news/news/2025/09/press-release-apertus-a-fully-open-transparent-multilingual-language-model.html", "type": "primary" }, { "label": "Apertus v1 technical report", "url": "https://arxiv.org/html/2509.14233v2", "type": "paper" }, { "label": "Pinned checkpoint configuration", "url": "https://huggingface.co/swiss-ai/Apertus-70B-Instruct-2509/blob/0f2767662a013c0fb631079cea883155252bc3b6/config.json", "type": "primary" }, { "label": "Pinned checkpoint tensor index", "url": "https://huggingface.co/swiss-ai/Apertus-70B-Instruct-2509/blob/0f2767662a013c0fb631079cea883155252bc3b6/model.safetensors.index.json", "type": "weights" }, { "label": "Apertus model implementation", "url": "https://github.com/huggingface/transformers/blob/v4.56.0/src/transformers/models/apertus/modeling_apertus.py", "type": "primary" }, { "label": "xIELU parameters and buffers", "url": "https://github.com/huggingface/transformers/blob/v4.56.0/src/transformers/activations.py", "type": "primary" }, { "label": "Embedding tying implementation", "url": "https://github.com/huggingface/transformers/blob/v4.56.0/src/transformers/modeling_utils.py", "type": "primary" }, { "label": "Checkpoint license", "url": "https://huggingface.co/swiss-ai/Apertus-70B-Instruct-2509/blob/0f2767662a013c0fb631079cea883155252bc3b6/LICENSE.txt", "type": "primary" }, { "label": "Swiss AI clarification on thinking support", "url": "https://huggingface.co/swiss-ai/Apertus-70B-Instruct-2509/discussions/18", "type": "primary" }, { "label": "Kolibri: exact Apertus 2509 nominal active-count comparison", "url": "https://arxiv.org/pdf/2610.31945v1", "type": "paper" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/swiss-ai/apertus-70b.toml", "type": "catalog" }, { "label": "Paper", "url": "https://arxiv.org/abs/2509.14233", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Apertus-70B-Instruct-2509: 70.59986448B learned parameters, including 160 learned activation scalars and excluding 160 buffers. Nominal 70B active parameters follow Aleph Alpha's Kolibri comparison convention for this exact checkpoint; no exact per-token participation count is established.", "field_status": {}, "audited_at": "2026-10-27T17:57:43.336407Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/498.json" }, { "id": "swiss-ai/apertus-8b", "slug": "swiss-ai--apertus-8b", "name": "Apertus 8B Instruct 2509", "developer": "Swiss AI", "developer_id": "swiss-ai", "family": "apertus", "description": "Instruction-tuned multilingual text model from the Swiss AI Initiative, released with Apache-2.0 weights, a 65,536-token context window and tool calling.", "release_date": "2025-09-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 65536, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://huggingface.co/swiss-ai/Apertus-8B-Instruct-2509", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/swiss-ai/Apertus-8B-Instruct-2509", "type": "weights" }, { "label": "Apertus technical report", "url": "https://arxiv.org/html/2509.14233v2", "type": "paper" }, { "label": "Official checkpoint configuration", "url": "https://huggingface.co/swiss-ai/Apertus-8B-Instruct-2509/blob/main/config.json", "type": "docs" }, { "label": "ETH Zurich release announcement", "url": "https://ethz.ch/en/news-and-events/eth-news/news/2025/09/press-release-apertus-a-fully-open-transparent-multilingual-language-model.html", "type": "announcement" }, { "label": "Developer thinking-mode clarification", "url": "https://huggingface.co/swiss-ai/Apertus-70B-Instruct-2509/discussions/18", "type": "docs" }, { "label": "Apertus release documentation", "url": "https://huggingface.co/swiss-ai/Apertus-70B-2509/blob/main/Apertus_EU_Code_of_Practice.pdf", "type": "docs" }, { "label": "Original entropy study: exact 8B comparison", "url": "https://arxiv.org/html/2602.20052v1", "type": "paper" }, { "label": "Apertus 1.5 architecture and comparison report", "url": "https://arxiv.org/html/2610.15542v1", "type": "paper" }, { "label": "Kolibri original comparison: Apertus 70B only", "url": "https://arxiv.org/pdf/2610.31945v1", "type": "paper" }, { "label": "Olmo 3 original Apertus comparisons", "url": "https://arxiv.org/html/2512.13961v2", "type": "paper" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/swiss-ai/apertus-8b.toml", "type": "catalog" }, { "label": "Paper", "url": "https://arxiv.org/abs/2509.14233", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The 8B parameter count is the developer's nominal total for Apertus-8B-Instruct-2509. Active parameters are not separately established by the original documentation or later comparisons checked. Apertus 1.5 has a different multimodal vocabulary and checkpoint.", "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T17:59:54.820Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/499.json" }, { "id": "vispark/vision-large", "slug": "vispark--vision-large", "name": "Vision Large", "developer": "Vispark", "developer_id": "vispark", "family": "Vision", "description": "Vispark’s large Vision API model, with text, image, audio, video and PDF input, text output, a one-million-token context window and tool calling.", "release_date": "2025-09-01", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 550, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "transformer", "context_window": 1000000, "modalities": { "input": [ "text", "image", "audio", "video", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://blog.vispark.in/vision-ai/", "source_type": "primary", "sources": [ { "label": "Vispark Lab public model roster", "url": "https://api.lab.vispark.in/v1/models", "type": "primary" }, { "label": "Vispark Lab deployed API documentation", "url": "https://lab.vispark.in/assets/index-g5phCnAo.js", "type": "primary" }, { "label": "Vision AI by Vispark: a foundational Indian multimodal AI", "url": "https://blog.vispark.in/vision-ai/", "type": "primary" }, { "label": "Vispark timeline and Terms of Service in deployed application", "url": "https://vispark.in/assets/index-BgxKNlqk.js", "type": "primary" }, { "label": "A Culturally Aware Multimodal AI Model", "url": "https://www.opastpublishers.com/open-access-articles/a-culturally-aware-multimodal-ai-model.pdf", "type": "primary" }, { "label": "A Self-Improving Experiential AI Model: A Path to Continual Learning", "url": "https://www.opastpublishers.com/open-access-articles/a-selfimproving-experiential-ai-model-a-path-to-continual-learning.pdf", "type": "primary" }, { "label": "Vision CLv1 model specifications and family", "url": "https://www.aivoco.com/", "type": "primary" }, { "label": "Continual Learning Models: An Introduction to Weight Management", "url": "https://blog.aivoco.com/continual-learning-models-an-introduction-to-weight-management/", "type": "primary" }, { "label": "Vision CLv1 ARC-AGI-3 evaluation repository", "url": "https://github.com/vansh-one/arc-agi-3_Vision-CLv1", "type": "primary" }, { "label": "Original Vision Large catalog record", "url": "https://raw.githubusercontent.com/anomalyco/models.dev/dev/models/vispark/vision-large.toml", "type": "catalog" }, { "label": "Vision CLv1 evaluation API client", "url": "https://raw.githubusercontent.com/vansh-one/arc-agi-3_Vision-CLv1/main/src/vision_arc_agi/vision.py", "type": "primary" }, { "label": "Vision CLv1 evaluation harness architecture", "url": "https://raw.githubusercontent.com/vansh-one/arc-agi-3_Vision-CLv1/main/docs/ARCHITECTURE.md", "type": "primary" }, { "label": "AIVOCO model library", "url": "https://www.aivoco.com/models", "type": "primary" }, { "label": "AIVOCO published research index", "url": "https://www.aivoco.com/research", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/vispark/vision-large.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Vision Large has a nominal developer-published 550B total for the later CLv1 model, linked through the developer’s evaluation repository. Active parameters and dense/MoE structure are not separately disclosed. Earlier Vision papers describe different generations and do not supply this count. The developer’s September 1, 2025 launch article explicitly offers the Large API; this dates the alias’s documented public launch, not its later CLv1 backend. The August 29 Lab launch does not enumerate these sizes, and generic May 2024 creation metadata is not used. Returned learning state and the open evaluation harness do not make the base weights open. Capabilities reflect provider documentation.", "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T19:28:30.009793Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/519.json" }, { "id": "vispark/vision-medium", "slug": "vispark--vision-medium", "name": "Vision Medium", "developer": "Vispark", "developer_id": "vispark", "family": "Vision", "description": "Medium-size Vision model served through Vispark Lab, accepting text, image, audio, video and PDF inputs and returning text, with reasoning and tool calling.", "release_date": "2025-09-01", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 175, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "transformer", "context_window": 1000000, "modalities": { "input": [ "text", "image", "audio", "video", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://blog.vispark.in/vision-ai/", "source_type": "primary", "sources": [ { "label": "Public model roster", "url": "https://api.lab.vispark.in/v1/models", "type": "primary" }, { "label": "Vision playground and API documentation", "url": "https://lab.vispark.in/#vision", "type": "primary" }, { "label": "Vision CLv1 model family table", "url": "https://www.aivoco.com/", "type": "primary" }, { "label": "Continual Learning Models: An Introduction to Weight Management", "url": "https://blog.aivoco.com/continual-learning-models-an-introduction-to-weight-management/", "type": "primary" }, { "label": "Vision AI by Vispark: a foundational Indian multimodal AI", "url": "https://blog.vispark.in/vision-ai/", "type": "primary" }, { "label": "Vispark product timeline", "url": "https://vispark.in/timeline", "type": "primary" }, { "label": "Terms of Service, September 26, 2025", "url": "https://vispark.in/terms", "type": "primary" }, { "label": "Vispark official timeline: Lab platform launch", "url": "https://vispark.in/assets/index-BgxKNlqk.js", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/vispark/vision-medium.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "175B is the nominal developer-published total for the later Vision CLv1 Medium served at Vispark Lab, described as a modified Transformer. Active parameters and dense/MoE structure remain undisclosed. The API alias has no dated checkpoint or transition history; CLv1 specifications are not backdated to 2024. The September 1, 2025 developer introduction documents the public launch of the Medium API alias; it does not date the later CLv1 transition. The August 29 Lab platform launch does not enumerate the sizes. Context is 1,000,000 tokens with a separate 980,000-token combined-input limit. Continual-learning state files are not publicly released base weights.", "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T19:28:47.333932Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/520.json" }, { "id": "vispark/vision-small", "slug": "vispark--vision-small", "name": "Vision Small", "developer": "Vispark", "developer_id": "vispark", "family": "Vision", "description": "Small Vision model served through Vispark Lab, with text, image, audio, video and PDF inputs, text output and tool calling.", "release_date": "2025-09-01", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 45, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "transformer", "context_window": 1000000, "modalities": { "input": [ "text", "image", "audio", "video", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://blog.vispark.in/vision-ai/", "source_type": "primary", "sources": [ { "label": "Vispark public model roster", "url": "https://api.lab.vispark.in/v1/models", "type": "primary" }, { "label": "Vispark Lab model documentation and playground", "url": "https://lab.vispark.in/", "type": "primary" }, { "label": "AIVOCO Vision CLv1 model family", "url": "https://www.aivoco.com/", "type": "primary" }, { "label": "Continual Learning Models: An Introduction to Weight Management", "url": "https://blog.aivoco.com/continual-learning-models-an-introduction-to-weight-management/", "type": "primary" }, { "label": "AIVOCO model playground destination", "url": "https://www.aivoco.com/playground", "type": "primary" }, { "label": "Vispark product timeline", "url": "https://vispark.in/timeline", "type": "primary" }, { "label": "Vision AI by Vispark: a foundational Indian multimodal AI", "url": "https://blog.vispark.in/vision-ai/", "type": "primary" }, { "label": "Vispark Terms of Service", "url": "https://vispark.in/terms", "type": "primary" }, { "label": "Vispark official timeline: Lab platform launch", "url": "https://vispark.in/assets/index-BgxKNlqk.js", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/vispark/vision-small.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "AIVOCO identifies the current Vision Small alias with its later CLv1 Small model and reports a nominal 45B total. Active parameters and dense/MoE layout are undisclosed. The family is described as a modified Transformer. The September 1, 2025 developer introduction documents the public launch of the Small API alias, not the later CLv1 checkpoint. The August 29 Lab platform launch does not enumerate the sizes; the CLv1 transition remains undated. Downloadable continual-learning state is not the complete hosted checkpoint.", "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T19:28:57.058963Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/521.json" }, { "id": "alibaba/qwen3-next-80b-a3b-instruct", "slug": "alibaba--qwen3-next-80b-a3b-instruct", "name": "Qwen3-Next 80B-A3B Instruct", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Non-thinking Qwen3-Next text model with hybrid attention and sparse MoE, supporting multilingual chat and tool use.", "release_date": "2025-09", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 80, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-next-80b-a3b-instruct.toml", "source_type": "catalog", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Instruct", "type": "weights" }, { "label": "Qwen checkpoint configuration", "url": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Instruct/raw/main/config.json", "type": "primary" }, { "label": "Alibaba Cloud model releases", "url": "https://help.aliyun.com/en/model-studio/newly-released-models", "type": "primary" }, { "label": "Official release announcement", "url": "https://qwen.ai/blog?id=qwen3-next", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-next-80b-a3b-instruct.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "80B total and 3B activated are nominal developer-reported counts. Architecture combines Gated DeltaNet and Gated Attention with MoE. Native context is 262144 tokens. Release month is September 2025; indexed announcement and API availability dates differ by one day.", "field_status": {}, "audited_at": "2026-10-26T14:34:56Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/038.json" }, { "id": "alibaba/qwen3-next-80b-a3b-thinking", "slug": "alibaba--qwen3-next-80b-a3b-thinking", "name": "Qwen3-Next 80B-A3B (Thinking)", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Open-weight Qwen3-Next thinking model with 80B total and 3B active parameters, hybrid attention, and tool-calling support.", "release_date": "2025-09", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 80, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-next-80b-a3b-thinking.toml", "source_type": "catalog", "sources": [ { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-next-80b-a3b-thinking.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Thinking", "type": "weights" }, { "label": "Official model configuration", "url": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Thinking/blob/main/config.json", "type": "primary" }, { "label": "Alibaba Cloud release log", "url": "https://help.aliyun.com/en/model-studio/newly-released-models", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "80B total and 3B active are nominal developer counts for this exact thinking checkpoint. Hybrid attention combines Gated DeltaNet and gated attention with MoE. Native context is 262144. Release month is September 2025; indexed announcement displays September 10 and API table says September 11.", "field_status": {}, "audited_at": "2026-10-26T14:34:39Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/039.json" }, { "id": "mistral/magistral-medium-2509", "slug": "mistral--magistral-medium-2509", "name": "Magistral Medium 1.2", "developer": "Mistral", "developer_id": "mistral", "family": "magistral", "description": null, "release_date": "2025-09", "date_precision": "month", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 128000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://docs.mistral.ai/models/magistral-medium-1-2-25-09", "source_type": "primary", "sources": [ { "label": "Official version-specific model documentation", "url": "https://docs.mistral.ai/models/magistral-medium-1-2-25-09", "type": "docs" }, { "label": "Dated publisher API changelog", "url": "https://docs.mistral.ai/resources/changelogs", "type": "announcement" }, { "label": "Publisher model lifecycle", "url": "https://legal.mistral.ai/ai-governance/models/magistral-medium-1/", "type": "docs" } ], "aliases": [], "first_seen": "2026-10-29T10:46:35.789Z", "last_seen": "2026-10-29T10:46:35.789Z", "upstream_present": false, "availability": "public", "notes": "Original Mistral reasoning model served through its API; this version is now retired or deprecated. Parameter count and internal architecture are undisclosed. Context is the publisher’s rounded 128k API figure. The changelog dates release September 17, while version documentation and legal lifecycle say September 18, so only month precision is asserted." }, { "id": "cohere/command-a-translate-08-2025", "slug": "cohere--command-a-translate-08-2025", "name": "Command A Translate", "developer": "Cohere", "developer_id": "cohere", "family": "command-a", "description": "Translation model for multilingual conversion, localization, and cross-language workflows", "release_date": "2025-08-28", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-4.0", "parameters_b": 111, "active_parameters_b": 111, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://docs.cohere.com/docs/command-a-translate", "source_type": "primary", "sources": [ { "label": "Cohere model card", "url": "https://huggingface.co/CohereLabs/command-a-translate-08-2025", "type": "weights" }, { "label": "Cohere release announcement", "url": "https://docs.cohere.com/changelog/2025-08-28-command-a-translate", "type": "primary" }, { "label": "Cohere model documentation", "url": "https://docs.cohere.com/docs/command-a-translate", "type": "primary" }, { "label": "Cohere technical report", "url": "https://aclanthology.org/2025.wmt-1.55.pdf", "type": "primary" }, { "label": "Cohere models overview", "url": "https://cohere.com/models-overview", "type": "primary" }, { "label": "Later Cohere report: Command A Translate total and active parameters", "url": "https://arxiv.org/html/2609.13916v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/cohere/command-a-translate-08-2025.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/c4ai-command-a-translate-08-2025", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Cohere reports 111B total and 111B active parameters for Command A Translate in its later North Small Translate technical report; both are nominal counts. Architecture is a dense decoder-only Transformer fine-tuned from Command A. The API lists 8,000 input/context tokens and 8,000 maximum output tokens, also described as 16K combined. Research weights use CC-BY-NC-4.0 with Cohere Labs acceptable-use conditions. Reasoning=false refers to the explicitly disabled dedicated reasoning capability; the separate CommandA-WMT and Deep Translation workflows do not change this model specification.", "field_status": {}, "audited_at": "2026-10-26T18:33:12.881346+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/149.json" }, { "id": "openai/gpt-realtime-2025-08-28", "slug": "openai--gpt-realtime-2025-08-28", "name": "GPT-realtime-2025-08-28", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-realtime", "description": null, "release_date": "2025-08-28", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 32000, "modalities": { "input": [ "text", "audio", "image" ], "output": [ "text", "audio" ] }, "reasoning": null, "tool_call": null, "source_url": "https://developers.openai.com/api/docs/changelog", "source_type": "primary", "sources": [ { "label": "Dated publisher API releases", "url": "https://developers.openai.com/api/docs/changelog", "type": "documentation" }, { "label": "Publisher model documentation and snapshot mapping", "url": "https://developers.openai.com/api/docs/models/gpt-realtime", "type": "documentation" }, { "label": "Dated launch", "url": "https://openai.com/index/introducing-gpt-realtime/", "type": "announcement" } ], "aliases": [ "openai/gpt-realtime" ], "first_seen": "2026-10-29T10:31:17.277Z", "last_seen": "2026-10-29T10:31:17.277Z", "upstream_present": false, "availability": "public", "notes": "Original publisher conversational audio model. " }, { "id": "google/gemini-2.5-flash-image", "slug": "google--gemini-2.5-flash-image", "name": "Gemini 2.5 flash image", "developer": "Google", "developer_id": "google", "family": "gemini", "description": null, "release_date": "2025-08-26", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 65536, "modalities": { "input": [ "text", "image" ], "output": [ "text", "image" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.google.dev/gemini-api/docs/changelog", "source_type": "primary", "sources": [ { "label": "Dated publisher API releases", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "documentation" }, { "label": "Publisher interleaved text and image capability", "url": "https://ai.google.dev/gemini-api/docs/image-generation", "type": "documentation" }, { "label": "Publisher model specifications", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-flash-image", "type": "documentation" } ], "aliases": [ "google/gemini-2.5-flash-image-preview" ], "first_seen": "2026-10-29T10:31:17.277Z", "last_seen": "2026-10-29T10:31:17.277Z", "upstream_present": false, "availability": "public", "notes": "Generates interleaved text and images. Date records first preview availability; general availability followed 2025-10-02. Preview and stable serving names are retained as aliases, without evidence of a separate trained checkpoint. Context and capabilities describe current published model documentation; preview limits may have differed." }, { "id": "nousresearch/hermes-4-405b", "slug": "nousresearch--hermes-4-405b", "name": "Hermes 4 405B", "developer": "Nous Research", "developer_id": "nousresearch", "family": "hermes", "description": "Open-weight Llama 3.1 405B fine-tune with hybrid reasoning and function calling.", "release_date": "2025-08-26", "date_precision": "day", "open_weights": true, "license": "Llama 3.1 Community License", "parameters_b": 405, "active_parameters_b": 405, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/NousResearch/Hermes-4-405B", "source_type": "primary", "sources": [ { "label": "Hermes 4 405B model card", "url": "https://huggingface.co/NousResearch/Hermes-4-405B/blob/88e3dce03c4a5535e2f4a2bcc08e939a2b302f82/README.md", "type": "primary" }, { "label": "Hermes 4 405B configuration", "url": "https://huggingface.co/NousResearch/Hermes-4-405B/blob/88e3dce03c4a5535e2f4a2bcc08e939a2b302f82/config.json", "type": "primary" }, { "label": "Hermes 4 Technical Report, version 2", "url": "https://arxiv.org/pdf/2508.18255v2", "type": "primary" }, { "label": "Nous Research releases", "url": "https://nousresearch.com/releases", "type": "primary" }, { "label": "Introducing Llama 3.1: Our most capable models to date", "url": "https://ai.meta.com/blog/meta-llama-3-1/", "type": "primary" }, { "label": "Llama 3.1 405B Community License Agreement", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/LICENSE", "type": "primary" }, { "label": "Hermes 4 405B BF16 checkpoint index", "url": "https://huggingface.co/NousResearch/Hermes-4-405B/blob/88e3dce03c4a5535e2f4a2bcc08e939a2b302f82/model.safetensors.index.json", "type": "primary" }, { "label": "Jamba: Hybrid Transformer-Mamba Language Models, ICLR 2025", "url": "https://proceedings.iclr.cc/paper_files/paper/2025/file/a9ed43fa31dc8b4a7d7a673d713dcb5f-Paper-Conference.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/nousresearch/hermes-4-405b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "405B total is Nous's published nominal size. Nominal 405B active is inferred through the documented Llama 3.1 405B post-training lineage from AI21's explicit Jamba comparison-table convention; Nous does not separately report an exact active count. This record covers the original BF16 checkpoint. The exact Llama 3.1 base license covers derivatives; Nous retains a conflicting llama3 metadata tag and publishes no separate LICENSE file. Tool calling is a checkpoint capability and may depend on serving support.", "field_status": {}, "audited_at": "2026-10-27T15:48:41.844294Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/373.json" }, { "id": "aisingapore/gemma-sea-lion-v4-27b-it", "slug": "aisingapore--gemma-sea-lion-v4-27b-it", "name": "Gemma-SEA-LION-v4-27B-IT", "developer": "AI Singapore", "developer_id": "aisingapore", "family": "gemma", "description": "Gemma 3 27B tuned by AI Singapore for Southeast Asian languages and instruction following", "release_date": "2025-08-25", "date_precision": "day", "open_weights": true, "license": "gemma", "parameters_b": 27.43240664, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://sea-lion.ai/blog/sea-lion-v4-multimodal/", "source_type": "primary", "sources": [ { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/aisingapore/gemma-sea-lion-v4-27b-it.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/aisingapore/Gemma-SEA-LION-v4-27B-IT", "type": "weights" }, { "label": "Exact released checkpoint configuration", "url": "https://huggingface.co/aisingapore/Gemma-SEA-LION-v4-27B-IT/raw/c41c291558bdd8cbc37d77868cec0bc5b6921476/config.json", "type": "primary" }, { "label": "SEA-LION v4: Our First Multimodal Release, Open, Powerful & Small", "url": "https://sea-lion.ai/blog/sea-lion-v4-multimodal/", "type": "primary" }, { "label": "Exact checkpoint repository and tensor metadata", "url": "https://huggingface.co/api/models/aisingapore/Gemma-SEA-LION-v4-27B-IT", "type": "repository_metadata" }, { "label": "Published checkpoint tensor index", "url": "https://huggingface.co/aisingapore/Gemma-SEA-LION-v4-27B-IT/raw/c41c291558bdd8cbc37d77868cec0bc5b6921476/model.safetensors.index.json", "type": "primary" }, { "label": "Gemma 3 Technical Report", "url": "https://goo.gle/Gemma3Report", "type": "primary" }, { "label": "Transformers 4.50.0 Gemma 3 configuration implementation", "url": "https://raw.githubusercontent.com/huggingface/transformers/v4.50.0/src/transformers/models/gemma3/configuration_gemma3.py", "type": "primary" }, { "label": "Exact checkpoint tokenizer and conversation template", "url": "https://huggingface.co/aisingapore/Gemma-SEA-LION-v4-27B-IT/raw/c41c291558bdd8cbc37d77868cec0bc5b6921476/tokenizer_config.json", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "27B is the developer's nominal size; published BF16 tensor metadata totals 27.43240664B including vision. Dense architecture has no separately reported active-per-token count. 128K resolves to 131072 positions in the Gemma 3 configuration. Dedicated reasoning-mode support is unverified; ordinary reasoning tasks do not establish a thinking-model flag.", "field_status": { "active_parameters_b": "not_applicable", "reasoning": "unverified" }, "audited_at": "2026-10-26T13:56:47.851663Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/010.json" }, { "id": "cohere/command-a-reasoning-08-2025", "slug": "cohere--command-a-reasoning-08-2025", "name": "Command A Reasoning", "developer": "Cohere", "developer_id": "cohere", "family": "command-a", "description": "Cohere reasoning model for multilingual enterprise agents, tools, and complex workflows", "release_date": "2025-08-21", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-4.0 with Acceptable Use Addendum", "parameters_b": 111, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 256000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://docs.cohere.com/changelog/2025-08-21-command-a-reasoning", "source_type": "primary", "sources": [ { "label": "Cohere release announcement", "url": "https://docs.cohere.com/changelog/2025-08-21-command-a-reasoning", "type": "primary" }, { "label": "Official model card and weights", "url": "https://huggingface.co/CohereLabs/command-a-reasoning-08-2025", "type": "weights" }, { "label": "Weights license with Acceptable Use Addendum", "url": "https://cohere.com/cohere-labs-cc-by-nc-license", "type": "primary" }, { "label": "Official checkpoint metadata", "url": "https://huggingface.co/api/models/CohereLabs/command-a-reasoning-08-2025", "type": "primary" }, { "label": "Cohere2 Transformer implementation", "url": "https://raw.githubusercontent.com/huggingface/transformers/v4.55.0/src/transformers/models/cohere2/modeling_cohere2.py", "type": "primary" }, { "label": "Later Cohere family comparison", "url": "https://cohere.com/blog/command-a-plus", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/cohere/command-a-reasoning-08-2025.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/c4ai-command-a-reasoning-08-2025", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "111B is Cohere's nominal whole-model parameter count. Active parameters are not separately disclosed, including in the later Command A+ comparison. The exact checkpoint is a dense autoregressive Transformer with interleaved sliding-window and global attention. Open weights use CC-BY-NC 4.0 with an Acceptable Use Addendum. The correct checkpoint repository omits the c4ai- prefix.", "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T12:26:23.900655+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/148.json" }, { "id": "deepseek/deepseek-v3.1", "slug": "deepseek--deepseek-v3.1", "name": "DeepSeek-V3.1", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek", "description": "Hybrid-reasoning DeepSeek model with thinking and non-thinking modes", "release_date": "2025-08-21", "date_precision": "day", "open_weights": true, "license": "MIT License", "parameters_b": 682.636487424, "active_parameters_b": 37, "parameters_source": "primary_source", "architecture": "moe", "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.1", "source_type": "primary", "sources": [ { "label": "Official DeepSeek-V3.1 model card", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.1/blob/c0781d039fb7a1ba2abc4add0bdc293e92d2b8db/README.md", "type": "primary" }, { "label": "DeepSeek-V3.1 checkpoint configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.1/blob/c0781d039fb7a1ba2abc4add0bdc293e92d2b8db/config.json", "type": "weights" }, { "label": "DeepSeek-V3.1 release change log", "url": "https://api-docs.deepseek.com/updates/#date-2025-08-21", "type": "primary" }, { "label": "DeepSeek weight structure and sharing", "url": "https://github.com/deepseek-ai/DeepSeek-V3/blob/main/README_WEIGHTS.md", "type": "primary" }, { "label": "DeepSeek-V3.1 tensor index and bounded headers", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.1/resolve/c0781d039fb7a1ba2abc4add0bdc293e92d2b8db/model.safetensors.index.json", "type": "weights" }, { "label": "DeepSeek-V3.1 official model code", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.1/blob/c0781d039fb7a1ba2abc4add0bdc293e92d2b8db/modeling_deepseek.py", "type": "primary" }, { "label": "model.norm.weight bounded comparison", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.1/resolve/c0781d039fb7a1ba2abc4add0bdc293e92d2b8db/model-00160-of-000163.safetensors", "type": "weights" }, { "label": "model.layers.61.shared_head.norm.weight bounded comparison", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.1/resolve/c0781d039fb7a1ba2abc4add0bdc293e92d2b8db/model-00163-of-000163.safetensors", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepseek/deepseek-v3.1.toml", "type": "catalog" }, { "label": "Alias documentation", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.1", "type": "docs" } ], "aliases": [ "deepseek/deepseek-chat-v3.1" ], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete count: 671.0264192B main + 11.610068224B unique MTP; shared embedding/head counted once and FP8 scales excluded. The MTP norm differs from the main norm. Nominal 37B active applies to the main model. Advertised context is 128K; config positions are 163840. General tool calls use non-thinking mode; thinking has a separate search-tool format.", "active_parameters_scope": "main", "parameter_components_b": { "main": 671.0264192, "mtp_unique": 11.610068224 }, "field_status": {}, "audited_at": "2026-10-26T19:45:21.462057+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/175.json", "alias_sources": { "deepseek/deepseek-chat-v3.1": { "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.1", "note": "Duplicate catalog names for the same DeepSeek-V3.1 chat checkpoint." } } }, { "id": "bytedance-seed/seed-oss-36b-instruct", "slug": "bytedance-seed--seed-oss-36b-instruct", "name": "Seed-OSS-36B-Instruct", "developer": "ByteDance", "developer_id": "bytedance-seed", "family": "seed", "description": "Text model with long-context instruction following, adjustable reasoning, and tool calling.", "release_date": "2025-08-20", "date_precision": "day", "open_weights": true, "license": "Apache 2.0", "parameters_b": 36.151104512, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 524288, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/ByteDance-Seed/Seed-OSS-36B-Instruct", "source_type": "primary", "sources": [ { "label": "Official Seed-OSS Instruct model card", "url": "https://huggingface.co/ByteDance-Seed/Seed-OSS-36B-Instruct/blob/497f1dca95ebdec98e41d517b9f060ee753c902f/MODEL_CARD.md", "type": "primary" }, { "label": "Official Seed-OSS Instruct README", "url": "https://huggingface.co/ByteDance-Seed/Seed-OSS-36B-Instruct/blob/497f1dca95ebdec98e41d517b9f060ee753c902f/README.md", "type": "primary" }, { "label": "Official Instruct tensor index", "url": "https://huggingface.co/ByteDance-Seed/Seed-OSS-36B-Instruct/blob/497f1dca95ebdec98e41d517b9f060ee753c902f/model.safetensors.index.json", "type": "weights" }, { "label": "Official Instruct configuration", "url": "https://huggingface.co/ByteDance-Seed/Seed-OSS-36B-Instruct/blob/497f1dca95ebdec98e41d517b9f060ee753c902f/config.json", "type": "weights" }, { "label": "SeedOss implementation recommended by ByteDance", "url": "https://github.com/huggingface/transformers/blob/56d68c6706ee052b445e1e476056ed92ac5eb383/src/transformers/models/seed_oss/modeling_seed_oss.py", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/bytedance-seed/seed-oss-36b-instruct.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/ByteDance-Seed/Seed-OSS-36B-Instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "36B is the nominal label; the original BF16 checkpoint contains exactly 36,151,104,512 learned parameters, confirmed by its index metadata and the prior complete header census. ByteDance does not separately specify active parameters in the checked sources; SiliconFlow lists 36B activated without a source or counting convention, so that claim remains unverified.", "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T11:10:21.021337Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/140.json" }, { "id": "xai/grok-code-fast-1", "slug": "xai--grok-code-fast-1", "name": "Grok Code Fast 1", "developer": "xAI", "developer_id": "xai", "family": "grok", "description": "Text-only reasoning model for coding tasks and tool use in agentic programming environments.", "release_date": "2025-08-20", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 256000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://forum.cursor.com/t/sonic-ghost-model-discussion/131085", "source_type": "primary", "sources": [ { "label": "Original developer announcement", "url": "https://x.ai/news/grok-code-fast-1", "type": "primary" }, { "label": "Original model card", "url": "https://data.x.ai/2025-08-26-grok-code-fast-1-model-card.pdf", "type": "primary" }, { "label": "Cursor identifies sonic as Grok Code Fast 1", "url": "https://forum.cursor.com/t/grok-code-out-now-free-week/132063", "type": "primary" }, { "label": "Public sonic preview in Cursor", "url": "https://forum.cursor.com/t/sonic-ghost-model-discussion/131085", "type": "primary" }, { "label": "Original dated public-preview post", "url": "https://cdn.syndication.twimg.com/tweet-result?id=1957977611646439628&lang=en&token=1", "type": "primary" }, { "label": "Named Copilot rollout", "url": "https://github.blog/changelog/2025-08-26-grok-code-fast-1-is-rolling-out-in-public-preview-for-github-copilot/", "type": "provider" }, { "label": "Historical Code Fast 1 specifications", "url": "https://docs.oracle.com/en-us/iaas/Content/generative-ai/xai-grok-code-fast-1.htm", "type": "provider" }, { "label": "Original slug retirement and redirect", "url": "https://docs.x.ai/developers/migration/may-15-retirement", "type": "primary" }, { "label": "Release-period service terms", "url": "https://x.ai/legal/terms-of-service-enterprise/previous-2025-06-27", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/xai/grok-code-fast-1.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "August 20, 2025 is the earliest documented sonic preview, later identified as Grok Code Fast 1; named Cursor/Copilot rollout followed August 26 and xAI announcement August 28. Preview checkpoints changed. Total and active parameters remain undisclosed after the October 27 review. xAI says this model was built from scratch with a new architecture; Grok-1’s 314B and activation fraction do not transfer. Original text-only 256000-token specifications are preserved. Current original-slug routing to Grok Build 0.1 describes a replacement.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T18:33:53.969443+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/536.json" }, { "id": "nvidia/nemotron-nano-9b-v2", "slug": "nvidia--nemotron-nano-9b-v2", "name": "Nemotron Nano 9B v2", "developer": "NVIDIA", "developer_id": "nvidia", "family": "nemotron", "description": "Text model pruned and distilled from Nemotron Nano 12B v2, combining Mamba-2, attention and dense MLP layers with configurable reasoning and tool calling.", "release_date": "2025-08-18", "date_precision": "day", "open_weights": true, "license": "NVIDIA Open Model License", "parameters_b": 8.888227328, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "hybrid-ssm-transformer", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-9B-v2/blob/6533e8de2c68e4536bf7c411d7a3ce5734111476/README.md", "source_type": "primary", "sources": [ { "label": "Pinned Nemotron Nano 9B v2 model card", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-9B-v2/blob/6533e8de2c68e4536bf7c411d7a3ce5734111476/README.md", "type": "primary" }, { "label": "Pinned Nano 9B v2 checkpoint configuration", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-9B-v2/blob/6533e8de2c68e4536bf7c411d7a3ce5734111476/config.json", "type": "primary" }, { "label": "Pinned exact-model implementation", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-9B-v2/blob/6533e8de2c68e4536bf7c411d7a3ce5734111476/modeling_nemotron_h.py", "type": "primary" }, { "label": "Pinned checkpoint tensor index", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-9B-v2/blob/6533e8de2c68e4536bf7c411d7a3ce5734111476/model.safetensors.index.json", "type": "primary" }, { "label": "Nano 2 technical report, revision 4", "url": "https://arxiv.org/html/2508.14444v4", "type": "primary" }, { "label": "August 18 Nano 2 release announcement", "url": "https://research.nvidia.com/labs/adlr/NVIDIA-Nemotron-Nano-2/", "type": "primary" }, { "label": "NVIDIA Open Model License Agreement", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/", "type": "primary" }, { "label": "Pinned NVIDIA checkpoint parameter metadata", "url": "https://huggingface.co/api/models/nvidia/NVIDIA-Nemotron-Nano-9B-v2/revision/6533e8de2c68e4536bf7c411d7a3ce5734111476", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/nvidia/nemotron-nano-9b-v2.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The complete checkpoint has 8.888227328B learned parameters, corroborated by pinned Hugging Face metadata and NVIDIA’s rounded 8.89B specification. NVIDIA does not separately specify active parameters; unverified third-party 9B active claims are not adopted. August 18 is the explicit checkpoint/API release and announcement date; an experimental NGC container lists August 15 without corroborating earlier public availability.", "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T16:09:04.983906Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/393.json" }, { "id": "google/gemma-3-270m", "slug": "google--gemma-3-270m", "name": "gemma-3-270m", "developer": "Google", "developer_id": "google", "family": "gemma", "description": null, "release_date": "2025-08-14", "date_precision": "day", "open_weights": true, "license": "Gemma", "parameters_b": 0.27, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://developers.googleblog.com/introducing-gemma-3-270m/", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://developers.googleblog.com/introducing-gemma-3-270m/", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/google/gemma-3-270m/blob/9b0cfec892e2bc2afd938c98eabe4e4a7b1e0ca1/README.md", "type": "model_card" }, { "label": "Developer model card", "url": "https://ai.google.dev/gemma/docs/core/model_card_3", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Pretrained text-only checkpoint. The launch states 270M total parameters, including 170M embedding parameters. Weights require agreement to Google’s Gemma terms." }, { "id": "google/gemma-3-270m-it", "slug": "google--gemma-3-270m-it", "name": "gemma-3-270m-it", "developer": "Google", "developer_id": "google", "family": "gemma", "description": null, "release_date": "2025-08-14", "date_precision": "day", "open_weights": true, "license": "Gemma", "parameters_b": 0.27, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://developers.googleblog.com/introducing-gemma-3-270m/", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://developers.googleblog.com/introducing-gemma-3-270m/", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/google/gemma-3-270m-it/blob/ac82b4e820549b854eebf28ce6dedaf9fdfa17b3/README.md", "type": "model_card" }, { "label": "Developer model card", "url": "https://ai.google.dev/gemma/docs/core/model_card_3", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Instruction-tuned text-only checkpoint. The launch states 270M total parameters, including 170M embedding parameters. Weights require agreement to Google’s Gemma terms." }, { "id": "mistral/mistral-medium-3.1", "slug": "mistral--mistral-medium-3.1", "name": "Mistral Medium 3.1", "developer": "Mistral", "developer_id": "mistral", "family": "mistral-medium", "description": "Mistral's August 2025 multimodal model for text and image inputs, text responses, structured outputs, and function calling.", "release_date": "2025-08-12", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 123, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": null, "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://docs.mistral.ai/models/mistral-medium-3-1-25-08", "source_type": "primary", "sources": [ { "label": "Mistral Medium 3.1 model card", "url": "https://docs.mistral.ai/models/mistral-medium-3-1-25-08", "type": "primary" }, { "label": "Mistral release changelog", "url": "https://docs.mistral.ai/resources/changelogs", "type": "primary" }, { "label": "Mistral native context limitations", "url": "https://docs.mistral.ai/resources/known-limitations", "type": "primary" }, { "label": "OpenRouter exact Medium 3.1 serving endpoints", "url": "https://openrouter.ai/api/v1/models/mistralai/mistral-medium-3.1/endpoints", "type": "primary" }, { "label": "Mistral Document QnA processing", "url": "https://docs.mistral.ai/studio/document-processing/document_qna", "type": "primary" }, { "label": "Mistral commercial terms", "url": "https://legal.mistral.ai/terms/commercial-terms-of-service/", "type": "primary" }, { "label": "Phoenix-VL derivative technical report", "url": "https://arxiv.org/html/2605.10391v1", "type": "primary" }, { "label": "Phoenix-VL exact-base lineage and training stages (EMNLP 2026)", "url": "https://aclanthology.org/2026.emnlp-industry.260.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/mistral-medium-3.1.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Exact revision: mistral-medium-2508. Total 123B is a nominal lineage inference from the HTX/Mistral Phoenix reports, which identify this base checkpoint and describe continued training with the multimodal backbone preserved. Active parameters and dense versus MoE routing remain undisclosed. Medium 3.5 specifications are not transferred. Context 131072 is OpenRouter's Mistral-provider hosted capacity; Mistral's native card states nominal 128k. Native inputs are text and images; PDF QnA uses OCR preprocessing. No dedicated reasoning mode was verified. The card is deprecated, and model data records retirement on 2026-08-31; endpoint operation was not tested.", "field_status": { "active_parameters_b": "undisclosed", "architecture": "unverified", "reasoning": "unverified" }, "audited_at": "2026-10-27T15:37:59.237562Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/339.json" }, { "id": "baichuan/baichuan-m2-32b", "slug": "baichuan--baichuan-m2-32b", "name": "Baichuan M2 32B", "developer": "Baichuan", "developer_id": "baichuan", "family": "baichuan", "description": "Open-weight medical language model based on Qwen2.5-32B, with dense architecture and selectable thinking modes.", "release_date": "2025-08-11", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 32.763876352, "active_parameters_b": 32, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://finance.sina.com.cn/roll/2025-08-11/doc-infkqwrc9743137.shtml", "source_type": "secondary", "sources": [ { "label": "Baichuan-M2-32B model card", "url": "https://huggingface.co/baichuan-inc/Baichuan-M2-32B/blob/360f684d3cd527a4ef986c54071015eb0e6def58/README.md", "type": "primary" }, { "label": "Baichuan-M2-32B target configuration", "url": "https://huggingface.co/baichuan-inc/Baichuan-M2-32B/blob/360f684d3cd527a4ef986c54071015eb0e6def58/config.json", "type": "primary" }, { "label": "Baichuan-M2-32B Apache License", "url": "https://huggingface.co/baichuan-inc/Baichuan-M2-32B/blob/360f684d3cd527a4ef986c54071015eb0e6def58/LICENSE", "type": "primary" }, { "label": "Baichuan-M2-32B repository metadata", "url": "https://huggingface.co/api/models/baichuan-inc/Baichuan-M2-32B", "type": "primary" }, { "label": "Baichuan-M2-32B target index and all 14 shard headers", "url": "https://huggingface.co/baichuan-inc/Baichuan-M2-32B/blob/360f684d3cd527a4ef986c54071015eb0e6def58/model.safetensors.index.json", "type": "primary" }, { "label": "Baichuan-M2-32B chat and tool template", "url": "https://huggingface.co/baichuan-inc/Baichuan-M2-32B/blob/360f684d3cd527a4ef986c54071015eb0e6def58/tokenizer_config.json", "type": "primary" }, { "label": "Baichuan-M2 technical blog", "url": "https://www.baichuan-ai.com/blog/baichuan-M2", "type": "primary" }, { "label": "Baichuan-M2: Scaling Medical Capability with Large Verifier System", "url": "https://arxiv.org/pdf/2509.02208", "type": "primary" }, { "label": "Baichuan-M2-32B weight repository history", "url": "https://huggingface.co/api/models/baichuan-inc/Baichuan-M2-32B/commits/main", "type": "primary" }, { "label": "Baichuan-M2-32B original public-project README commit", "url": "https://github.com/baichuan-inc/Baichuan-M2-32B/commit/28c2cc732cdfc7897601f71c9ee9f1620bb30e10", "type": "primary" }, { "label": "Hongxin Electronics announcement about the Baichuan-M2 launch", "url": "https://finance.sina.com.cn/roll/2025-08-18/doc-infmkcqi1648547.shtml", "type": "secondary" }, { "label": "Contemporaneous Baichuan-M2 release coverage", "url": "https://finance.sina.com.cn/roll/2025-08-11/doc-infkqwrc9743137.shtml", "type": "secondary" }, { "label": "Baichuan official model catalog", "url": "https://www.baichuan-ai.com/", "type": "primary" }, { "label": "Baichuan-M3 technical report: exact M2 comparison", "url": "https://arxiv.org/pdf/2602.06570", "type": "primary" }, { "label": "Baichuan-M4 technical report", "url": "https://arxiv.org/pdf/2606.08982", "type": "primary" }, { "label": "Developer-shipped Qwen2 target implementation for speculative serving", "url": "https://huggingface.co/baichuan-inc/Baichuan-M2-32B/blob/360f684d3cd527a4ef986c54071015eb0e6def58/draft/qwen2.py", "type": "primary" }, { "label": "Qwen3 Technical Report, Table 4: Qwen2.5-32B-Base activated parameters", "url": "https://arxiv.org/html/2505.09388v1#S3.T4", "type": "primary" }, { "label": "Qwen2.5-32B named base configuration", "url": "https://huggingface.co/Qwen/Qwen2.5-32B/blob/1818d35814b8319459f4bd55ed1ac8709630f003/config.json", "type": "primary" }, { "label": "Qwen2.5-32B repository metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2.5-32B", "type": "primary" }, { "label": "Baichuan-M2-32B current official project README", "url": "https://github.com/baichuan-inc/Baichuan-M2-32B", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/baichuan/baichuan-m2-32b.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/baichuan-inc/Baichuan-M2-32B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Target checkpoint count is 32.763876352B learned parameters; nominal developer label is 32B. Optional EAGLE3 draft is separate (0.766798848B learned parameters, excluding vocabulary mapping buffers) and is not included. Nominal active count is 32B, inferred from the explicit Qwen2.5-32B base lineage and its developer-published activated count; parameter-bearing dimensions are unchanged. This is not an exact per-token count. Config permits 131072 positions, while the official homepage advertises 32K and the report describes 32K SFT. Tool-call serialization is present but the developer says tool invocation is not fully optimized. Public launch is August 11, 2025, supported by contemporary news and a syndicated compute-partner announcement rather than repository modification dates.", "parameter_components_b": { "input_embeddings": 0.77856768, "normalization": 0.00066048, "dense_mlp": 27.17908992, "attention": 4.026990592, "output_head": 0.77856768 }, "field_status": {}, "audited_at": "2026-10-27T10:25:56.792487+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/122.json" }, { "id": "zhipuai/glm-4.5v", "slug": "zhipuai--glm-4.5v", "name": "GLM-4.5V", "developer": "Z.ai", "developer_id": "zhipuai", "family": "glm", "description": "MoE vision-language model based on GLM-4.5-Air, with image and video understanding, switchable reasoning, and text tool calls.", "release_date": "2025-08-11", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 107.71092736, "active_parameters_b": 12, "parameters_source": "primary_source", "architecture": "moe", "context_window": 65536, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/zai-org/GLM-4.5V/blob/ed47433b37111465ec527affaaddceff371bca04/README.md", "source_type": "primary", "sources": [ { "label": "GLM-4.5V official model card at audited revision", "url": "https://huggingface.co/zai-org/GLM-4.5V/blob/ed47433b37111465ec527affaaddceff371bca04/README.md", "type": "weights" }, { "label": "Exact GLM-4.5V checkpoint configuration", "url": "https://huggingface.co/zai-org/GLM-4.5V/blob/ed47433b37111465ec527affaaddceff371bca04/config.json", "type": "weights" }, { "label": "Complete exact GLM-4.5V checkpoint tensor index", "url": "https://huggingface.co/zai-org/GLM-4.5V/blob/ed47433b37111465ec527affaaddceff371bca04/model.safetensors.index.json", "type": "weights" }, { "label": "GLM4V MoE reference implementation matching checkpoint configuration", "url": "https://raw.githubusercontent.com/huggingface/transformers/v4.57.1/src/transformers/models/glm4v_moe/modeling_glm4v_moe.py", "type": "primary" }, { "label": "Exact GLM-4.5V chat template", "url": "https://huggingface.co/zai-org/GLM-4.5V/blob/ed47433b37111465ec527affaaddceff371bca04/chat_template.jinja", "type": "weights" }, { "label": "Exact GLM-4.5V tokenizer configuration", "url": "https://huggingface.co/zai-org/GLM-4.5V/blob/ed47433b37111465ec527affaaddceff371bca04/tokenizer_config.json", "type": "weights" }, { "label": "GLM-4.5V native developer documentation", "url": "https://docs.z.ai/guides/vlm/glm-4.5v", "type": "primary" }, { "label": "GLM-4.5V and GLM-4.1V-Thinking technical report, revision6", "url": "https://arxiv.org/html/2507.01006v6", "type": "primary" }, { "label": "Official checkpoint repository metadata", "url": "https://huggingface.co/api/models/zai-org/GLM-4.5V", "type": "weights" }, { "label": "Checkpoint shard 1 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00001-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 2 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00002-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 3 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00003-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 4 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00004-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 5 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00005-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 6 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00006-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 7 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00007-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 8 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00008-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 9 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00009-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 10 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00010-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 11 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00011-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 12 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00012-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 13 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00013-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 14 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00014-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 15 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00015-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 16 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00016-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 17 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00017-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 18 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00018-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 19 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00019-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 20 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00020-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 21 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00021-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 22 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00022-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 23 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00023-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 24 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00024-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 25 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00025-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 26 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00026-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 27 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00027-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 28 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00028-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 29 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00029-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 30 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00030-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 31 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00031-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 32 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00032-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 33 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00033-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 34 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00034-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 35 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00035-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 36 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00036-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 37 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00037-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 38 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00038-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 39 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00039-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 40 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00040-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 41 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00041-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 42 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00042-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 43 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00043-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 44 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00044-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 45 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00045-of-00046.safetensors", "type": "weights" }, { "label": "Checkpoint shard 46 of 46, bounded tensor header", "url": "https://huggingface.co/zai-org/GLM-4.5V/resolve/ed47433b37111465ec527affaaddceff371bca04/model-00046-of-00046.safetensors", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/zhipuai/glm-4.5v.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/GLM-4.5V", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Exact full checkpoint total is 107,710,927,360 learned parameters, including 106,852,245,504 language parameters with untied embeddings/output and 858,681,856 vision/projector parameters. All 46 bounded headers and 18,061 learned tensors were reconciled; 45 router correction buffers (5,760 F32 values) are excluded. This exact checkpoint contains no MTP module. Published 106B total and 12B active are nominal; 12B is retained as the author-reported active figure, with no exact VLM or vision-phase boundary asserted and no Air MTP counting convention transferred. Model position limit 65536 conflicts with tokenizer limit 128000; 16K in native API documentation is maximum output. Native outputs are text; PDF/file handling does not add a neural modality.", "parameter_components_b": { "language_backbone": 106.852245504, "vision_encoder_projector": 0.858681856 }, "field_status": {}, "audited_at": "2026-10-27T06:07:09.606234Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/550.json" }, { "id": "openai/gpt-5", "slug": "openai--gpt-5", "name": "GPT-5", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5", "description": "Reasoning model with text and image input, text output, a 400,000-token context window and function calling.", "release_date": "2025-08-07", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 400000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "primary", "sources": [ { "label": "Introducing GPT-5 for developers", "url": "https://openai.com/index/introducing-gpt-5-for-developers/", "type": "primary" }, { "label": "GPT-5 model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5.md", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "OpenAI API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "All models", "url": "https://developers.openai.com/api/docs/models/all", "type": "primary" }, { "label": "GPT-5 System Card", "url": "https://cdn.openai.com/gpt-5-system-card.pdf", "type": "primary" }, { "label": "From Hard Refusals to Safe-Completions: Toward Output-Centric Safety Training", "url": "https://cdn.openai.com/pdf/be60c07b-6bc2-4f54-bcee-4141e1d6c69a/gpt-5-safe_completions.pdf", "type": "primary" }, { "label": "GPT-5.1 Instant and GPT-5.1 Thinking System Card Addendum", "url": "https://cdn.openai.com/pdf/4173ec8d-1229-47db-96de-06d87147e07e/5_1_system_card.pdf", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201", "type": "primary" }, { "label": "Studying metagaming latents in language models", "url": "https://alignment.openai.com/metagaming-latents/", "type": "primary" }, { "label": "Learning Powerful Models: From Transformers to Reasoners and Beyond", "url": "https://forum.openai.com/en/public/videos/event-replay-learning-powerful-models-from-transformers-to-reasoners-and-beyond-2025-10-06", "type": "primary" }, { "label": "Reasoning Models Struggle to Control their Chains of Thought", "url": "https://cdn.openai.com/pdf/a21c39c1-fa07-41db-9078-973a12620117/cot_controllability.pdf", "type": "primary" }, { "label": "ZAYA1-8B Technical Report", "url": "https://arxiv.org/html/2605.05365v1", "type": "primary" }, { "label": "Cleaning Maintenance Logs with LLM Agents for Improved Predictive Maintenance", "url": "https://papers.phmsociety.org/index.php/phmap/article/download/4486/phmap_25_4486", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The gpt-5 alias defaults to gpt-5-2025-08-07, the reasoning API model released August 7, 2025; it is distinct from ChatGPT routing and gpt-5-chat-latest. Context is 400,000 tokens, with separate 272,000 maximum input and 128,000 maximum output limits. PDF files are processed into text and page images. Broad Transformer attention is inferred from exact-model prompt-caching support and its documented mechanism. Total and active parameter counts remain undisclosed; later Zyphra comparisons leave both GPT-5 counts blank. Complete structure and dense/MoE routing are undisclosed. OpenAI schedules this alias and snapshot for API shutdown December 11, 2026; no live endpoint request was made.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T16:31:17.179769Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/412.json" }, { "id": "openai/gpt-5-chat-latest", "slug": "openai--gpt-5-chat-latest", "name": "GPT-5 Chat (latest)", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5", "description": "Non-reasoning GPT-5 Chat API model with text and image input, text output and function calling.", "release_date": "2025-08-07", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 128000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "primary", "sources": [ { "label": "Introducing GPT-5 for developers", "url": "https://openai.com/index/introducing-gpt-5-for-developers/", "type": "primary" }, { "label": "GPT-5 Chat model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5-chat-latest", "type": "primary" }, { "label": "API deprecations: July 2026 shutdown", "url": "https://developers.openai.com/api/docs/deprecations", "type": "primary" }, { "label": "GPT-5 System Card, August 13, 2025 revision", "url": "https://cdn.openai.com/gpt-5-system-card.pdf", "type": "primary" }, { "label": "GPT-5.1 Instant and GPT-5.1 Thinking System Card Addendum", "url": "https://cdn.openai.com/pdf/4173ec8d-1229-47db-96de-06d87147e07e/5_1_system_card.pdf", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "Foundry Models sold by Azure: GPT-5 Chat historical specifications", "url": "https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "AgenticEval: Toward Agentic and Self-Evolving Safety Evaluation of Large Language Models", "url": "https://aclanthology.org/2026.findings-acl.727.pdf", "type": "primary" }, { "label": "A Self-Controlled Benchmark of Retrieval-Augmented Generation for Large Language Models on Clinical Guideline Questions", "url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13465324/", "type": "primary" }, { "label": "LLMRouterBench: A Massive Benchmark and Unified Framework for LLM Routing", "url": "https://aclanthology.org/2026.findings-acl.1881.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5-chat-latest.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The alias first became available August 7, 2025 as the non-reasoning GPT-5 model used in ChatGPT, separate from GPT-5 thinking and the full router system. Its documentation does not bind the last alias target to an immutable date. OpenAI labels its context 128,000 tokens and output cap 16,384; the Markdown also lists 272,000 maximum input, an unresolved internal inconsistency. The 128,000 context label is corroborated by Microsoft historical GPT-5 Chat specifications. Function calling is supported in the last-documented exact model page. PDF denotes API input processed as text and page images. OpenAI lists API shutdown on July 23, 2026; no live endpoint request was made. Total and active parameter counts remain undisclosed for this distinct chat model in OpenAI documentation and the inspected later research comparisons; the ChatGPT model-selection router does not establish an MoE count. Exact architecture remains unverified.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "unverified" }, "audited_at": "2026-10-27T16:31:20.716985Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/413.json" }, { "id": "openai/gpt-5-mini", "slug": "openai--gpt-5-mini", "name": "GPT-5 Mini", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5", "description": "GPT-5 reasoning variant with text and image input, text output, a 400,000-token context window and function calling.", "release_date": "2025-08-07", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 400000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "primary", "sources": [ { "label": "Introducing GPT-5 for developers", "url": "https://openai.com/index/introducing-gpt-5-for-developers/", "type": "primary" }, { "label": "GPT-5 Mini API model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5-mini.md", "type": "primary" }, { "label": "GPT-5 System Card", "url": "https://cdn.openai.com/gpt-5-system-card.pdf", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201.md", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "Introducing GPT-5.4 mini and nano", "url": "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", "type": "primary" }, { "label": "All OpenAI API models", "url": "https://developers.openai.com/api/docs/models/all.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "EXAONE 4.5 Technical Report: LG’s First Open-Weight Vision-Language Model for Industrial Intelligence", "url": "https://arxiv.org/html/2604.08644v1", "type": "primary" }, { "label": "EXAONE 4.5 33B official model card", "url": "https://huggingface.co/LGAI-EXAONE/EXAONE-4.5-33B", "type": "primary" }, { "label": "GPT-5.4 Thinking System Card, including GPT-5.4 mini appendix", "url": "https://deploymentsafety.openai.com/gpt-5-4-thinking/gpt-5-4-thinking.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5-mini.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The gpt-5-mini API alias identifies snapshot gpt-5-mini-2025-08-07 and the GPT-5 thinking-mini variant. Total and active parameters remain undisclosed; later LG AI Research comparison tables also leave both counts blank. The 400,000-token context is distinct from 272,000 maximum input and 128,000 maximum output. PDF input is converted to text and page images. Transformer is a limited operational inference from prompt-caching support and OpenAI’s applicable KV-cache documentation; block structure and dense/MoE routing are unspecified. The model is deprecated, with shutdown announced for December 11, 2026; no endpoint call was made.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T16:38:00Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/415.json" }, { "id": "openai/gpt-5-nano", "slug": "openai--gpt-5-nano", "name": "GPT-5 Nano", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5", "description": "GPT-5 nano is an API reasoning model with text and image inputs, text output, and function calling.", "release_date": "2025-08-07", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 400000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-gpt-5-for-developers/", "source_type": "primary", "sources": [ { "label": "GPT-5 nano model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5-nano.md", "type": "primary" }, { "label": "Introducing GPT-5 for developers", "url": "https://openai.com/index/introducing-gpt-5-for-developers/", "type": "primary" }, { "label": "GPT-5 System Card", "url": "https://cdn.openai.com/gpt-5-system-card.pdf", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201.md", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "Deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "All models", "url": "https://developers.openai.com/api/docs/models/all.md", "type": "primary" }, { "label": "GPT-5.1 Instant and GPT-5.1 Thinking System Card Addendum", "url": "https://cdn.openai.com/pdf/4173ec8d-1229-47db-96de-06d87147e07e/5_1_system_card.pdf", "type": "primary" }, { "label": "Introducing GPT-5.4 mini and nano", "url": "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/", "type": "primary" }, { "label": "Task-Aware Evaluation and Error-Overlap Analysis for Large Language Models", "url": "https://aclanthology.org/2025.chomps-main.1.pdf", "type": "primary" }, { "label": "Evaluating Large Language Models on the 2026 Korean CSAT Mathematics Exam", "url": "https://isoft.cnu.ac.kr/csat2026/assets/_2026__CSAT_ArXiv.pdf", "type": "primary" }, { "label": "Incompressible Knowledge Probes: Estimating Black-Box LLM Parameter Counts via Factual Capacity", "url": "https://arxiv.org/html/2604.24827v2", "type": "primary" }, { "label": "Nemotron 3 Nano: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning", "url": "https://arxiv.org/html/2512.20848v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5-nano.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The original API nano is gpt-5-thinking-nano in the GPT-5 system card, not the ChatGPT router or a GPT-5.4 nano variant. Published nominal total and active parameter counts remain undisclosed in the checked sources; performance-based size estimates are excluded. The broad transformer label is a cross-source inference from exact nano prompt-caching support and OpenAI's pre-GPT-5.6 KV-attention explanation, which also names nano; it does not establish the complete block design, dense/MoE routing or counts. PDF denotes API extraction to text and page images; native input modalities are text and image, with text output. The 400,000 context window has separate 272,000 maximum-input and 128,000 maximum-output limits. As of 2026-10-27 the model is deprecated, with shutdown scheduled for 2026-12-11; endpoint operation was not tested.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T16:37:54.934924Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/416.json" }, { "id": "openai/gpt-5-pro", "slug": "openai--gpt-5-pro", "name": "GPT-5 Pro", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-5", "description": "GPT-5 reasoning variant that uses additional inference compute, with text and image input and tool calling.", "release_date": "2025-08-07", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 400000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-gpt-5/", "source_type": "primary", "sources": [ { "label": "GPT-5 Pro — model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5-pro", "type": "primary" }, { "label": "Introducing GPT-5", "url": "https://openai.com/index/introducing-gpt-5/", "type": "primary" }, { "label": "API changelog — October 6, 2025", "url": "https://developers.openai.com/api/docs/changelog", "type": "primary" }, { "label": "GPT-5 System Card", "url": "https://cdn.openai.com/gpt-5-system-card.pdf", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs", "type": "primary" }, { "label": "Deprecations — GPT-5 and o3 models", "url": "https://developers.openai.com/api/docs/deprecations", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "All models", "url": "https://developers.openai.com/api/docs/models/all", "type": "primary" }, { "label": "Early science acceleration experiments with GPT-5", "url": "https://cdn.openai.com/pdf/4a25f921-e4e0-479a-9b38-5367b47e8fd0/early-science-acceleration-experiments-with-gpt-5.pdf", "type": "primary" }, { "label": "GPT-5.1 Instant and GPT-5.1 Thinking System Card Addendum", "url": "https://cdn.openai.com/pdf/4173ec8d-1229-47db-96de-06d87147e07e/5_1_system_card.pdf", "type": "primary" }, { "label": "GPT-5 — model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-5", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201", "type": "primary" }, { "label": "GeneBench: Assessing AI Agents for Multi-Stage Inference Problems in Genomics and Quantitative Biology", "url": "https://cdn.openai.com/pdf/6dc7175d-d9e7-4b8d-96b8-48fe5798cd5b/oai_genebench_benchmark.pdf", "type": "primary" }, { "label": "LongCat-Flash-Thinking Technical Report — Table 2", "url": "https://arxiv.org/html/2509.18883v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-5-pro.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "GPT-5 Pro first became publicly available in ChatGPT on 2025-08-07; the gpt-5-pro API alias and gpt-5-pro-2025-10-06 snapshot launched on 2025-10-06. The 400,000-token context and API capabilities refer to that documented API snapshot, not original ChatGPT plan limits. The system card defines Pro as GPT-5-thinking with parallel test-time compute. Total and active parameter counts remain undisclosed in the checked primary sources; no published base count supports a numerical lineage inference, and extra inference compute does not establish a larger checkpoint or MoE topology. Broad Transformer-based architecture is a qualified inference from that explicit model relationship, GPT-5 prompt-caching support, and OpenAI's pre-GPT-5.6 attention-KV caching explanation; Pro endpoint caching support is not asserted. PDF denotes API preprocessing into text and page images. Only high reasoning effort is supported; code interpreter is unsupported. The alias and snapshot were deprecated on 2026-06-11 with announced API shutdown on 2026-12-11, still in the future at review time. No live inference request was made.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T16:37:29.488005Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/417.json" }, { "id": "anthropic/claude-opus-4-1-20250805", "slug": "anthropic--claude-opus-4-1-20250805", "name": "Claude Opus 4.1", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude-opus", "description": "Claude Opus model with extended thinking, coding and tool-use capabilities, released as an update to Opus 4.", "release_date": "2025-08-05", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 200000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.anthropic.com/news/claude-opus-4-1", "source_type": "primary", "sources": [ { "label": "Anthropic release announcement", "url": "https://www.anthropic.com/news/claude-opus-4-1", "type": "primary" }, { "label": "Claude Opus 4.1 system card", "url": "https://www.anthropic.com/claude-opus-4-1-system-card", "type": "primary" }, { "label": "Anthropic introspection research", "url": "https://transformer-circuits.pub/2025/introspection/index.html", "type": "primary" }, { "label": "Amazon Bedrock exact-version documentation", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-anthropic-claude-opus-4-1.html", "type": "docs" }, { "label": "Google Cloud exact-model documentation (indexed)", "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/partner-models/claude/opus-4-1", "type": "docs" }, { "label": "Anthropic commercial terms", "url": "https://www.anthropic.com/legal/commercial-terms", "type": "primary" }, { "label": "Anthropic model deprecations", "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "type": "docs" }, { "label": "Anthropic statement on Claude weight availability", "url": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities", "type": "research" }, { "label": "Anthropic pricing table with conflicting context tier", "url": "https://www-cdn.anthropic.com/files/4zrzovbb/website/5678bc2f5978e5bcd4f1fe7c14b2c72284dcf9f8.pdf", "type": "documentation" }, { "label": "Claude Opus 4.5 System Card", "url": "https://assets.claude.com/bf10f64990cfda0ba858290be7b8cc6317685f47.pdf", "type": "primary" }, { "label": "Claude Sonnet 4.5 System Card", "url": "https://assets.claude.com/963373e433e489a87a10c823c52a0a013e9172dd.pdf", "type": "primary" }, { "label": "Claude 4 System Card", "url": "https://www.anthropic.com/claude-4-system-card", "type": "primary" }, { "label": "Claude Opus 5.5 System Card", "url": "https://www.anthropic.com/claude-opus-5-5-system-card", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/anthropic/claude-opus-4-1-20250805.toml", "type": "catalog" }, { "label": "Alias documentation", "url": "https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions", "type": "docs" } ], "aliases": [ "anthropic/claude-opus-4-1" ], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "First-party interpretability research supports Transformer structure for production Opus 4.1; dense/MoE topology and total/active parameter counts are undisclosed. Anthropic identifies it as an Opus 4 update, but publishes no nominal size for that predecessor in the cited card. The 200K context and input capabilities describe the released model; Anthropic documents retirement of this dated ID from its API on August 5, 2026. PDF input is a document format processed through text and vision. An Anthropic pricing PDF effective May 12, 2026 also lists a 1M API tier, conflicting with the retained 200K model specification; availability of that tier was not verified.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T10:33:52.696826Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/104.json", "alias_sources": { "anthropic/claude-opus-4-1": { "url": "https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions", "note": "Alias or alternate catalog name for the same released model; only the retained ID counts as a release." } } }, { "id": "openai/gpt-oss-120b", "slug": "openai--gpt-oss-120b", "name": "gpt-oss-120b", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-oss", "description": "Open-weight text reasoning model with 36 MoE layers, four active experts per token, and support for function calling.", "release_date": "2025-08-05", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 116.829156672, "active_parameters_b": 5.13, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-gpt-oss/", "source_type": "primary", "sources": [ { "label": "OpenAI original release", "url": "https://openai.com/index/introducing-gpt-oss/", "type": "primary" }, { "label": "Original OpenAI technical model card", "url": "https://arxiv.org/html/2508.10925v1", "type": "paper" }, { "label": "Pinned checkpoint configuration", "url": "https://huggingface.co/openai/gpt-oss-120b/blob/b5c939de8f754692c1647ca79fbf85e8c1e70f8a/config.json", "type": "primary" }, { "label": "Pinned original weight index", "url": "https://huggingface.co/openai/gpt-oss-120b/blob/b5c939de8f754692c1647ca79fbf85e8c1e70f8a/original/model.safetensors.index.json", "type": "weights" }, { "label": "Official reference model", "url": "https://github.com/openai/gpt-oss/blob/5681a6b6552635e7b29cde94a61ca6a7aee90ebd/gpt_oss/torch/model.py", "type": "primary" }, { "label": "Official MXFP4 loader", "url": "https://github.com/openai/gpt-oss/blob/5681a6b6552635e7b29cde94a61ca6a7aee90ebd/gpt_oss/torch/weights.py", "type": "primary" }, { "label": "Pinned Apache-2.0 license", "url": "https://huggingface.co/openai/gpt-oss-120b/blob/b5c939de8f754692c1647ca79fbf85e8c1e70f8a/LICENSE", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-oss-120b.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/openai/gpt-oss-120b", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Total is the complete logical checkpoint count, including untied embeddings, learned norms/sinks and all biases; MXFP4 packing and quantization scales do not add parameters. The official nominal 5.13B active count follows OpenAI’s per-token convention: four experts per layer and output head included, input embeddings excluded. Harmony formatting is required for the documented reasoning and tool behavior.", "field_status": {}, "audited_at": "2026-10-27T02:49:10.535481Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/446.json" }, { "id": "openai/gpt-oss-20b", "slug": "openai--gpt-oss-20b", "name": "gpt-oss-20b", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-oss", "description": "Open-weight text reasoning model with 32 experts per layer, four selected per token, and function calling in Harmony format.", "release_date": "2025-08-05", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 20.914757184, "active_parameters_b": 3.61, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-gpt-oss/", "source_type": "primary", "sources": [ { "label": "Introducing gpt-oss — August 5, 2025", "url": "https://openai.com/index/introducing-gpt-oss/", "type": "primary" }, { "label": "gpt-oss-120b & gpt-oss-20b Model Card — August 5, 2025", "url": "https://cdn.openai.com/pdf/419b6906-9da6-406c-a19d-1bb078ac7637/oai_gpt-oss_model_card.pdf", "type": "primary" }, { "label": "Pinned gpt-oss-20b model card", "url": "https://huggingface.co/openai/gpt-oss-20b/resolve/6cee5e81ee83917806bbde320786a8fb61efebee/README.md", "type": "weights" }, { "label": "Pinned gpt-oss-20b Transformers configuration", "url": "https://huggingface.co/openai/gpt-oss-20b/resolve/6cee5e81ee83917806bbde320786a8fb61efebee/config.json", "type": "weights" }, { "label": "Pinned gpt-oss-20b reference configuration", "url": "https://huggingface.co/openai/gpt-oss-20b/resolve/6cee5e81ee83917806bbde320786a8fb61efebee/original/config.json", "type": "weights" }, { "label": "Pinned OpenAI PyTorch reference model", "url": "https://raw.githubusercontent.com/openai/gpt-oss/5681a6b6552635e7b29cde94a61ca6a7aee90ebd/gpt_oss/torch/model.py", "type": "technical" }, { "label": "Pinned OpenAI MXFP4 checkpoint loader", "url": "https://raw.githubusercontent.com/openai/gpt-oss/5681a6b6552635e7b29cde94a61ca6a7aee90ebd/gpt_oss/torch/weights.py", "type": "technical" }, { "label": "Pinned full checkpoint shard index", "url": "https://huggingface.co/openai/gpt-oss-20b/resolve/6cee5e81ee83917806bbde320786a8fb61efebee/model.safetensors.index.json", "type": "weights" }, { "label": "Pinned checkpoint shard 00000 bounded tensor header", "url": "https://huggingface.co/openai/gpt-oss-20b/resolve/6cee5e81ee83917806bbde320786a8fb61efebee/model-00000-of-00002.safetensors", "type": "weights" }, { "label": "Pinned checkpoint shard 00001 bounded tensor header", "url": "https://huggingface.co/openai/gpt-oss-20b/resolve/6cee5e81ee83917806bbde320786a8fb61efebee/model-00001-of-00002.safetensors", "type": "weights" }, { "label": "Pinned checkpoint shard 00002 bounded tensor header", "url": "https://huggingface.co/openai/gpt-oss-20b/resolve/6cee5e81ee83917806bbde320786a8fb61efebee/model-00002-of-00002.safetensors", "type": "weights" }, { "label": "Pinned original checkpoint bounded tensor header", "url": "https://huggingface.co/openai/gpt-oss-20b/resolve/6cee5e81ee83917806bbde320786a8fb61efebee/original/model.safetensors", "type": "weights" }, { "label": "Pinned gpt-oss-20b license", "url": "https://huggingface.co/openai/gpt-oss-20b/resolve/6cee5e81ee83917806bbde320786a8fb61efebee/LICENSE", "type": "weights" }, { "label": "Current official gpt-oss model card publication", "url": "https://deploymentsafety.openai.com/gpt-oss", "type": "primary" }, { "label": "Current OpenAI gpt-oss-20b model reference", "url": "https://developers.openai.com/api/docs/models/gpt-oss-20b", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-oss-20b.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/openai/gpt-oss-20b", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete checkpoint count: 20,914,757,184 logical parameters, including separate input/output embeddings, norms, attention biases/sinks, routers and expert biases. MXFP4 packed weights are expanded; quantization scales are excluded. Official active count is rounded 3.61B per token per forward pass, including unembedding but excluding input embedding (model-card Table 1). Harmony format is required. Release: August 5, 2025; Apache 2.0 weights.", "field_status": {}, "audited_at": "2026-10-27T02:49:16.964809Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/447.json" }, { "id": "alibaba/qwen-flash", "slug": "alibaba--qwen-flash", "name": "Qwen Flash", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Hosted Qwen3 text model with optional thinking and a one-million-token context window.", "release_date": "2025-08-01", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1000000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://help.aliyun.com/en/model-studio/qwen-flash", "source_type": "primary", "sources": [ { "label": "Alibaba Cloud Qwen Flash model documentation", "url": "https://help.aliyun.com/en/model-studio/qwen-flash", "type": "primary" }, { "label": "Alibaba Cloud model release log", "url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "type": "primary" }, { "label": "Alibaba Cloud thinking-mode support", "url": "https://www.alibabacloud.com/help/en/model-studio/deep-thinking", "type": "primary" }, { "label": "Alibaba Cloud function-calling support", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen-function-calling", "type": "primary" }, { "label": "Alibaba Cloud commercial-edition classification", "url": "https://www.alibabacloud.com/help/en/model-studio/savings-plan-and-resource-package", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen-flash.toml", "type": "catalog" }, { "label": "Alibaba Cloud Qwen Flash international model documentation", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen-flash", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The qwen-flash alias is functionally equivalent to qwen-flash-2025-07-28. Official availability began August 1, 2025 (International/Global), August 5 (Beijing). Total/active parameters and architecture are undisclosed; no explicit named-base lineage supports a size inference. Later numbered Flash models are separate entries. Function calling is documented generally and for Beijing, but other exact-model regional tables mark it unsupported.", "field_status": { "license": "undisclosed", "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T07:28:27.235Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/012.json" }, { "id": "bytedance-seed/seed-1-6-vision", "slug": "bytedance-seed--seed-1-6-vision", "name": "Seed 1.6 Vision", "developer": "ByteDance", "developer_id": "bytedance-seed", "family": "seed", "description": "ByteDance Seed multimodal model for image understanding, visual reasoning, and tool-assisted tasks", "release_date": "2025-08", "date_precision": "month", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 256000, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/bytedance-seed/seed-1-6-vision.toml", "source_type": "catalog", "sources": [ { "label": "Volcengine 2025 release archive", "url": "https://docs.volcengine.com/docs/ark/model-release-announcement-2025?lang=zh", "type": "primary" }, { "label": "Volcengine legacy thinking support", "url": "https://docs.volcengine.com/docs/ark/model-deprecation-migration-guide?lang=en", "type": "primary" }, { "label": "Coze exact Vision 250815 specifications", "url": "https://docs.coze.cn/guides_model_release_note#77badcb7", "type": "primary" }, { "label": "ByteDance native tool-call example", "url": "https://raw.githubusercontent.com/bytedance/UI-TARS-desktop/main/multimodal/tarko/agent/examples/streaming/tool-calls.ts", "type": "primary" }, { "label": "Doubao model service agreement", "url": "https://docs.volcengine.com/docs/ark/doubao-model-service-agreement?lang=zh", "type": "primary" }, { "label": "302.AI exact-ID API specifications", "url": "https://nuxt.302.ai/product/detail/doubao-seed-1-6-vision-250815", "type": "primary" }, { "label": "ByteDance Seed1.6 nominal counts and named post-training variants", "url": "https://seed.bytedance.com/en/blog/introduction-to-techniques-used-in-seed1-6", "type": "primary" }, { "label": "Feishu model and version comparison", "url": "https://aily.feishu.cn/hc/1u7kleqg_897wr1b8", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/bytedance-seed/seed-1-6-vision.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Audited as doubao-seed-1-6-vision-250815. Public launch is verified to August 2025 only; the snapshot suffix is not treated as a release-day announcement. Supports switchable deep thinking, video understanding and native tools. Nominal parameter counts remain undisclosed: the June Seed1.6 230B-total/23B-active disclosure names Thinking and Adaptive variants, but no developer source maps Vision250815 to that base. Architecture remains undisclosed. Specifications describe the legacy model, not a tested current endpoint.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T10:42:47.803243+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/130.json" }, { "id": "alibaba/qwen3-coder-30b-a3b-instruct", "slug": "alibaba--qwen3-coder-30b-a3b-instruct", "name": "Qwen3-Coder 30B-A3B Instruct", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Qwen3 MoE coding model with 30.5B total and 3.3B active parameters, native 256K context, non-thinking mode and tool calling.", "release_date": "2025-07-31", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 30.5, "active_parameters_b": 3.3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct", "type": "weights" }, { "label": "Official model configuration", "url": "https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct/raw/main/config.json", "type": "docs" }, { "label": "Checkpoint license", "url": "https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct/blob/main/LICENSE", "type": "docs" }, { "label": "Alibaba Cloud release history", "url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "type": "docs" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-coder-30b-a3b-instruct.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "field_status": {}, "audited_at": "2026-10-26T14:26:06Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/031.json" }, { "id": "cohere/command-a-vision-07-2025", "slug": "cohere--command-a-vision-07-2025", "name": "Command A Vision", "developer": "Cohere", "developer_id": "cohere", "family": "command-a", "description": "Cohere vision model for multilingual document analysis, OCR, and image understanding", "release_date": "2025-07-31", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-4.0", "parameters_b": 112, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 128000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": false, "tool_call": false, "source_url": "https://docs.cohere.com/v2/changelog/2025-07-31-command-a-vision", "source_type": "primary", "sources": [ { "label": "Cohere Command A Vision release", "url": "https://docs.cohere.com/v2/changelog/2025-07-31-command-a-vision", "type": "primary" }, { "label": "Command A Vision architecture and training", "url": "https://huggingface.co/blog/CohereLabs/introducing-command-a-vision-07-2025", "type": "primary" }, { "label": "Exact Command A Vision model card", "url": "https://huggingface.co/CohereLabs/command-a-vision-07-2025", "type": "weights" }, { "label": "Command A Vision API documentation", "url": "https://docs.cohere.com/v2/docs/command-a-vision", "type": "primary" }, { "label": "Later Command A family comparison", "url": "https://cohere.com/blog/command-a-plus", "type": "primary" }, { "label": "Exact capability source", "url": "https://raw.githubusercontent.com/cohere-ai/cohere-developer-experience/main/fern/pages/models/the-command-family-of-models/command-a-vision.mdx", "type": "primary" }, { "label": "Capability renderer", "url": "https://raw.githubusercontent.com/cohere-ai/cohere-developer-experience/main/fern/components/model-showcase.tsx", "type": "primary" }, { "label": "Conflicting current license overview", "url": "https://cohere.com/models-overview", "type": "primary" }, { "label": "Checkpoint metadata and license gate", "url": "https://huggingface.co/api/models/CohereLabs/command-a-vision-07-2025", "type": "primary" }, { "label": "License terms linked by the exact checkpoint gate", "url": "https://cohere.com/cohere-labs-cc-by-nc-license", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/cohere/command-a-vision-07-2025.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/c4ai-command-a-vision-07-2025", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "112B is Cohere’s nominal complete-model total, including a 111B language tower. Active parameters are not separately disclosed for either scope. API context is 128,000 tokens; downloadable configuration defaults to 32K according to the card. License records the exact checkpoint card/gate declaration and linked public CC-BY-NC-4.0 terms with acceptable-use addendum; Cohere's current overview contradictorily says Apache 2.0 for the same repository. Relicensing was not verified. Native reasoning and tool use are explicitly disabled. The baseline c4ai-prefixed weights URL is not the release's canonical repository.", "parameter_components_b": { "language_model": 111 }, "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T12:27:55.783775+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/150.json" }, { "id": "mistral/codestral-2508", "slug": "mistral--codestral-2508", "name": "Codestral 25.08", "developer": "Mistral", "developer_id": "mistral", "family": "codestral", "description": "Code model for fill-in-the-middle completion and code generation, with chat, function calling and structured outputs.", "release_date": "2025-07-30", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 22, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": null, "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://mistral.ai/news/codestral-25-08/", "source_type": "primary", "sources": [ { "label": "Codestral 25.08 official model card", "url": "https://docs.mistral.ai/models/codestral-25-08.md", "type": "primary" }, { "label": "Codestral 25.08 release announcement", "url": "https://mistral.ai/news/codestral-25-08/", "type": "primary" }, { "label": "Codestral lifecycle and technical-documentation information", "url": "https://legal.mistral.ai/ai-governance/models/codestral/", "type": "primary" }, { "label": "Official Mistral Codestral public checkpoint listing", "url": "https://huggingface.co/api/models?author=mistralai&search=Codestral&limit=100", "type": "weights" }, { "label": "Mistral current reasoning support guide", "url": "https://docs.mistral.ai/studio/conversations/reasoning.md", "type": "primary" }, { "label": "Mistral legacy native reasoning guide", "url": "https://docs.mistral.ai/resources/deprecated/native-reasoning.md", "type": "primary" }, { "label": "Later Devstral 2 official model card", "url": "https://huggingface.co/mistralai/Devstral-2-123B-Instruct-2512/blob/main/README.md", "type": "weights" }, { "label": "Later Mistral Large 3 official model card", "url": "https://huggingface.co/mistralai/Mistral-Large-3-675B-Instruct-2512/blob/main/README.md", "type": "weights" }, { "label": "OpenRouter exact Codestral 2508 route metadata", "url": "https://openrouter.ai/api/v1/models/mistralai/codestral-2508/endpoints", "type": "primary" }, { "label": "Mistral FIM API reference", "url": "https://docs.mistral.ai/api/endpoint/fim", "type": "primary" }, { "label": "IBM watsonx exact codestral-2508 model specification", "url": "https://www.ibm.com/docs/en/watsonx/w-and-w/2.3.x?topic=models-third-party-foundation", "type": "primary" }, { "label": "Google Cloud Mistral partner-model documentation", "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/partner-models/mistral?hl=en", "type": "primary" }, { "label": "Microsoft Foundry partner-model documentation", "url": "https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/models-from-partners", "type": "primary" }, { "label": "Mistral Codestral 25.01 predecessor release", "url": "https://mistral.ai/news/codestral-2501/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/codestral-2508.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "IBM watsonx documents nominal 22B parameters for exact Codestral 2508; active parameters and dense/MoE structure remain undisclosed. The developer card lists 128k context (normalized to 128000; exact native boundary unverified), while OpenRouter advertises a 256000-token Mistral-backed route. Text input/output and function calling are explicit; dedicated native-thinking status remains unverified. Commercial VPC/on-prem deployment is offered, with no public exact-25.08 weights found.", "field_status": { "active_parameters_b": "undisclosed", "architecture": "undisclosed", "reasoning": "unverified" }, "audited_at": "2026-10-27T15:30:18.974584Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/313.json" }, { "id": "alibaba/qwen3-30b-a3b-instruct-2507", "slug": "alibaba--qwen3-30b-a3b-instruct-2507", "name": "Qwen3 30B-A3B Instruct 2507", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Updated compact Qwen3 MoE instruct model for multilingual chat, coding, and tool use", "release_date": "2025-07-29", "date_precision": "day", "open_weights": true, "license": "Apache 2.0", "parameters_b": 30.5, "active_parameters_b": 3.3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "source_type": "primary", "sources": [ { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-30b-a3b-instruct-2507.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-30B-A3B-Instruct-2507", "type": "weights" }, { "label": "Official model configuration", "url": "https://huggingface.co/Qwen/Qwen3-30B-A3B-Instruct-2507/blob/main/config.json", "type": "primary" }, { "label": "Official Qwen3 release news", "url": "https://github.com/QwenLM/Qwen3", "type": "primary" }, { "label": "Official release log", "url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Developer-reported counts are 30.5B total and 3.3B active. July 29 availability is explicit in the Alibaba release table; GitHub news followed July 30. Native context is 262144; 1M requires optional extension configuration.", "field_status": {}, "audited_at": "2026-10-26T14:22:40Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/028.json" }, { "id": "alibaba/qwen3-coder-flash", "slug": "alibaba--qwen3-coder-flash", "name": "Qwen3 Coder Flash", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Qwen3-based code generation model for repository understanding and multi-turn tool interaction.", "release_date": "2025-07-29", "date_precision": "day", "open_weights": true, "license": null, "parameters_b": 30.5, "active_parameters_b": 3.3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-coder-flash.toml", "source_type": "catalog", "sources": [ { "label": "Alibaba Cloud exact model card", "url": "https://docs.modelstudio.console.alibabacloud.com/en/model-studio/qwen3-coder-flash", "type": "primary" }, { "label": "Alibaba Cloud model release history", "url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "type": "primary" }, { "label": "Alibaba Cloud text-generation capabilities", "url": "https://www.alibabacloud.com/help/en/model-studio/text-generation-model/", "type": "primary" }, { "label": "Alibaba Cloud deep-thinking documentation", "url": "https://www.alibabacloud.com/help/en/model-studio/deep-thinking", "type": "primary" }, { "label": "Qwen Coder-Flash model identification", "url": "https://x.com/Alibaba_Qwen/status/1950925444057792808", "type": "primary" }, { "label": "Qwen identified model parameter specifications", "url": "https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct", "type": "primary" }, { "label": "Qwen Flash API announcement", "url": "https://x.com/Alibaba_Qwen/status/1952767585596145773", "type": "primary" }, { "label": "Qwen identified model configuration", "url": "https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct/raw/main/config.json", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=qwen3-coder-flash&limit=100", "label": "Official Qwen inventory search for the exact hosted model", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=Qwen3-Coder&limit=1000", "label": "Official Qwen Qwen3-Coder weight inventory", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-coder-flash.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "30.5B total and 3.3B activated are nominal counts inferred from Qwen's explicit identification of Coder-Flash as Qwen3-Coder-30B-A3B-Instruct, a MoE model. Alibaba still maps the API alias to its July 28 snapshot; Global/International service release was July 29, 2025. Qwen explicitly identifies Flash with the publicly downloadable Qwen3-Coder-30B-A3B-Instruct checkpoint. Official thinking-mode and regional function-calling documentation conflict.", "field_status": { "license": "undisclosed", "reasoning": "unverified" }, "audited_at": "2026-10-28T08:13:49.514266+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/033.json" }, { "id": "zhipuai/glm-4.5", "slug": "zhipuai--glm-4.5", "name": "GLM-4.5", "developer": "Z.ai", "developer_id": "zhipuai", "family": "glm", "description": "Text MoE model with switchable reasoning, tool calling and a 128K-token context window.", "release_date": "2025-07-28", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 355, "active_parameters_b": 32, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://z.ai/blog/glm-4.5", "source_type": "primary", "sources": [ { "label": "GLM-4.5 original release announcement", "url": "https://z.ai/blog/glm-4.5", "type": "primary" }, { "label": "GLM-4.5 technical report, architecture section and Table 1", "url": "https://arxiv.org/html/2508.06471v1", "type": "primary" }, { "label": "GLM-4.5 official checkpoint card at pinned revision", "url": "https://huggingface.co/zai-org/GLM-4.5/blob/cbb2c7cfb52fa128a9660cb1a7a78e017899e115/README.md", "type": "weights" }, { "label": "GLM-4.5 checkpoint configuration", "url": "https://huggingface.co/zai-org/GLM-4.5/blob/cbb2c7cfb52fa128a9660cb1a7a78e017899e115/config.json", "type": "weights" }, { "label": "GLM-4.5 checkpoint chat template", "url": "https://huggingface.co/zai-org/GLM-4.5/blob/cbb2c7cfb52fa128a9660cb1a7a78e017899e115/chat_template.jinja", "type": "weights" }, { "label": "Native GLM-4.5 API model documentation", "url": "https://docs.z.ai/guides/llm/glm-4.5", "type": "primary" }, { "label": "Official GLM-4.5 checkpoint repository metadata", "url": "https://huggingface.co/api/models/zai-org/GLM-4.5", "type": "weights" }, { "label": "Official GLM repository with later model comparison/download table", "url": "https://github.com/zai-org/GLM-4.5", "type": "primary" }, { "label": "Z.ai original corporate release distributed by PR Newswire", "url": "https://www.prnewswire.com/news-releases/zai-releases-glm-4-5--setting-new-standards-for-ai-performance-and-accessibility-while-improving-affordability-302514803.html", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/zhipuai/glm-4.5.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/GLM-4.5", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "GLM-4.5 launched July 28, 2025. Z.ai reports nominal 355B total and 32B activated parameters; its technical report explicitly includes the MTP layer but excludes word embeddings and the output layer. These scoped figures are not exact complete-checkpoint totals or measured inference-phase counts. The exact BF16 checkpoint supports 131072 positions; the native API advertises 128K context and a separate 96K maximum output. The model accepts and produces text, with switchable thinking and structured tool calls.", "parameters_scope": "nonembedding_with_mtp", "parameter_components_b": { "nonembedding_with_mtp": 355 }, "active_parameters_scope": "nonembedding_with_mtp", "active_parameters_by_phase_b": { "nonembedding_with_mtp": 32 }, "field_status": {}, "audited_at": "2026-10-27T06:00:42.073577Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/547.json" }, { "id": "zhipuai/glm-4.5-air", "slug": "zhipuai--glm-4.5-air", "name": "GLM-4.5-Air", "developer": "Z.ai", "developer_id": "zhipuai", "family": "glm-air", "description": "Text MoE model with thinking and non-thinking modes, tool calling and a 128K-token context window.", "release_date": "2025-07-28", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 106, "active_parameters_b": 12, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://z.ai/blog/glm-4.5", "source_type": "primary", "sources": [ { "label": "GLM-4.5 original release announcement", "url": "https://z.ai/blog/glm-4.5", "type": "primary" }, { "label": "Pinned GLM-4.5-Air model card", "url": "https://huggingface.co/zai-org/GLM-4.5-Air/blob/a24ceef6ce4f3536971efe9b778bdaa1bab18daa/README.md", "type": "weights" }, { "label": "GLM-4.5 technical report, Table 1", "url": "https://arxiv.org/html/2508.06471v1", "type": "primary" }, { "label": "Pinned GLM-4.5-Air configuration", "url": "https://huggingface.co/zai-org/GLM-4.5-Air/blob/a24ceef6ce4f3536971efe9b778bdaa1bab18daa/config.json", "type": "weights" }, { "label": "Pinned GLM-4.5-Air chat template", "url": "https://huggingface.co/zai-org/GLM-4.5-Air/blob/a24ceef6ce4f3536971efe9b778bdaa1bab18daa/chat_template.jinja", "type": "weights" }, { "label": "GLM-4.5 and Air API overview", "url": "https://docs.z.ai/guides/llm/glm-4.5", "type": "primary" }, { "label": "Official Air checkpoint inventory", "url": "https://huggingface.co/api/models/zai-org/GLM-4.5-Air", "type": "weights" }, { "label": "Pinned Air checkpoint tensor index", "url": "https://huggingface.co/zai-org/GLM-4.5-Air/blob/a24ceef6ce4f3536971efe9b778bdaa1bab18daa/model.safetensors.index.json", "type": "weights" }, { "label": "Z.ai original corporate release distributed by PR Newswire", "url": "https://www.prnewswire.com/news-releases/zai-releases-glm-4-5--setting-new-standards-for-ai-performance-and-accessibility-while-improving-affordability-302514803.html", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/zhipuai/glm-4.5-air.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/zai-org/GLM-4.5-Air", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The author-reported 106B total and 12B active counts include MTP layers but exclude word embeddings and the output layer, as specified in the technical report. They are nominal scoped counts, not an exact complete-checkpoint tally. The Air model has one dense and 45 MoE main layers, plus one MTP layer; eight of 128 routed experts and one shared expert are used per MoE layer. Context capacity is 131072 tokens; API output limits are separate.", "parameters_scope": "nonembedding_with_mtp", "parameter_components_b": { "nonembedding_with_mtp": 106 }, "active_parameters_scope": "nonembedding_with_mtp", "active_parameters_by_phase_b": { "nonembedding_with_mtp": 12 }, "field_status": {}, "audited_at": "2026-10-27T06:00:57.363377+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/548.json" }, { "id": "zhipuai/glm-4.5-flash", "slug": "zhipuai--glm-4.5-flash", "name": "GLM-4.5-Flash", "developer": "Z.ai", "developer_id": "zhipuai", "family": "glm-flash", "description": "Hosted GLM text model with optional reasoning and function calling.", "release_date": "2025-07-28", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": null, "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://docs.bigmodel.cn/cn/guide/models/free/glm-4.5-flash", "source_type": "primary", "sources": [ { "label": "GLM-4.5-Flash official model guide", "url": "https://docs.bigmodel.cn/cn/guide/models/free/glm-4.5-flash", "type": "primary" }, { "label": "July 28, 2025 archived original GLM-4.5-Flash launch guide", "url": "https://web.archive.org/web/20250728123929/https://docs.bigmodel.cn/cn/guide/models/free/glm-4.5-flash", "type": "primary" }, { "label": "BigModel model and product release log", "url": "https://docs.bigmodel.cn/cn/update/new-releases", "type": "primary" }, { "label": "GLM-4.5 official API family guide", "url": "https://docs.z.ai/guides/llm/glm-4.5", "type": "primary" }, { "label": "GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models", "url": "https://arxiv.org/html/2508.06471v1", "type": "primary" }, { "label": "Official Z.ai public model repository inventory", "url": "https://huggingface.co/api/models?author=zai-org&limit=1000", "type": "weights" }, { "label": "Official GLM-4.5 model card", "url": "https://huggingface.co/zai-org/GLM-4.5/blob/main/README.md", "type": "weights" }, { "label": "Later GLM-4.7-Flash official model card", "url": "https://huggingface.co/zai-org/GLM-4.7-Flash/blob/main/README.md", "type": "weights" }, { "label": "Z.ai Terms of Use and API service terms", "url": "https://docs.z.ai/legal-agreement/terms-of-use", "type": "primary" }, { "label": "Original Dify Z.ai integration for exact glm-4.5-flash identifier", "url": "https://github.com/langgenius/dify-official-plugins/blob/0d33548940ad41cc72f978a831dc660a0f9c148d/models/z_ai/models/llm/glm-4.5-flash.yaml", "type": "primary" }, { "label": "GLM-4.7-Flash open and free: January 19, 2026 original release notice", "url": "https://www.zhipuai.cn/zh/news/148", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/zhipuai/glm-4.5-flash.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Original GLM-4.5-Flash was publicly documented on July 28, 2025. Total and active parameters remain undisclosed; no named-base relationship establishes inheritance from standard GLM-4.5 or Air. Architecture is also undisclosed. Free API access does not establish open weights. BigModel announced January 30, 2026 retirement and rerouting to GLM-4.7-Flash; that successor’s 30B/3B counts do not describe the original model. The notice is specific to BigModel.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T18:37:24.927093Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/549.json" }, { "id": "alibaba/qwen3-235b-a22b-thinking-2507", "slug": "alibaba--qwen3-235b-a22b-thinking-2507", "name": "Qwen3 235B-A22B Thinking 2507", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Updated large open Qwen3 MoE thinking model for reasoning, math, coding, and tool use", "release_date": "2025-07-25", "date_precision": "day", "open_weights": true, "license": "Apache 2.0", "parameters_b": 235, "active_parameters_b": 22, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507", "type": "weights" }, { "label": "Qwen checkpoint configuration", "url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507/raw/main/config.json", "type": "primary" }, { "label": "Qwen release news", "url": "https://github.com/QwenLM/Qwen3", "type": "primary" }, { "label": "Hugging Face weights", "url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507/tree/main", "type": "weights" }, { "label": "Checkpoint license", "url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507/blob/main/LICENSE", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-235b-a22b-thinking-2507.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "field_status": {}, "audited_at": "2026-10-26T14:19:06Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/026.json" }, { "id": "nvidia/llama-3.3-nemotron-super-49b-v1.5", "slug": "nvidia--llama-3.3-nemotron-super-49b-v1.5", "name": "Llama 3.3 Nemotron Super 49B v1.5", "developer": "NVIDIA", "developer_id": "nvidia", "family": "nemotron", "description": "NVIDIA’s NAS-derived Llama 3.3 model, updated in v1.5 with supervised fine-tuning and reinforcement learning for reasoning, chat and tool calling.", "release_date": "2025-07-25", "date_precision": "day", "open_weights": true, "license": "NVIDIA Open Model License + Llama 3.3 Community License", "parameters_b": 49.867145216, "active_parameters_b": 49, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1_5/blob/420ba7d28211abf116b8b103ab700d92619daf98/README.md", "source_type": "primary", "sources": [ { "label": "Exact v1.5 model card", "url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1_5/blob/420ba7d28211abf116b8b103ab700d92619daf98/README.md", "type": "primary" }, { "label": "Exact v1.5 checkpoint configuration", "url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1_5/resolve/420ba7d28211abf116b8b103ab700d92619daf98/config.json", "type": "weights" }, { "label": "NVIDIA checkpoint implementation", "url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1_5/resolve/420ba7d28211abf116b8b103ab700d92619daf98/modeling_decilm.py", "type": "weights" }, { "label": "Exact BF16 checkpoint tensor index", "url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1_5/resolve/420ba7d28211abf116b8b103ab700d92619daf98/model.safetensors.index.json", "type": "weights" }, { "label": "Official safetensors headers, first of 21 indexed shards", "url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1_5/resolve/420ba7d28211abf116b8b103ab700d92619daf98/model-00001-of-00021.safetensors", "type": "weights" }, { "label": "Exact v1.5 tokenizer and chat template", "url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1_5/resolve/420ba7d28211abf116b8b103ab700d92619daf98/tokenizer_config.json", "type": "weights" }, { "label": "NVIDIA tool-call parser", "url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1_5/resolve/420ba7d28211abf116b8b103ab700d92619daf98/llama_nemotron_toolcall_parser_no_streaming.py", "type": "weights" }, { "label": "NVIDIA v1.5 release article", "url": "https://developer.nvidia.com/blog/build-more-accurate-and-efficient-ai-agents-with-the-new-nvidia-llama-nemotron-super-v1-5/", "type": "primary" }, { "label": "NVIDIA Open Model License Agreement", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/", "type": "primary" }, { "label": "Llama 3.3 Community License Agreement", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_3/LICENSE", "type": "primary" }, { "label": "Llama-Nemotron technical report, September 2025 revision", "url": "https://arxiv.org/html/2505.00949v5", "type": "primary" }, { "label": "Vultr Nemotron Super 49B deployment and HGX B200 benchmarks", "url": "https://docs.vultr.com/inference-cookbook/cuda/model-guides/nemotron-super-49b", "type": "primary" }, { "label": "NVIDIA exact v1.5 FP8 model card", "url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1_5-FP8/blob/04822723e77e036ddf2d24e83c6d469d3b009252/README.md", "type": "primary" }, { "label": "NVIDIA exact v1.5 FP8 checkpoint configuration", "url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1_5-FP8/resolve/04822723e77e036ddf2d24e83c6d469d3b009252/config.json", "type": "weights" }, { "label": "NVIDIA v1.5 FP8 quantization configuration", "url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1_5-FP8/resolve/04822723e77e036ddf2d24e83c6d469d3b009252/hf_quant_config.json", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/nvidia/llama-3.3-nemotron-super-49b-v1.5.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Total is the exact 49.867145216B BF16 checkpoint count; 49B is nominal. Active 49B follows Vultr’s nominal deployment convention for v1.5 FP8, whose architecture matches BF16; NVIDIA does not separately disclose an active count. The NAS architecture has 49 attention blocks, 31 attention skips and 80 variable-width dense FFNs. Reasoning defaults on; /no_think disables it. Both NVIDIA and Llama 3.3 licenses apply.", "field_status": {}, "audited_at": "2026-10-27T15:57:18.286190+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/378.json" }, { "id": "mistral/magistral-medium-2507", "slug": "mistral--magistral-medium-2507", "name": "Magistral Medium 1.1", "developer": "Mistral", "developer_id": "mistral", "family": "magistral", "description": null, "release_date": "2025-07-24", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 40000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://docs.mistral.ai/models/magistral-medium-1-1-25-07", "source_type": "primary", "sources": [ { "label": "Official version-specific model documentation", "url": "https://docs.mistral.ai/models/magistral-medium-1-1-25-07", "type": "docs" }, { "label": "Dated publisher API changelog", "url": "https://docs.mistral.ai/resources/changelogs", "type": "announcement" }, { "label": "Publisher model lifecycle", "url": "https://legal.mistral.ai/ai-governance/models/magistral-medium-1/", "type": "docs" } ], "aliases": [], "first_seen": "2026-10-29T10:46:35.789Z", "last_seen": "2026-10-29T10:46:35.789Z", "upstream_present": false, "availability": "public", "notes": "Original Mistral reasoning model served through its API; this version is now retired or deprecated. Parameter count and internal architecture are undisclosed. Context is the publisher’s rounded 40k API figure. " }, { "id": "mistral/magistral-small-2507", "slug": "mistral--magistral-small-2507", "name": "Magistral Small 1.1", "developer": "Mistral", "developer_id": "mistral", "family": "magistral", "description": null, "release_date": "2025-07-24", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 24, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 40960, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://huggingface.co/mistralai/Magistral-Small-2507", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/mistralai/Magistral-Small-2507/blob/b3583a426ca23186681c4297926dfe9219be6161/README.md", "type": "model_card" }, { "label": "Official native configuration", "url": "https://huggingface.co/mistralai/Magistral-Small-2507/blob/b3583a426ca23186681c4297926dfe9219be6161/params.json", "type": "config" }, { "label": "Dated publisher API changelog", "url": "https://docs.mistral.ai/resources/changelogs", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:46:35.789Z", "last_seen": "2026-10-29T10:46:35.789Z", "upstream_present": false, "availability": "public", "notes": "Mistral trained this reasoning update from its own Small 3.1 model. The card gives a nominal 24B parameters and recommends a 40k limit despite describing 128k capability; the native configuration sets 40960 positions, which is recorded here." }, { "id": "alibaba/qwen3-coder-480b-a35b-instruct", "slug": "alibaba--qwen3-coder-480b-a35b-instruct", "name": "Qwen3-Coder 480B-A35B Instruct", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Open-weight Qwen MoE model for agentic coding and repository-scale understanding.", "release_date": "2025-07-22", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 480, "active_parameters_b": 35, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://qwenlm.github.io/blog/qwen3-coder/", "source_type": "primary", "sources": [ { "label": "Qwen launch announcement", "url": "https://qwenlm.github.io/blog/qwen3-coder/", "type": "primary" }, { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct", "type": "weights" }, { "label": "Qwen checkpoint configuration", "url": "https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct/raw/main/config.json", "type": "primary" }, { "label": "Qwen checkpoint license", "url": "https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct/blob/main/LICENSE", "type": "primary" }, { "label": "Alibaba release announcement", "url": "https://www.alibabacloud.com/en/press-room/alibaba-unveils-cutting-edge-ai-coding-model-qwen3", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-coder-480b-a35b-instruct.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "480B total and 35B activated per token are developer-reported nominal counts. Native context is 262,144 tokens; 1M requires extrapolation. Non-thinking output mode.", "field_status": {}, "audited_at": "2026-10-26T14:25:56Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/032.json" }, { "id": "alibaba/qwen3-coder-plus", "slug": "alibaba--qwen3-coder-plus", "name": "Qwen3 Coder Plus", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Hosted Qwen3 coding model for code generation and tool-driven programming.", "release_date": "2025-07-22", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": 480, "active_parameters_b": 35, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-coder-plus.toml", "source_type": "catalog", "sources": [ { "label": "Alibaba model lifecycle and release dates", "url": "https://help.aliyun.com/en/model-studio/newly-released-models", "type": "primary" }, { "label": "Qwen3-Coder launch announcement", "url": "https://qwenlm.github.io/blog/qwen3-coder/", "type": "primary" }, { "label": "Historical official Qwen3-Coder-Plus model card (indexed)", "url": "https://docs.modelstudio.console.alibabacloud.com/id/model-studio/qwen3-coder-plus", "type": "primary" }, { "label": "Alibaba text generation capability table", "url": "https://www.alibabacloud.com/help/en/model-studio/text-generation-model/", "type": "primary" }, { "label": "Alibaba Dataphin model capabilities", "url": "https://help.aliyun.com/zh/dataphin/semimanaged-v4/user-guide/untitled-document-1753169796047", "type": "primary" }, { "label": "DataWorks model retirement notice", "url": "https://www.alibabacloud.com/help/en/dataworks/announcement-on-model-retirement-for-dataworks-data-agent", "type": "primary" }, { "label": "Qwen3-Coder-480B-A35B official parameter specification", "url": "https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct", "type": "primary" }, { "label": "Alibaba Coder Plus snapshot update and lineage", "url": "https://www.alibabacloud.com/en/notice/model_studio_update_of_qwen3coderplus_5c3?_p_lc=1", "type": "primary" }, { "label": "Alibaba regional Coder Plus alias mapping", "url": "https://www.alibabacloud.com/en/notice/model_studio_update_of_qwen3coderplus_60c?_p_lc=1", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=qwen3-coder-plus&limit=100", "label": "Official Qwen inventory search for the exact hosted model", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=Qwen3-Coder&limit=1000", "label": "Official Qwen Qwen3-Coder weight inventory", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-coder-plus.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Nominal 480B total / 35B active MoE inferred from Qwen’s explicit launch mapping to Qwen3-Coder-480B-A35B-Instruct and the documented September snapshot update; no structural change is disclosed, but the later snapshot has no separate count publication. The current hosted September revision has no identified public weight release; its weight license remains undisclosed. Context and modalities use historical official documentation; current pages omit or redirect this alias, and present endpoint operation is unverified. Official sources conflict on thinking mode and regional tool-calling support. Release date is earliest China availability; international/global service followed July 23. Classified as closed weights: no public release of this hosted variant was found in the developer documentation and weight inventory.", "field_status": { "license": "undisclosed", "reasoning": "unverified" }, "audited_at": "2026-10-28T08:13:49.514266+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/035.json" }, { "id": "alibaba/qwen3-235b-a22b-instruct-2507", "slug": "alibaba--qwen3-235b-a22b-instruct-2507", "name": "Qwen3 235B-A22B Instruct 2507", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Updated large open Qwen3 MoE instruct model for multilingual chat, coding, and tool use", "release_date": "2025-07-21", "date_precision": "day", "open_weights": true, "license": "Apache 2.0", "parameters_b": 235, "active_parameters_b": 22, "parameters_source": "primary_source", "architecture": "moe", "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Instruct-2507", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Instruct-2507", "type": "weights" }, { "label": "Qwen checkpoint configuration", "url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Instruct-2507/raw/main/config.json", "type": "primary" }, { "label": "Qwen official release news", "url": "https://github.com/QwenLM/Qwen3", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-235b-a22b-instruct-2507.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "field_status": {}, "audited_at": "2026-10-26T14:18:56Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/025.json" }, { "id": "mistral/voxtral-mini-3b-2507", "slug": "mistral--voxtral-mini-3b-2507", "name": "Voxtral Mini 3B 2507", "developer": "Mistral", "developer_id": "mistral", "family": "voxtral", "description": "Open audio-language model for speech transcription, audio understanding, and voice-driven tool use", "release_date": "2025-07-15", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 4.7, "active_parameters_b": 4, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text", "audio" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://mistral.ai/news/voxtral/", "source_type": "primary", "sources": [ { "label": "Voxtral launch announcement", "url": "https://mistral.ai/news/voxtral/", "type": "primary" }, { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Voxtral-Mini-3B-2507", "type": "weights" }, { "label": "Voxtral technical report, sections 2 and 3 and Table 1", "url": "https://arxiv.org/html/2507.13264v1", "type": "primary" }, { "label": "Official Voxtral Mini 25.07 documentation schema, pinned revision", "url": "https://github.com/mistralai/platform-docs-public/blob/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/schema/models/models/voxtral-mini-25-07.ts", "type": "primary" }, { "label": "Official model-card weights table renderer, pinned revision", "url": "https://github.com/mistralai/platform-docs-public/blob/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/app/%5Blocale%5D/(docs)/model-card/%5Bslug%5D/components/weights-table.tsx", "type": "primary" }, { "label": "Native Voxtral Mini 3B 2507 configuration", "url": "https://huggingface.co/mistralai/Voxtral-Mini-3B-2507/blob/3060fe34b35ba5d44202ce9ff3c097642914f8f3/params.json", "type": "weights" }, { "label": "Transformers conversion configuration for Voxtral Mini 3B 2507", "url": "https://huggingface.co/mistralai/Voxtral-Mini-3B-2507/blob/3060fe34b35ba5d44202ce9ff3c097642914f8f3/config.json", "type": "weights" }, { "label": "Official separate Voxtral Mini Transcribe 25.07 documentation schema", "url": "https://github.com/mistralai/platform-docs-public/blob/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/schema/models/models/voxtral-mini-transcribe-25-07.ts", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/voxtral-mini-3b-2507.toml", "type": "catalog" }, { "label": "Exact Voxtral Mini checkpoint tensor index", "url": "https://huggingface.co/mistralai/Voxtral-Mini-3B-2507/blob/3060fe34b35ba5d44202ce9ff3c097642914f8f3/model.safetensors.index.json", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The 3B name is not the full multimodal checkpoint count: the technical report gives 4.7B total. The later official weights table instead lists 4B total and 4B active without defining component or inference-phase scope; 4B active is retained only as that published nominal figure. Native context is 32768 tokens (about 30 minutes transcription or 40 minutes audio understanding), despite 131072 in the converted Transformers text configuration. The public audio-chat checkpoint is distinct from Voxtral Mini Transcribe, which shares an API identifier. Dedicated thinking-mode support remains unverified. Component figures are independently rounded in the technical report and are not an exact-sum reconstruction.", "parameter_components_b": { "audio_encoder": 0.64, "audio_adapter": 0.025, "text_embeddings": 0.4, "language_decoder": 3.6 }, "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-26T23:40:58.845360Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/351.json" }, { "id": "mistral/voxtral-small-24b-2507", "slug": "mistral--voxtral-small-24b-2507", "name": "Voxtral Small 24B 2507", "developer": "Mistral", "developer_id": "mistral", "family": "voxtral", "description": "Open audio-language model for speech transcription, audio understanding, and voice-driven tool use", "release_date": "2025-07-15", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 24.3, "active_parameters_b": 24, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text", "audio" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://mistral.ai/news/voxtral/", "source_type": "primary", "sources": [ { "label": "Voxtral release announcement, July 15, 2025", "url": "https://mistral.ai/news/voxtral/", "type": "primary" }, { "label": "Voxtral Small 1.0 (24B) - 2507 model card", "url": "https://huggingface.co/mistralai/Voxtral-Small-24B-2507/blob/da5b42409f279fdd92febee0511a6c32828569c1/README.md", "type": "weights" }, { "label": "Voxtral technical report, Table 1 and sections 1–2", "url": "https://arxiv.org/html/2507.13264v1", "type": "primary" }, { "label": "Voxtral Small native configuration", "url": "https://huggingface.co/mistralai/Voxtral-Small-24B-2507/blob/da5b42409f279fdd92febee0511a6c32828569c1/params.json", "type": "weights" }, { "label": "Voxtral Small Transformers conversion configuration", "url": "https://huggingface.co/mistralai/Voxtral-Small-24B-2507/blob/da5b42409f279fdd92febee0511a6c32828569c1/config.json", "type": "weights" }, { "label": "Official Voxtral Small 25.07 documentation source", "url": "https://github.com/mistralai/platform-docs-public/blob/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/schema/models/models/voxtral-small-25-07.ts", "type": "primary" }, { "label": "Official model weights table renderer", "url": "https://github.com/mistralai/platform-docs-public/blob/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/app/%5Blocale%5D/(docs)/models/model-cards/%5Bslug%5D/components/weights-table.tsx", "type": "primary" }, { "label": "Native Voxtral Small checkpoint tensor header", "url": "https://huggingface.co/mistralai/Voxtral-Small-24B-2507/resolve/da5b42409f279fdd92febee0511a6c32828569c1/consolidated.safetensors", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/voxtral-small-24b-2507.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Voxtral-Small-24B-2507", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Technical report Table 1 reports a complete 24.3B checkpoint (rounded); its independently rounded components are 0.640B audio encoder, 0.052B adapter, 0.670B text embeddings and 22.9B language decoder. Native BF16 header shapes total 24.259880960B. The official documentation independently labels 24B active, a nominal figure without modality-phase detail. Retain native/documented 32,768-token context despite the converted Transformers config listing 131,072. Function calling is native; the model card calls its example experimental. Dedicated reasoning-mode classification remains unverified.", "parameter_components_b": { "audio_encoder": 0.64, "audio_adapter": 0.052, "text_embeddings": 0.67, "language_decoder": 22.9 }, "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-26T23:44:35.909765+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/352.json" }, { "id": "moonshotai/kimi-k2", "slug": "moonshotai--kimi-k2", "name": "Kimi K2 0711", "developer": "Moonshot AI", "developer_id": "moonshotai", "family": "kimi-k2", "description": "Original July 2025 Kimi K2 Instruct: an open-weight text MoE Transformer with MLA, 128K context, and native tool calling for chat, coding, and agentic tasks; explicitly released without long thinking.", "release_date": "2025-07-11", "date_precision": "day", "open_weights": true, "license": "Modified MIT", "parameters_b": 1040, "active_parameters_b": 32.6, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct", "source_type": "primary", "sources": [ { "label": "Kimi-K2-Instruct model card", "url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct", "type": "primary" }, { "label": "Kimi K2: Open Agentic Intelligence, section 2.3 and Table 2 (v2)", "url": "https://arxiv.org/html/2507.20534v2#S2.SS3", "type": "primary" }, { "label": "Kimi K2: Open Agentic Intelligence, original July 2025 technical report", "url": "https://arxiv.org/html/2507.20534v1#S2.SS3", "type": "primary" }, { "label": "Original Kimi-K2-Instruct checkpoint configuration", "url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct/blob/fd1984e2b7a3350dbf7305fe73a4ede25c14de50/config.json", "type": "primary" }, { "label": "Kimi-K2-Instruct official model implementation", "url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct/blob/fd1984e2b7a3350dbf7305fe73a4ede25c14de50/modeling_deepseek.py", "type": "primary" }, { "label": "Original Kimi-K2-Instruct safetensors checkpoint index", "url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct/blob/fd1984e2b7a3350dbf7305fe73a4ede25c14de50/model.safetensors.index.json", "type": "primary" }, { "label": "Public metadata for the original Kimi-K2-Instruct repository", "url": "https://huggingface.co/api/models/moonshotai/Kimi-K2-Instruct", "type": "primary" }, { "label": "Kimi-K2-Instruct Modified MIT License", "url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct/blob/fd1984e2b7a3350dbf7305fe73a4ede25c14de50/LICENSE", "type": "primary" }, { "label": "Kimi research publication index", "url": "https://www.kimi.com/en/blog/", "type": "primary" }, { "label": "Kimi K2: Open Agentic Intelligence launch announcement", "url": "https://www.kimi.com/en/blog/kimi-k2", "type": "primary" }, { "label": "Kimi K2 official tool calling guide", "url": "https://github.com/MoonshotAI/Kimi-K2/blob/main/docs/tool_call_guidance.md", "type": "primary" }, { "label": "Kimi API model list and deprecations", "url": "https://platform.kimi.ai/docs/models", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/moonshotai/kimi-k2.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Counts are nominal: the original Kimi K2 technical report (Table 2, also retained in v2) specifies 1.04T total and 32.6B activated; the card uses 1T/32B. These are not tensor-exact checkpoint counts. FP8 scale tensors, rotary buffers and checkpoint byte sizes were not interpreted as learned parameters. Original July Instruct supports 128K context and native tool calling, without long thinking. Modified MIT includes a prominent-name condition above 100M monthly active users or US$20M monthly revenue. Public weights remain available; the original 0711 hosted API is listed as deprecated.", "field_status": {}, "audited_at": "2026-10-26T23:47:34.094255Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/354.json" }, { "id": "moonshotai/kimi-k2-instruct", "slug": "moonshotai--kimi-k2-instruct", "name": "Kimi K2 Instruct", "developer": "Moonshot AI", "developer_id": "moonshotai", "family": "kimi-k2", "description": "Non-thinking Kimi K2 MoE instruction model for coding and tool-using agents", "release_date": "2025-07-11", "date_precision": "day", "open_weights": true, "license": "Modified MIT", "parameters_b": 1040, "active_parameters_b": 32.6, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct", "type": "weights" }, { "label": "July Kimi K2 technical report, Table 2", "url": "https://arxiv.org/html/2507.20534v1", "type": "primary" }, { "label": "February 2026 report revision, Table 2", "url": "https://arxiv.org/html/2507.20534v2", "type": "primary" }, { "label": "Original Instruct configuration", "url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct/blob/fd1984e2b7a3350dbf7305fe73a4ede25c14de50/config.json", "type": "primary" }, { "label": "Original Instruct license", "url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct/blob/fd1984e2b7a3350dbf7305fe73a4ede25c14de50/LICENSE", "type": "primary" }, { "label": "Kimi K2 release announcement", "url": "https://www.kimi.com/en/blog/kimi-k2", "type": "primary" }, { "label": "Official dated release index", "url": "https://platform.kimi.com/blog", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/moonshotai/kimi-k2-instruct.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Original July 11, 2025 Instruct checkpoint, distinct from Instruct-0905, K2 Thinking and K2.5. July technical-report Table 2, unchanged in the February 2026 revision, reports 1.04T total and 32.6B active: rounded published figures, versus nominal 1T/32B in the card, not exact tensor counts. Table 4's 1043B is explicitly for Base and is not transferred. Original context is 131072; reasoning=false means no dedicated long-thinking mode. Modified MIT includes a Kimi K2 UI-branding condition above 100M monthly active users or US$20M monthly revenue.", "field_status": {}, "audited_at": "2026-10-26T23:51:04.441698Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/356.json" }, { "id": "liquid/lfm2-1.2b", "slug": "liquid--lfm2-1.2b", "name": "LFM2-1.2B", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm", "description": null, "release_date": "2025-07-10", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 1.170340608, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "hybrid", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://www.liquid.ai/blog/liquid-foundation-models-v2-our-second-series-of-generative-ai-models", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/liquid-foundation-models-v2-our-second-series-of-generative-ai-models", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/LiquidAI/LFM2-1.2B/blob/40f3da0d0164913923aee9462c23077868b816a3/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/LiquidAI/LFM2-1.2B/blob/40f3da0d0164913923aee9462c23077868b816a3/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Post-trained dense hybrid convolution/attention model. Exact total and 32,768-token supported context are from the developer model card; a larger configuration limit is not substituted." }, { "id": "liquid/lfm2-350m", "slug": "liquid--lfm2-350m", "name": "LFM2-350M", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm", "description": null, "release_date": "2025-07-10", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 0.354483968, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "hybrid", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://www.liquid.ai/blog/liquid-foundation-models-v2-our-second-series-of-generative-ai-models", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/liquid-foundation-models-v2-our-second-series-of-generative-ai-models", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/LiquidAI/LFM2-350M/blob/f37d3f5c8c5484bc01dad379a595cf4c68c4e70e/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/LiquidAI/LFM2-350M/blob/f37d3f5c8c5484bc01dad379a595cf4c68c4e70e/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Post-trained dense hybrid convolution/attention model. Exact total and 32,768-token supported context are from the developer model card; a larger configuration limit is not substituted." }, { "id": "liquid/lfm2-700m", "slug": "liquid--lfm2-700m", "name": "LFM2-700M", "developer": "Liquid AI", "developer_id": "liquid", "family": "lfm", "description": null, "release_date": "2025-07-10", "date_precision": "day", "open_weights": true, "license": "LFM Open License v1.0", "parameters_b": 0.742489344, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "hybrid", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://www.liquid.ai/blog/liquid-foundation-models-v2-our-second-series-of-generative-ai-models", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://www.liquid.ai/blog/liquid-foundation-models-v2-our-second-series-of-generative-ai-models", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/LiquidAI/LFM2-700M/blob/86f49fc9a3800c3a325b7320bde179c318062583/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/LiquidAI/LFM2-700M/blob/86f49fc9a3800c3a325b7320bde179c318062583/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Post-trained dense hybrid convolution/attention model. Exact total and 32,768-token supported context are from the developer model card; a larger configuration limit is not substituted." }, { "id": "mistral/devstral-medium-2507", "slug": "mistral--devstral-medium-2507", "name": "Devstral Medium 1.0", "developer": "Mistral", "developer_id": "mistral", "family": "devstral", "description": "Text model for software-engineering agents, with function calling for codebase exploration and multi-file edits.", "release_date": "2025-07-10", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 48, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": null, "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://mistral.ai/news/devstral-2507/", "source_type": "primary", "sources": [ { "label": "Upgrading agentic coding capabilities with the new Devstral models", "url": "https://mistral.ai/news/devstral-2507/", "type": "primary" }, { "label": "Devstral Medium 1.0", "url": "https://docs.mistral.ai/models/devstral-medium-1-0-25-07", "type": "primary" }, { "label": "Mistral documentation model catalog used by the model-selection guide", "url": "https://docs.mistral.ai/_next/static/chunks/2894-884782bf479f4ba0.js", "type": "primary" }, { "label": "Devstral Medium 1 legal model record", "url": "https://legal.mistral.ai/ai-governance/models/devstral-medium-1/", "type": "primary" }, { "label": "Third-party foundation models — watsonx.ai 2.4.x, devstral-medium-2507", "url": "https://www.ibm.com/docs/en/watsonx/w-and-w/2.4.x?topic=models-third-party-foundation", "type": "primary" }, { "label": "Third-party foundation models — watsonx.ai 2.3.x, devstral-medium-2507", "url": "https://www.ibm.com/docs/en/watsonx/w-and-w/2.3.x?topic=models-third-party-foundation", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/devstral-medium-2507.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Mistral developed this release in collaboration with All Hands AI. Public weights were not released; private enterprise deployment and customization were offered. IBM, the licensed deployment partner, specifies a nominal 48B total for this exact release; Mistral itself does not publish a count in the reviewed materials. Active parameters and dense-versus-MoE architecture remain undisclosed. Context is the developer-advertised 128k, normalized to 128000; the exact token boundary is unverified. Dedicated native-thinking status is unverified. Developer docs mark this API revision deprecated, while the legal lifecycle page still labels it Active; current endpoint operation was not verified.", "field_status": { "active_parameters_b": "undisclosed", "architecture": "undisclosed", "reasoning": "unverified" }, "audited_at": "2026-10-27T15:31:00.419937Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/315.json" }, { "id": "mistral/devstral-small-2507", "slug": "mistral--devstral-small-2507", "name": "Devstral Small 1.1", "developer": "Mistral", "developer_id": "mistral", "family": "devstral", "description": "Devstral Small 1.1 is a text-only coding model developed by Mistral AI with All Hands AI for codebase exploration, file editing and software engineering agents.", "release_date": "2025-07-10", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 24, "active_parameters_b": 24, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://mistral.ai/news/devstral-2507/", "source_type": "primary", "sources": [ { "label": "Upgrading agentic coding capabilities with the new Devstral models", "url": "https://mistral.ai/news/devstral-2507/", "type": "primary" }, { "label": "Devstral Small 1.1 — model documentation and Weights table", "url": "https://docs.mistral.ai/models/devstral-small-1-1-25-07", "type": "primary" }, { "label": "Devstral Small 1.1 — pinned official model card", "url": "https://huggingface.co/mistralai/Devstral-Small-2507/blob/bd165ab26cebbcc2eea2c4ecbfc07f3ac42b3c39/README.md", "type": "weights" }, { "label": "Devstral-Small-2507 — pinned configuration", "url": "https://huggingface.co/mistralai/Devstral-Small-2507/blob/bd165ab26cebbcc2eea2c4ecbfc07f3ac42b3c39/config.json", "type": "weights" }, { "label": "Devstral-Small-2507 — pinned checkpoint tensor index", "url": "https://huggingface.co/mistralai/Devstral-Small-2507/blob/bd165ab26cebbcc2eea2c4ecbfc07f3ac42b3c39/model.safetensors.index.json", "type": "weights" }, { "label": "Devstral-Small-2507 — pinned system prompt", "url": "https://huggingface.co/mistralai/Devstral-Small-2507/blob/bd165ab26cebbcc2eea2c4ecbfc07f3ac42b3c39/SYSTEM_PROMPT.txt", "type": "weights" }, { "label": "Devstral Small 1.1 — official Markdown model documentation", "url": "https://docs.mistral.ai/models/devstral-small-1-1-25-07.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/devstral-small-2507.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Devstral-Small-2507", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The 24B total and 24B active counts are nominal values explicitly reported by Mistral. The checkpoint supports 131,072 positions; API documentation advertises 128k. Mistral documents API deprecation on 2026-02-27 and retirement on 2026-05-31. A separate native reasoning mode remains unverified.", "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-26T22:51:31.417966Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/317.json" }, { "id": "huggingface/smollm3-3b", "slug": "huggingface--smollm3-3b", "name": "SmolLM3-3B", "developer": "Hugging Face", "developer_id": "huggingface", "family": "smollm", "description": null, "release_date": "2025-07-08", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 3, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 65536, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/blog/smollm3", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://huggingface.co/blog/smollm3", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/HuggingFaceTB/SmolLM3-3B/blob/a07cc9a04f16550a088caea529712d1d335b0ac1/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/HuggingFaceTB/SmolLM3-3B/blob/a07cc9a04f16550a088caea529712d1d335b0ac1/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Instruction and reasoning checkpoint. Nominal 3B parameters; the model configuration defaults to 65,536 tokens, with longer contexts available through documented YaRN scaling." }, { "id": "huggingface/smollm3-3b-base", "slug": "huggingface--smollm3-3b-base", "name": "SmolLM3-3B-Base", "developer": "Hugging Face", "developer_id": "huggingface", "family": "smollm", "description": null, "release_date": "2025-07-08", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 3, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 65536, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/blog/smollm3", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://huggingface.co/blog/smollm3", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/HuggingFaceTB/SmolLM3-3B-Base/blob/d78a42f79198603e614095753484a04c10c2b940/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/HuggingFaceTB/SmolLM3-3B-Base/blob/d78a42f79198603e614095753484a04c10c2b940/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Pretrained checkpoint. Nominal 3B parameters; the model configuration defaults to 65,536 tokens, with longer contexts available through documented YaRN scaling." }, { "id": "ai21/jamba-large", "slug": "ai21--jamba-large", "name": "Jamba Large", "developer": "AI21 Labs", "developer_id": "ai21", "family": "jamba", "description": "AI21's Jamba Large 1.7, a text-only hybrid Mamba-Transformer mixture-of-experts model for long-context instruction following and grounded generation", "release_date": "2025-07-03", "date_precision": "day", "open_weights": true, "license": "Jamba Open Model License", "parameters_b": 398, "active_parameters_b": 94, "parameters_source": "primary_source", "architecture": "hybrid-ssm-transformer-moe", "context_window": 256000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/ai21/jamba-large.toml", "source_type": "catalog", "sources": [ { "label": "AI21 model details and alias mapping", "url": "https://docs.ai21.com/docs/jamba-foundation-models", "type": "primary" }, { "label": "Hugging Face", "url": "https://huggingface.co/ai21labs/AI21-Jamba-Large-1.7", "type": "weights" }, { "label": "AI21 release changelog", "url": "https://docs.ai21.com/changelog", "type": "primary" }, { "label": "Jamba Open Model License", "url": "https://www.ai21.com/jamba-open-model-license/", "type": "primary" }, { "label": "AI21 function calling", "url": "https://docs.ai21.com/docs/function-calling", "type": "primary" }, { "label": "AI21 chat API", "url": "https://docs.ai21.com/reference/jamba-1-6-api-ref", "type": "primary" }, { "label": "Exact checkpoint public metadata", "url": "https://huggingface.co/api/models/ai21labs/AI21-Jamba-Large-1.7", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/ai21/jamba-large.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Reviewed alias target: jamba-large-1.7-2025-07. Counts are AI21's nominal 398B total / 94B active; public tensor metadata counts 398,555,145,696 parameters. Context 256000 normalizes the advertised 256K, not an independently verified exact implementation limit. Weights require acceptance of custom license conditions. Dedicated reasoning-mode status remains unverified.", "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-26T13:48:17Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/004.json" }, { "id": "cognitivecomputations/dolphin-mistral-24b-venice-edition-v1.1", "slug": "cognitivecomputations--dolphin-mistral-24b-venice-edition-v1.1", "name": "Venice Uncensored 1.1", "developer": "Dolphin (Cognitive Computations) and Venice", "developer_id": "cognitivecomputations", "family": "mistral", "description": "Dolphin and Venice conversational revision trained with new datasets and reinforcement learning.", "release_date": "2025-07-03", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": 24, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://venice.ai/blog/introducing-dolphin-mistral-24b-venice-edition-the-most-uncensored-ai-model-yet", "source_type": "primary", "sources": [ { "label": "Venice announces version 1.1 in the July 3 update", "url": "https://venice.ai/blog/introducing-dolphin-mistral-24b-venice-edition-the-most-uncensored-ai-model-yet", "type": "primary" } ], "aliases": [ "venice/venice-uncensored-1.1" ], "first_seen": "2026-10-29T11:32:56.160Z", "last_seen": "2026-10-29T11:32:56.160Z", "upstream_present": false, "availability": "public", "notes": "The dated publisher update explicitly announces a new trained revision of Dolphin Mistral 24B Venice Edition. The 24B count is the publisher’s nominal model size, not a checkpoint census. No independently identified v1.1 weight download or license was verified, so this row records hosted access. Context, active count, tools and reasoning mode remain unverified. June 2025 weights predate this revision and are not used as its technical evidence." }, { "id": "baidu/ernie-4.5-21b-a3b", "slug": "baidu--ernie-4.5-21b-a3b", "name": "ERNIE 4.5 21B A3B", "developer": "Baidu", "developer_id": "baidu", "family": "ernie", "description": "Open-weight, post-trained ERNIE 4.5 text MoE with about 21.95B checkpoint parameters and a nominal 3B active per token.", "release_date": "2025-06-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 21.948655808, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": null, "source_url": "https://ai.baidu.com/ai-doc/WENXINWORKSHOP/flxu4ej5u", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/baidu/ERNIE-4.5-21B-A3B-PT", "type": "weights" }, { "label": "Complete checkpoint tensor index", "url": "https://huggingface.co/baidu/ERNIE-4.5-21B-A3B-PT/blob/87db95487941cb39592ee0abca3b9155a6d19c5c/model.safetensors.index.json", "type": "weights" }, { "label": "Router bias training implementation", "url": "https://github.com/PaddlePaddle/ERNIE/blob/30ad9866a4fb89bd7184dc01aff2236e1f1fd1ec/ernie/callbacks/moe_correction_bias_adjust_callback.py", "type": "primary" }, { "label": "ERNIE 4.5 technical report", "url": "https://ernie.baidu.com/blog/publication/ERNIE_Technical_Report.pdf", "type": "paper" }, { "label": "Baidu Qianfan release history", "url": "https://ai.baidu.com/ai-doc/WENXINWORKSHOP/flxu4ej5u", "type": "primary" }, { "label": "Baidu open-weight announcement", "url": "https://ernie.baidu.com/blog/posts/ernie4.5/", "type": "primary" }, { "label": "Exact checkpoint chat template", "url": "https://huggingface.co/baidu/ERNIE-4.5-21B-A3B-PT/raw/87db95487941cb39592ee0abca3b9155a6d19c5c/chat_template.jinja", "type": "primary" }, { "label": "FastDeploy tool-calling documentation", "url": "https://paddlepaddle.github.io/FastDeploy/features/tool_calling/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/baidu/ernie-4.5-21b-a3b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "PT means PyTorch; this is the post-trained non-thinking model, distinct from Base and Thinking. Complete checkpoint: 21.948655808B parameters including 0.123217920B MTP and 1728 training-adjusted routing biases; advertised 21B total / 3B active are nominal. Qianfan documents hosted release June 29, 2025; weights were announced June 30. Tool calling for this exact variant remains unverified because first-party examples conflict with the Thinking-only support table.", "parameter_components_b": { "main_model": 21.825437888, "multi_token_prediction": 0.12321792 }, "field_status": { "tool_call": "unverified" }, "audited_at": "2026-10-26T17:36:18.511564+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/124.json" }, { "id": "baidu/ernie-4.5-300b-a47b", "slug": "baidu--ernie-4.5-300b-a47b", "name": "ERNIE 4.5 300B A47B", "developer": "Baidu", "developer_id": "baidu", "family": "ernie", "description": "Baidu post-trained text MoE with approximately 300B parameters and a reported 47B activated per token.", "release_date": "2025-06-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 300.474051776, "active_parameters_b": 47, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": null, "source_url": "https://huggingface.co/baidu/ERNIE-4.5-300B-A47B-Paddle", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/baidu/ERNIE-4.5-300B-A47B-Paddle", "type": "weights" }, { "label": "Announcing the Open Source Release of the ERNIE 4.5 Model Family", "url": "https://ernie.baidu.com/blog/posts/ernie4.5/", "type": "primary" }, { "label": "Exact Paddle model configuration", "url": "https://huggingface.co/baidu/ERNIE-4.5-300B-A47B-Paddle/resolve/c96334ddf7c1dae19e50cf5f972d8820531f7e59/config.json", "type": "primary" }, { "label": "Exact Paddle checkpoint index and indexed safetensors headers", "url": "https://huggingface.co/baidu/ERNIE-4.5-300B-A47B-Paddle/resolve/c96334ddf7c1dae19e50cf5f972d8820531f7e59/model.safetensors.index.json", "type": "primary" }, { "label": "Auxiliary-loss-free routing-bias update implementation", "url": "https://github.com/PaddlePaddle/ERNIE/blob/release/v1.5/ernie/callbacks/moe_correction_bias_adjust_callback.py", "type": "primary" }, { "label": "ERNIE 4.5 Technical Report", "url": "https://ernie.baidu.com/blog/publication/ERNIE_Technical_Report.pdf", "type": "primary" }, { "label": "AI Studio large-model API capability documentation", "url": "https://ai.baidu.com/ai-doc/AISTUDIO/rm344erns", "type": "primary" }, { "label": "Exact model chat template", "url": "https://huggingface.co/baidu/ERNIE-4.5-300B-A47B-Paddle/resolve/c96334ddf7c1dae19e50cf5f972d8820531f7e59/tokenizer_config.json", "type": "primary" }, { "label": "PyTorch packaging configuration", "url": "https://huggingface.co/baidu/ERNIE-4.5-300B-A47B-PT/resolve/47bbaf1c9e356313c3cb2ded9078a494fb152f0e/config.json", "type": "primary" }, { "label": "PyTorch packaging checkpoint index", "url": "https://huggingface.co/baidu/ERNIE-4.5-300B-A47B-PT/resolve/47bbaf1c9e356313c3cb2ded9078a494fb152f0e/model.safetensors.index.json", "type": "primary" }, { "label": "Qianfan model release history", "url": "https://ai.baidu.com/ai-doc/WENXINWORKSHOP/flxu4ej5u", "type": "primary" }, { "label": "ERNIE open model launch announcement for Qianfan", "url": "https://ai.baidu.com/ai-doc/WENXINWORKSHOP/4mchtzl8s", "type": "primary" }, { "label": "BaiDu ERNIE 4.5 MoE Model API Now Live on Novita AI", "url": "https://blogs.novita.ai/ernie-4-5-on-novita-ai/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/baidu/ernie-4.5-300b-a47b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete learned checkpoint count is 300.474051776B including 0.989888512B auxiliary MTP; 300B total and 47B active are published rounded labels. Routing correction biases are included because training updates them. This is the post-trained, non-thinking text variant; PT means PyTorch packaging. Native tool calling is unverified. The model-card table contains conflicting Pretraining and Vision Experts rows; exact configuration and technical report resolve these as post-trained and text-only. Earliest exact-name hosted availability verified June29 on Novita; Baidu announced open weights June30. A June25 Qianfan alias existed, but its checkpoint identity at launch is unresolved.", "parameter_components_b": { "mtp": 0.989888512, "routed_experts": 287.494373376, "dense_ffn": 2.113929216, "attention": 8.153726976, "embeddings_and_head": 1.694498816, "routers": 0.026738688, "routing_biases": 0.000003264, "norms": 0.000892928 }, "field_status": { "tool_call": "unverified" }, "audited_at": "2026-10-26T17:37:04.534972Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/125.json" }, { "id": "baidu/ernie-4.5-vl-424b-a47b", "slug": "baidu--ernie-4.5-vl-424b-a47b", "name": "ERNIE-4.5-VL-424B-A47B", "developer": "Baidu", "developer_id": "baidu", "family": "ernie", "description": "Post-trained vision-language MoE chat model with text, image, and video inputs, text output, and thinking and non-thinking modes.", "release_date": "2025-06-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 423.526285184, "active_parameters_b": 47, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "image", "text", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://huggingface.co/baidu/ERNIE-4.5-VL-424B-A47B-PT", "source_type": "primary", "sources": [ { "label": "Announcing the Open Source Release of the ERNIE 4.5 Model Family", "url": "https://ernie.baidu.com/blog/posts/ernie4.5/", "type": "primary" }, { "label": "ERNIE-4.5-VL-424B-A47B official model card", "url": "https://huggingface.co/baidu/ERNIE-4.5-VL-424B-A47B-PT/blob/a3d5a8439d67c9a69003a1f7aac0961c6f1bc20d/README.md", "type": "primary" }, { "label": "Exact checkpoint configuration", "url": "https://huggingface.co/baidu/ERNIE-4.5-VL-424B-A47B-PT/blob/a3d5a8439d67c9a69003a1f7aac0961c6f1bc20d/config.json", "type": "primary" }, { "label": "Complete checkpoint index and referenced safetensors headers", "url": "https://huggingface.co/baidu/ERNIE-4.5-VL-424B-A47B-PT/blob/a3d5a8439d67c9a69003a1f7aac0961c6f1bc20d/model.safetensors.index.json", "type": "primary" }, { "label": "Official PyTorch model implementation", "url": "https://huggingface.co/baidu/ERNIE-4.5-VL-424B-A47B-PT/blob/a3d5a8439d67c9a69003a1f7aac0961c6f1bc20d/modeling_ernie4_5_vl.py", "type": "primary" }, { "label": "Training update for routing correction biases", "url": "https://github.com/PaddlePaddle/ERNIE/blob/30ad9866a4fb89bd7184dc01aff2236e1f1fd1ec/ernie/callbacks/moe_correction_bias_adjust_callback.py", "type": "primary" }, { "label": "Exact checkpoint chat template", "url": "https://huggingface.co/baidu/ERNIE-4.5-VL-424B-A47B-PT/blob/a3d5a8439d67c9a69003a1f7aac0961c6f1bc20d/chat_template.json", "type": "primary" }, { "label": "AI Studio large-model API: model and feature support", "url": "https://ai.baidu.com/ai-doc/AISTUDIO/rm344erns", "type": "primary" }, { "label": "Novita exact-name hosted release", "url": "https://blogs.novita.ai/ernie-4-5-on-novita-ai/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/baidu/ernie-4.5-vl-424b-a47b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The post-trained model PT suffix means PyTorch. The complete checkpoint contains 423.526285184B parameters, including the vision encoder and adapter; 424B is nominal. The published 47B active count does not define exact text/vision or embedding accounting. Novita documents this exact model available June 29, 2025; Baidu announced weights June 30. Earlier family and rolling-endpoint dates do not prove checkpoint identity. Native tool calling remains unverified; Baidu excludes the mapped endpoint from its function-call support list.", "field_status": { "tool_call": "unverified" }, "audited_at": "2026-10-26T17:38:24.471Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/126.json" }, { "id": "google/gemma-3n-e2b-it", "slug": "google--gemma-3n-e2b-it", "name": "Gemma 3n E2B IT", "developer": "Google", "developer_id": "google", "family": "gemma", "description": "Instruction-tuned Gemma 3n E2B model with a MatFormer Transformer, text, image, audio and video input, and text output.", "release_date": "2025-06-26", "date_precision": "day", "open_weights": true, "license": "Gemma Terms of Use", "parameters_b": 5.44, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "transformer", "context_window": 32768, "modalities": { "input": [ "text", "image", "audio", "video" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://developers.googleblog.com/introducing-gemma-3n-developer-guide/", "source_type": "primary", "sources": [ { "label": "Google full-release announcement", "url": "https://developers.googleblog.com/introducing-gemma-3n-developer-guide/", "type": "primary" }, { "label": "Google model card", "url": "https://ai.google.dev/gemma/docs/gemma-3n/model_card", "type": "primary" }, { "label": "Google parameter definitions", "url": "https://ai.google.dev/gemma/docs/gemma-3n", "type": "primary" }, { "label": "Google E2B parameter diagram", "url": "https://ai.google.dev/static/gemma/docs/images/gemma-3n-parameters.png", "type": "primary" }, { "label": "Hugging Face", "url": "https://huggingface.co/google/gemma-3n-E2B-it", "type": "weights" }, { "label": "Checkpoint public metadata", "url": "https://huggingface.co/api/models/google/gemma-3n-E2B-it", "type": "weights" }, { "label": "Exact variant release list", "url": "https://huggingface.co/blog/gemma3n", "type": "primary" }, { "label": "Gemma Terms of Use", "url": "https://ai.google.dev/gemma/terms", "type": "primary" }, { "label": "Keras E2B IT parameter count", "url": "https://keras.io/keras_hub/api/models/gemma3n/gemma3n_causal_lm/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemma-3n-e2b-it.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "5.44B is Google’s rounded complete E2B count, including text, vision, audio and per-layer embeddings, corroborated by Keras and checkpoint metadata. E2B and 1.91B describe effective memory loading; a separate per-token active count is undisclosed. The model card’s 6B figure conflicts with those sources. June 26 dates the complete release; May offered a separate text/vision LiteRT preview.", "parameter_components_b": { "text": 1.91, "vision": 0.3, "audio": 0.68, "per_layer_embedding": 2.55 }, "field_status": { "active_parameters_b": "undisclosed", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T13:18:23Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/220.json" }, { "id": "google/gemma-3n-e4b-it", "slug": "google--gemma-3n-e4b-it", "name": "Gemma 3n E4B IT", "developer": "Google", "developer_id": "google", "family": "gemma", "description": "Google’s instruction-tuned Gemma 3n E4B model accepts text, images, audio and video and generates text. It uses a nested MatFormer architecture designed for devices with limited memory.", "release_date": "2025-06-26", "date_precision": "day", "open_weights": true, "license": "Gemma Terms of Use", "parameters_b": 8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "transformer", "context_window": 32768, "modalities": { "input": [ "text", "image", "audio", "video" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://developers.googleblog.com/en/introducing-gemma-3n-developer-guide/", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/google/gemma-3n-E4B-it", "type": "weights" }, { "label": "Google full release, June 26, 2025", "url": "https://developers.googleblog.com/en/introducing-gemma-3n-developer-guide/", "type": "primary" }, { "label": "Google parameter and effective-memory definitions", "url": "https://ai.google.dev/gemma/docs/gemma-3n", "type": "primary" }, { "label": "Google Gemma 3n model card", "url": "https://ai.google.dev/gemma/docs/gemma-3n/model_card", "type": "primary" }, { "label": "Official Hub parameter metadata", "url": "https://huggingface.co/api/models/google/gemma-3n-E4B-it?blobs=true", "type": "weights" }, { "label": "Google MatFormer Lab", "url": "https://github.com/google-gemma/cookbook/blob/main/.archive/Gemma/%5BGemma_3n%5DMatFormer_Lab.ipynb", "type": "primary" }, { "label": "Google AI Edge preview tool-calling announcement", "url": "https://developers.googleblog.com/google-ai-edge-small-language-models-multimodality-rag-function-calling/", "type": "primary" }, { "label": "KerasHub E4B total/effective parameter distinction", "url": "https://keras.io/keras_hub/api/base_classes/causal_lm_preprocessor/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemma-3n-e4b-it.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "8B is Google’s nominal complete E4B count. E4B and 3.98B describe effective memory loading; a separate active-parameter count is not disclosed. June 26 dates this complete launch checkpoint; May offered a limited preview. Dedicated reasoning-mode and exact-checkpoint tool support remain unverified.", "field_status": { "active_parameters_b": "undisclosed", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T13:19:18.947811+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/221.json" }, { "id": "mistral/mistral-small-2506", "slug": "mistral--mistral-small-2506", "name": "Mistral Small 3.2", "developer": "Mistral", "developer_id": "mistral", "family": "mistral-small", "description": "Mistral Small 3.2 is a 24B dense multimodal instruction model with improved instruction following, reduced repetition, and function calling.", "release_date": "2025-06-20", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 24, "active_parameters_b": 24, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://docs.mistral.ai/models/mistral-small-3-2-25-06", "source_type": "primary", "sources": [ { "label": "Mistral Small 3.2 model documentation", "url": "https://docs.mistral.ai/models/mistral-small-3-2-25-06", "type": "primary" }, { "label": "Pinned official Mistral Small 3.2 documentation data", "url": "https://github.com/mistralai/platform-docs-public/blob/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/schema/models/models/mistral-small-3-2-25-06.ts", "type": "primary" }, { "label": "Pinned official model weights table renderer", "url": "https://github.com/mistralai/platform-docs-public/blob/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/app/%5Blocale%5D/(docs)/model-card/%5Bslug%5D/components/weights-table.tsx", "type": "primary" }, { "label": "Mistral official changelog", "url": "https://docs.mistral.ai/resources/changelogs", "type": "primary" }, { "label": "Mistral Small 3.2 24B Instruct 2506 model card", "url": "https://huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506/blob/95a6d26c4bfb886c58daf9d3f7332c857cb27b43/README.md", "type": "weights" }, { "label": "Pinned Mistral Small 3.2 Transformers configuration", "url": "https://huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506/blob/95a6d26c4bfb886c58daf9d3f7332c857cb27b43/config.json", "type": "weights" }, { "label": "Pinned Mistral Small 3.2 native configuration", "url": "https://huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506/blob/95a6d26c4bfb886c58daf9d3f7332c857cb27b43/params.json", "type": "weights" }, { "label": "Pinned Mistral Small 3.2 weight index", "url": "https://huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506/blob/95a6d26c4bfb886c58daf9d3f7332c857cb27b43/model.safetensors.index.json", "type": "weights" }, { "label": "Pinned mistral-common chat template documentation source", "url": "https://github.com/mistralai/mistral-common/blob/d4364d04c684051a7a246e49bc699b9b1ed46420/src/mistral_common/integrations/chat_templates/chat_templates.py", "type": "primary" }, { "label": "Mistral Small family legal lifecycle", "url": "https://legal.mistral.ai/ai-governance/models/mistral-small-3/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/mistral-small-2506.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "24B total and 24B active are separate nominal disclosures in Mistral's official model table, not an inferred per-token census. Dense multimodal Transformer; native context is 131072 tokens, advertised as 128k. June 20, 2025 is the public Small 3.2 model release; the mistral-small-2506 API became available June 23. Dedicated reasoning is explicitly unsupported for this checkpoint. Official documentation records API retirement July 31, 2026; specifications here do not establish current endpoint operation.", "field_status": {}, "audited_at": "2026-10-26T23:30:31.622533Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/343.json" }, { "id": "mistral/mistral-small-3.2-24b-instruct", "slug": "mistral--mistral-small-3.2-24b-instruct", "name": "Mistral Small 3.2 24B Instruct", "developer": "Mistral", "developer_id": "mistral", "family": "mistral-small", "description": "Instruction-tuned multimodal Mistral Small model with improved instruction following, fewer repetition errors, and a more robust function-calling template than Small 3.1.", "release_date": "2025-06-20", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 24, "active_parameters_b": 24, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "image", "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506", "source_type": "primary", "sources": [ { "label": "Official exact-checkpoint model card", "url": "https://huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506/blob/95a6d26c4bfb886c58daf9d3f7332c857cb27b43/README.md", "type": "primary" }, { "label": "Official total and active parameter specification", "url": "https://github.com/mistralai/platform-docs-public/blob/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/schema/models/models/mistral-small-3-2-25-06.ts", "type": "primary" }, { "label": "Official parameter-table billion units", "url": "https://github.com/mistralai/platform-docs-public/blob/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/app/%5Blocale%5D/%28docs%29/model-card/%5Bslug%5D/components/weights-table.tsx", "type": "primary" }, { "label": "Native checkpoint configuration", "url": "https://huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506/blob/95a6d26c4bfb886c58daf9d3f7332c857cb27b43/params.json", "type": "weights" }, { "label": "Transformers checkpoint configuration", "url": "https://huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506/blob/95a6d26c4bfb886c58daf9d3f7332c857cb27b43/config.json", "type": "weights" }, { "label": "Published weight tensor index", "url": "https://huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506/blob/95a6d26c4bfb886c58daf9d3f7332c857cb27b43/model.safetensors.index.json", "type": "weights" }, { "label": "Official non-reasoning classification", "url": "https://github.com/mistralai/mistral-common/blob/d4364d04c684051a7a246e49bc699b9b1ed46420/src/mistral_common/integrations/chat_templates/chat_templates.py", "type": "primary" }, { "label": "Official model lifecycle and release date", "url": "https://legal.mistral.ai/ai-governance/models/mistral-small-3/", "type": "primary" }, { "label": "Official model and API release changelog", "url": "https://docs.mistral.ai/resources/changelogs", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/mistral-small-3.2-24b-instruct.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Total 24B and active 24B are nominal values separately published by Mistral for Mistral-Small-3.2-24B-Instruct-2506, not tensor-exact or phase-specific counts. Native context is 131072 tokens (advertised as 128k). Model release was June 20, 2025; June 23 was later API availability.", "field_status": {}, "audited_at": "2026-10-26T23:33:48Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/346.json" }, { "id": "google/gemini-2.5-flash-lite", "slug": "google--gemini-2.5-flash-lite", "name": "Gemini 2.5 Flash-Lite", "developer": "Google", "developer_id": "google", "family": "gemini-flash-lite", "description": "Gemini 2.5 multimodal model with optional thinking and function calling.", "release_date": "2025-06-17", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "audio", "video", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-flash-lite", "source_type": "primary", "sources": [ { "label": "Gemini 2.5 Flash-Lite Model Card (September 26, 2025)", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Flash-Lite-Model-Card.pdf", "type": "primary" }, { "label": "Gemini 2.5 Flash-Lite — Gemini API", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-flash-lite", "type": "primary" }, { "label": "We’re expanding our Gemini 2.5 family of models", "url": "https://blog.google/products-and-platforms/products/gemini/gemini-2-5-model-family-expands/", "type": "primary" }, { "label": "Gemini 2.5 Flash-Lite is now stable and generally available", "url": "https://developers.googleblog.com/en/gemini-25-flash-lite-is-now-stable-and-generally-available/", "type": "primary" }, { "label": "Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Model versions and lifecycle — Gemini Enterprise Agent Platform", "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/model-versions", "type": "primary" }, { "label": "Learn about supported models — Firebase AI Logic", "url": "https://firebase.google.com/docs/ai-logic/models", "type": "primary" }, { "label": "Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities", "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_v2_5_report.pdf", "type": "primary" }, { "label": "Gemini 3.1 Flash-Lite Model Card (March 2026)", "url": "https://deepmind.google/models/model-cards/gemini-3-1-flash-lite/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemini-2.5-flash-lite.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "First public preview: June 17, 2025. Stable API ID: July 22; Google explicitly says it uses the same underlying model. September Preview (09-2025) is separate. Sparse MoE Transformer; Google has not published total or active parameter counts. Availability documentation conflicts: the API page retains access for previous users while Firebase lists an October 16, 2026 shutdown warning.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T12:52:04.929388Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/194.json" }, { "id": "google/gemini-2.5-flash-lite-preview-06-17", "slug": "google--gemini-2.5-flash-lite-preview-06-17", "name": "Gemini 2.5 Flash-Lite Preview 06-17", "developer": "Google", "developer_id": "google", "family": "gemini-flash-lite", "description": "June 2025 preview of Gemini 2.5 Flash-Lite with optional thinking and function calling.", "release_date": "2025-06-17", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text", "image", "audio", "video", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ai.google.dev/gemini-api/docs/changelog", "source_type": "primary", "sources": [ { "label": "Gemini API release notes", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "primary" }, { "label": "Gemini 2.5: Updates to our family of thinking models (June 17, 2025)", "url": "https://developers.googleblog.com/en/gemini-2-5-thinking-model-updates/", "type": "primary" }, { "label": "Gemini 2.5 Flash-Lite is now stable and generally available (July 22, 2025)", "url": "https://developers.googleblog.com/en/gemini-25-flash-lite-is-now-stable-and-generally-available/", "type": "primary" }, { "label": "Gemini 2.5 Flash-Lite Model Card (September 26, 2025)", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Flash-Lite-Model-Card.pdf", "type": "primary" }, { "label": "Gemini models — historical indexed exact June preview specification", "url": "https://ai.google.dev/gemini-api/docs/models/gemini?hl=sl", "type": "primary" }, { "label": "Gemini 2.5 Flash-Lite — Gemini API", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-flash-lite", "type": "primary" }, { "label": "Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Gemini 3.1 Flash-Lite Model Card (March 2026)", "url": "https://deepmind.google/models/model-cards/gemini-3-1-flash-lite/", "type": "primary" }, { "label": "Gemini 2.5 technical report", "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_v2_5_report.pdf", "type": "primary" }, { "label": "Gemini 2.5 Flash and Flash-Lite September preview release", "url": "https://developers.googleblog.com/continuing-to-bring-you-our-latest-models-with-an-improved-gemini-2-5-flash-and-flash-lite-release/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemini-2.5-flash-lite-preview-06-17.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "June 17, 2025 preview; Google identifies the July stable release as the same underlying model. Sparse MoE Transformer; neither total nor active parameter count is published in the reviewed sources. The September preview is a separate update. Historical exact-preview docs list 1,000,000 input tokens; stable-ID docs list 1,048,576. The November 4 notice scheduled this endpoint for shutdown on November 18, superseding an earlier August removal plan.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T12:51:16.695Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/195.json" }, { "id": "minimax/minimax-m1-80k", "slug": "minimax--minimax-m1-80k", "name": "MiniMax M1 80K", "developer": "MiniMax", "developer_id": "minimax", "family": "minimax", "description": "Open-weight MoE reasoning model with hybrid Lightning and softmax attention, a one-million-token context and an 80K thinking budget.", "release_date": "2025-06-16", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 456, "active_parameters_b": 45.9, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.minimax.io/news/minimaxm1", "source_type": "primary", "sources": [ { "label": "MiniMax M1 launch announcement", "url": "https://www.minimax.io/news/minimaxm1", "type": "primary" }, { "label": "MiniMax-M1-80k official model card", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M1-80k/blob/8d1494b1a260e22040d5b9b2eb332eb44500b34d/README.md", "type": "weights" }, { "label": "MiniMax-M1: Scaling Test-Time Compute Efficiently with Lightning Attention", "url": "https://arxiv.org/html/2506.13585v1", "type": "primary" }, { "label": "MiniMax-M1-80k checkpoint configuration", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M1-80k/blob/8d1494b1a260e22040d5b9b2eb332eb44500b34d/config.json", "type": "weights" }, { "label": "MiniMax-M1-80k Apache License 2.0", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M1-80k/blob/8d1494b1a260e22040d5b9b2eb332eb44500b34d/LICENSE", "type": "weights" }, { "label": "MiniMax-M1 native function-call guide", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M1-80k/blob/8d1494b1a260e22040d5b9b2eb332eb44500b34d/docs/function_call_guide.md", "type": "weights" }, { "label": "MiniMax-M1-80k tokenizer and chat template", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M1-80k/blob/8d1494b1a260e22040d5b9b2eb332eb44500b34d/tokenizer_config.json", "type": "weights" }, { "label": "MiniMax-M1-80k reference implementation", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M1-80k/blob/8d1494b1a260e22040d5b9b2eb332eb44500b34d/modeling_minimax_m1.py", "type": "weights" }, { "label": "Official M1-80k repository metadata and bounded weight access", "url": "https://huggingface.co/api/models/MiniMaxAI/MiniMax-M1-80k", "type": "weights" }, { "label": "MiniMax M1 vLLM deployment guide", "url": "https://github.com/MiniMax-AI/MiniMax-M1/blob/main/docs/vllm_deployment_guide.md", "type": "primary" }, { "label": "Current MiniMax model list", "url": "https://platform.minimax.io/docs/faq/models", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/minimax/minimax-m1-80k.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/MiniMaxAI/MiniMax-M1-80k", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "456B total and 45.9B active are developer nominal figures; the active-component counting convention is not separately specified. Hybrid attention uses seven Lightning blocks per softmax block, with top-2-of-32 MoE routing. The 1M context is distinct from the 80K thinking/output budget. Configuration/tokenizer ceilings of 10.24M/40.96M and hardware-serving claims of 2M/5M are not substituted for the documented native context. Public launch was June 16, 2025; June 13 repository creation is not proof of earlier public release. The exact 80k weights use Apache-2.0. Current native API availability is not established by surviving launch/card links.", "field_status": {}, "audited_at": "2026-10-26T22:43:01.857589Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/311.json" }, { "id": "morph/morph-v3-fast", "slug": "morph--morph-v3-fast", "name": "Morph V3 Fast", "developer": "Morph", "developer_id": "morph", "family": "morph", "description": "Specialized code-edit application model that merges an instruction and proposed edit snippet into an existing source file and returns the complete updated file.", "release_date": "2025-06-15", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": null, "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.morphllm.com/blog/morph-gets-faster", "source_type": "primary", "sources": [ { "label": "Morph exact model API metadata", "url": "https://www.morphllm.com/api/models/json", "type": "primary" }, { "label": "Morph Fast Apply API documentation", "url": "https://docs.morphllm.com/llms.txt", "type": "primary" }, { "label": "Morph Fast Apply model overview", "url": "https://www.morphllm.com/fast-apply-model", "type": "primary" }, { "label": "Morph EULA", "url": "https://www.morphllm.com/eula", "type": "primary" }, { "label": "Morph Terms of Service", "url": "https://www.morphllm.com/privacy/tos", "type": "primary" }, { "label": "Morph SDK Fast Apply component", "url": "https://docs.morphllm.com/sdk/components/fast-apply", "type": "primary" }, { "label": "Morph speculative decoding and exact Fast production configuration", "url": "https://www.morphllm.com/speculative-decoding", "type": "primary" }, { "label": "Morph code-generation inference research", "url": "https://www.morphllm.com/blog/codegen-inference-research", "type": "primary" }, { "label": "Morph V3 Fast and Large introduction", "url": "https://www.morphllm.com/blog/morph-gets-faster", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/morph/morph-v3-fast.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Morph publishes nominal 7B for Fast Apply, matched to morph-v3-fast by the exact 10,500 tok/s and $0.80 row. Active parameters are not separately disclosed, and no named base lineage is established. Neural architecture remains unresolved. Morph documents 262,144 text tokens; the developer introduced V3 Fast and Large on June 15, 2025. Native reasoning/tool support remains unverified.", "field_status": { "active_parameters_b": "undisclosed", "architecture": "unverified", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T19:27:36.061408Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/366.json" }, { "id": "morph/morph-v3-large", "slug": "morph--morph-v3-large", "name": "Morph V3 Large", "developer": "Morph", "developer_id": "morph", "family": "morph", "description": "Specialized code-edit application model that merges edit snippets into existing source files, intended for complex changes.", "release_date": "2025-06-15", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.morphllm.com/blog/morph-gets-faster", "source_type": "primary", "sources": [ { "label": "Morph exact model metadata", "url": "https://www.morphllm.com/api/models/json", "type": "primary" }, { "label": "Morph Apply API documentation", "url": "https://docs.morphllm.com/llms-full.txt", "type": "primary" }, { "label": "Morph Fast Apply model guide", "url": "https://www.morphllm.com/fast-apply-model", "type": "primary" }, { "label": "Morph model lineup", "url": "https://www.morphllm.com/models", "type": "primary" }, { "label": "Morph EULA", "url": "https://www.morphllm.com/eula", "type": "primary" }, { "label": "Morph later inference research", "url": "https://www.morphllm.com/blog/codegen-inference-research", "type": "primary" }, { "label": "Morph dated speed announcement (body unavailable)", "url": "https://www.morphllm.com/blog/morph-gets-faster", "type": "primary" }, { "label": "OpenRouter provider listing", "url": "https://openrouter.ai/morph/morph-v3-large", "type": "catalog" }, { "label": "Morph Apply model specification", "url": "https://docs.morphllm.com/models/apply", "type": "primary" }, { "label": "Morph CLI Large model distinction", "url": "https://docs.morphllm.com/cli", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/morph/morph-v3-large.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "V3 Large is a specialized code-apply model, distinct from V3 Fast and Morph-hosted general coding models. The October 27 parameter re-audit reopened exact metadata, full documentation, the Fast Apply guide and later inference research, and searched for nominal counts and explicit base-model lineage. The generic developer 7B claim accompanies the Fast-speed/pricing offering; the CLI explicitly calls V3 Large the larger model. No nominal Large total or named base checkpoint was established, so total remains unverified and active count undisclosed. This is an attribution limitation, not a requirement for an exact tensor census. Backbone remains unverified. The fully rendered developer introduction establishes June 15, 2025 for V3 Fast and Large; July 7 is a later provider-catalog date. Reasoning/tool-call support remains unverified because absent metadata is not a negative classification. Current documented specifications were checked; endpoint operation was not tested.", "field_status": { "parameters_b": "unverified", "active_parameters_b": "undisclosed", "architecture": "unverified", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T19:27:48.785689Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/367.json" }, { "id": "bytedance-seed/seed-1-6", "slug": "bytedance-seed--seed-1-6", "name": "Seed 1.6", "developer": "ByteDance", "developer_id": "bytedance-seed", "family": "seed", "description": "ByteDance multimodal MoE model with adaptive reasoning, a 256K context window, and tool calling.", "release_date": "2025-06-11", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 230, "active_parameters_b": 23, "parameters_source": "primary_source", "architecture": "moe", "context_window": 256000, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://seed.bytedance.com/en/blog/introduction-to-techniques-used-in-seed1-6", "source_type": "primary", "sources": [ { "label": "Introduction to Techniques Used in Seed1.6", "url": "https://seed.bytedance.com/en/blog/introduction-to-techniques-used-in-seed1-6", "type": "primary" }, { "label": "更强模型效果！豆包大模型 1.6 系列上线边缘大模型网关，最高申领1000万免费 Tokens", "url": "https://developer.volcengine.com/articles/7518983626017472553", "type": "primary" }, { "label": "豆包大模型1.6升级：国内首个原生支持“分档调节思考长度”的大模型", "url": "https://www.volcengine.com/news/detail/23", "type": "primary" }, { "label": "Model list - BytePlus ModelArk Documentation", "url": "https://ai.byteplus.com/ark/region:ap-southeast-1/docs/model-list", "type": "primary" }, { "label": "Model list - image and video capability rows", "url": "https://docs.byteplus.com/zh-TW/docs/modelark/model-list", "type": "primary" }, { "label": "BytePlus Service Specific Terms", "url": "https://docs.byteplus.com/en/docs/legal/docs-service-specific-terms", "type": "primary" }, { "label": "seed-1.6 - recommended model documentation", "url": "https://ai.byteplus.com/ark/region:ap-southeast-1/docs/recommended-models-seed-1-6.md", "type": "primary" }, { "label": "Volcano Engine Edge Cloud author profile", "url": "https://developer.volcengine.com/user/26859967094408", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/bytedance-seed/seed-1-6.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The original Doubao-Seed-1.6 launch was June 11, 2025; June 25 is the Seed technical-introduction date and October 15 an upgrade. Parameter counts describe the standard 230B/23B Seed1.6 foundation, not Flash. Capabilities are documented specifications, not a live endpoint test. Context 256000 is the decimal normalization of the provider’s nominal 256K specification.", "field_status": {}, "audited_at": "2026-10-26T17:41:42.113970Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/128.json" }, { "id": "bytedance-seed/seed-1-6-flash", "slug": "bytedance-seed--seed-1-6-flash", "name": "Seed 1.6 Flash", "developer": "ByteDance", "developer_id": "bytedance-seed", "family": "seed", "description": "ByteDance multimodal model with optional thinking, text, image and video input, and tool calling.", "release_date": "2025-06-11", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 256000, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://developer.volcengine.com/articles/7518983626017472553", "source_type": "primary", "sources": [ { "label": "豆包大模型 1.6 系列上线边缘大模型网关", "url": "https://developer.volcengine.com/articles/7518983626017472553", "type": "primary" }, { "label": "Model releases - 2025", "url": "https://ai.byteplus.com/ark/region:ap-southeast-1/docs/model-release-announcement-2025", "type": "primary" }, { "label": "Model list", "url": "https://ai.byteplus.com/ark/region:ap-southeast-1/docs/model-list", "type": "primary" }, { "label": "Built on Seed1.6-Flash, Seed-1.6-Embedding Launched", "url": "https://seed.bytedance.com/en/blog/built-on-seed1-6-flash-seed-1-6-embedding-launched", "type": "primary" }, { "label": "BytePlus service-specific terms, June 16 2026 version", "url": "https://docs.byteplus.com/en/docs/legal/docs-service-specific-terms?TimeBefore=1781602227", "type": "primary" }, { "label": "Introduction to Techniques Used in Seed1.6", "url": "https://seed.bytedance.com/en/blog/introduction-to-techniques-used-in-seed1-6", "type": "primary" }, { "label": "Seed2.0 Model Card: Towards Intelligence Frontier for Real-World Complexity", "url": "https://arxiv.org/abs/2607.00248", "type": "primary" }, { "label": "Coze model release notes", "url": "https://docs.coze.cn/guides_model_release_note", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/bytedance-seed/seed-1-6-flash.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Nominal total and active counts remain undisclosed: no developer source maps Flash to the published 230B/23B Seed1.6 base. Flash topology also remains undisclosed. 256K context is the provider's nominal notation. Earliest public Flash launch was June 11, 2025; later provider dates and snapshot suffixes do not replace it.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T10:50:38.842220+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/129.json" }, { "id": "nvidia/mistral-nemotron", "slug": "nvidia--mistral-nemotron", "name": "Mistral Nemotron", "developer": "Mistral AI", "developer_id": "mistral", "family": "nemotron", "description": "Mistral AI text model optimized by NVIDIA for instruction following, coding, translation, summarization and tool use.", "release_date": "2025-06-11", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://build.nvidia.com/mistralai/mistral-nemotron.md", "source_type": "primary", "sources": [ { "label": "Official Mistral-Nemotron model card in Markdown", "url": "https://build.nvidia.com/mistralai/mistral-nemotron.md", "type": "primary" }, { "label": "Official endpoint page and embedded endpoint specification", "url": "https://build.nvidia.com/mistralai/mistral-nemotron", "type": "primary" }, { "label": "June 11, 2025 NVIDIA announcement", "url": "https://blogs.nvidia.com/blog/france-sovereign-ai-infrastructure/", "type": "primary" }, { "label": "Mistral partner-served deployment terms", "url": "https://legal.mistral.ai/terms/partner-served-deployment-terms/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/nvidia/mistral-nemotron.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Mistral AI developed this model; NVIDIA optimized and hosted it. The official card describes a modified Transformer with 128K context; total and active counts, dense/MoE structure and a named count-bearing base remain undisclosed. NeMo, Minitron and the later Nemotron coalition models do not establish this endpoint's size. Public weights are unavailable; deployment requires a Mistral license. The embedded API schema’s MIT label is not backed by an accessible license. Current capability metadata marks reasoning and function calling supported, but NVIDIA’s page also says this API was deprecated September 27, 2026 and unsupported after September 28; current endpoint operation was not tested. The catalog ID retains its original provider-based prefix.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T15:58:23.645782+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/379.json" }, { "id": "mistral/magistral-medium-2506", "slug": "mistral--magistral-medium-2506", "name": "Magistral Medium 1.0", "developer": "Mistral", "developer_id": "mistral", "family": "magistral", "description": null, "release_date": "2025-06-10", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 40000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://docs.mistral.ai/models/magistral-medium-1-0-25-06", "source_type": "primary", "sources": [ { "label": "Official version-specific model documentation", "url": "https://docs.mistral.ai/models/magistral-medium-1-0-25-06", "type": "docs" }, { "label": "Dated publisher API changelog", "url": "https://docs.mistral.ai/resources/changelogs", "type": "announcement" }, { "label": "Publisher model lifecycle", "url": "https://legal.mistral.ai/ai-governance/models/magistral-medium-1/", "type": "docs" }, { "label": "Original release announcement", "url": "https://mistral.ai/news/magistral/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:46:35.789Z", "last_seen": "2026-10-29T10:46:35.789Z", "upstream_present": false, "availability": "public", "notes": "Original Mistral reasoning model served through its API; this version is now retired or deprecated. Parameter count and internal architecture are undisclosed. Context is the publisher’s rounded 40k API figure. " }, { "id": "mistral/magistral-small-2506", "slug": "mistral--magistral-small-2506", "name": "Magistral Small 1.0", "developer": "Mistral", "developer_id": "mistral", "family": "magistral", "description": "Text reasoning model based on Mistral Small 3.1, trained with supervised reasoning traces and reinforcement learning.", "release_date": "2025-06-10", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 24, "active_parameters_b": 24, "parameters_source": "primary_source", "architecture": "dense", "context_window": 40960, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://mistral.ai/news/magistral/", "source_type": "primary", "sources": [ { "label": "Magistral — release announcement", "url": "https://mistral.ai/news/magistral/", "type": "primary" }, { "label": "Magistral Small 1.0 — pinned model card", "url": "https://huggingface.co/mistralai/Magistral-Small-2506/blob/ad2fce5b4113139e1877dcadfd826c9262ad8e8c/README.md", "type": "weights" }, { "label": "Magistral Small 1.0 — exact model reference and Weights table", "url": "https://docs.mistral.ai/models/magistral-small-1-0-25-06", "type": "primary" }, { "label": "Magistral Small 2506 — pinned Transformers configuration", "url": "https://huggingface.co/mistralai/Magistral-Small-2506/blob/ad2fce5b4113139e1877dcadfd826c9262ad8e8c/config.json", "type": "weights" }, { "label": "Magistral Small 2506 — pinned native configuration", "url": "https://huggingface.co/mistralai/Magistral-Small-2506/blob/ad2fce5b4113139e1877dcadfd826c9262ad8e8c/params.json", "type": "weights" }, { "label": "Magistral Small 2506 — pinned checkpoint tensor index", "url": "https://huggingface.co/mistralai/Magistral-Small-2506/blob/ad2fce5b4113139e1877dcadfd826c9262ad8e8c/model.safetensors.index.json", "type": "weights" }, { "label": "Magistral Small 1.0 — official Markdown reference", "url": "https://docs.mistral.ai/models/magistral-small-1-0-25-06.md", "type": "primary" }, { "label": "Magistral — technical report", "url": "https://arxiv.org/html/2506.10910v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/magistral-small-2506.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Magistral-Small-2506", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Mistral explicitly reports nominal 24B total and 24B active parameters. The released configurations specify 40,960 positions, matching the official 40k reference and model-card recommendation; the card mentions 128k nominal capacity but warns of quality degradation beyond 40k. Text-only released checkpoint with reasoning traces and native function calling. Mistral documents API retirement on November 30, 2025; public weights remain downloadable.", "field_status": {}, "audited_at": "2026-10-26T22:54:31.418003Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/318.json" }, { "id": "openai/o3-pro", "slug": "openai--o3-pro", "name": "o3-pro", "developer": "OpenAI", "developer_id": "openai", "family": "o3", "description": "OpenAI o3 reasoning variant that uses more inference compute, with text and image input and function calling.", "release_date": "2025-06-10", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 200000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://help.openai.com/en/articles/9624314-model-release-notes", "source_type": "primary", "sources": [ { "label": "Model Release Notes — o3-pro, June 10, 2025", "url": "https://help.openai.com/en/articles/9624314-model-release-notes", "type": "primary" }, { "label": "o3-pro model reference", "url": "https://developers.openai.com/api/docs/models/o3-pro", "type": "primary" }, { "label": "API changelog — June 10, 2025", "url": "https://developers.openai.com/api/docs/changelog", "type": "primary" }, { "label": "OpenAI o3 and o4-mini System Card", "url": "https://cdn.openai.com/pdf/2221c875-02dc-4789-800b-e7758f3722c1/o3-and-o4-mini-system-card.pdf", "type": "primary" }, { "label": "o3 model reference", "url": "https://developers.openai.com/api/docs/models/o3", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201", "type": "primary" }, { "label": "GPT-5 System Card", "url": "https://cdn.openai.com/gpt-5-system-card.pdf", "type": "primary" }, { "label": "gpt-oss-120b & gpt-oss-20b Model Card", "url": "https://arxiv.org/pdf/2508.10925", "type": "primary" }, { "label": "Detecting and reducing scheming in AI models", "url": "https://openai.com/index/detecting-and-reducing-scheming-in-ai-models/", "type": "primary" }, { "label": "Toward understanding and preventing misalignment generalization", "url": "https://openai.com/index/emergent-misalignment/", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs", "type": "primary" }, { "label": "API deprecations — GPT-5 and o3 model deprecations", "url": "https://developers.openai.com/api/docs/deprecations", "type": "primary" }, { "label": "Introducing GPT-5", "url": "https://openai.com/index/introducing-gpt-5/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/o3-pro.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The o3-pro alias points to o3-pro-2025-06-10. OpenAI explicitly identifies the same underlying o3 model with additional inference compute. Neither nominal total nor active parameters is disclosed for o3 or o3-pro; the published ~600B third-party estimate lacks a substantiated counting basis. Transformer is a broad inference from that binding, exact o3 prompt-caching support and OpenAI’s applicable pre-GPT-5.6 engineering explanation. It does not imply prompt caching in the o3-pro API or disclose dense/MoE topology. The API supports Responses and Batch, with 200,000 context tokens and 100,000 maximum output tokens. PDF is an API input format converted into text and page images; hosted image generation is not native image output. GPT-5 pro replaced o3-pro in ChatGPT in August 2025. The o3-pro API is deprecated with removal scheduled for December 11, 2026; endpoint operation was not tested.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T17:23:12.822668Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/456.json" }, { "id": "google/gemini-2.5-pro", "slug": "google--gemini-2.5-pro", "name": "Gemini 2.5 Pro", "developer": "Google", "developer_id": "google", "family": "gemini-pro", "description": "Google’s multimodal reasoning model with adaptive thinking, function calling, and support for coding, mathematics, and document analysis.", "release_date": "2025-06-05", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "audio", "video", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-pro", "source_type": "primary", "sources": [ { "label": "Gemini 2.5 Pro Model Card (June 27, 2025)", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Pro-Model-Card.pdf", "type": "primary" }, { "label": "Gemini 2.5 Pro — Gemini API", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-pro", "type": "primary" }, { "label": "Gemini API release notes", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "primary" }, { "label": "Gemini 2.5: Updates to our family of thinking models", "url": "https://developers.googleblog.com/en/gemini-2-5-thinking-model-updates/", "type": "primary" }, { "label": "Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Gemini 2.5 technical report", "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_v2_5_report.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemini-2.5-pro.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The GA model first appeared as the June 5, 2025 preview; Google explicitly states the June 17 stable release uses the same model. March 25 and May 6 were earlier revisions. Sparse MoE Transformer; total and active parameter counts are undisclosed in the reviewed sources. Context is the documented 1,048,576-token input limit; output limit is separate.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T12:55:05.297987+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/198.json" }, { "id": "google/gemini-2.5-pro-preview", "slug": "google--gemini-2.5-pro-preview", "name": "Gemini 2.5 Pro Preview 06-05", "developer": "Google", "developer_id": "google", "family": "gemini-pro", "description": "Google's June 2025 preview of Gemini 2.5 Pro, a multimodal thinking model with function calling and text output.", "release_date": "2025-06-05", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "audio", "video", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ai.google.dev/gemini-api/docs/changelog", "source_type": "primary", "sources": [ { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemini-2.5-pro-preview.toml", "type": "catalog" }, { "label": "Gemini API release notes", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "primary" }, { "label": "Gemini 2.5: Updates to our family of thinking models (June 17, 2025)", "url": "https://developers.googleblog.com/en/gemini-2-5-thinking-model-updates/", "type": "primary" }, { "label": "Gemini 2.5 Pro Model Card (updated June 27, 2025)", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Pro-Model-Card.pdf", "type": "primary" }, { "label": "Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities", "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_v2_5_report.pdf", "type": "primary" }, { "label": "Gemini API model details: indexed June 2025 page with preview versions", "url": "https://ai.google.dev/gemini-api/docs/models/gemini?hl=sl", "type": "primary" }, { "label": "Gemini 2.5 Pro: Gemini API model reference", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-pro", "type": "primary" }, { "label": "Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Gemini API deprecations", "url": "https://ai.google.dev/gemini-api/docs/deprecations/", "type": "primary" }, { "label": "Google responses to the December 2025 Foundation Model Transparency Index", "url": "https://crfm.stanford.edu/fmti/December-2025/company-reports/Google_FinalReport_FMTI2025.html", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "This catalog key identifies the June 5, 2025 preview; the official API ID was gemini-2.5-pro-preview-06-05. Google promoted it unchanged to stable on June 17. Sparse MoE Transformer; neither total nor active parameter count is disclosed in the reviewed sources. The dated preview was shut down December 2, 2025. Earlier March and May revisions are distinct.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T12:55:34.816739+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/199.json" }, { "id": "deepseek/deepseek-r1-0528", "slug": "deepseek--deepseek-r1-0528", "name": "DeepSeek R1 0528", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-thinking", "description": "May 2025 DeepSeek R1 update with improved mathematics, coding, and reasoning", "release_date": "2025-05-28", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 682.636487424, "active_parameters_b": 37, "parameters_source": "primary_source", "architecture": "moe", "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://deepseek.com/news/r1-0528/", "source_type": "primary", "sources": [ { "label": "DeepSeek May 28 release announcement", "url": "https://deepseek.com/news/r1-0528/", "type": "primary" }, { "label": "Exact R1-0528 model card", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528/blob/4236a6af538feda4548eca9ab308586007567f52/README.md", "type": "primary" }, { "label": "Exact checkpoint configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528/blob/4236a6af538feda4548eca9ab308586007567f52/config.json", "type": "primary" }, { "label": "DeepSeek V3 base architecture and nominal counts", "url": "https://github.com/deepseek-ai/DeepSeek-V3", "type": "primary" }, { "label": "DeepSeek weight structure and shared MTP documentation", "url": "https://github.com/deepseek-ai/DeepSeek-V3/blob/main/README_WEIGHTS.md", "type": "primary" }, { "label": "Pinned R1-0528 tensor index and all163 bounded headers", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528/resolve/4236a6af538feda4548eca9ab308586007567f52/model.safetensors.index.json", "type": "weights" }, { "label": "model.norm.weight bounded tensor comparison", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528/resolve/4236a6af538feda4548eca9ab308586007567f52/model-00160-of-000163.safetensors", "type": "weights" }, { "label": "model.layers.61.shared_head.norm.weight bounded tensor comparison", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528/resolve/4236a6af538feda4548eca9ab308586007567f52/model-00163-of-000163.safetensors", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepseek/deepseek-r1-0528.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete count: 671.0264192B main + 11.610068224B unique MTP; shared embedding/head counted once and FP8 scales excluded. The MTP norm differs from the main norm. Nominal 37B active applies to the main model. Advertised checkpoint context is 128K; config positions are 163840 and launch API context was 64K. Function calling is supported outside thinking.", "parameter_components_b": { "main": 671.0264192, "mtp_unique": 11.610068224 }, "active_parameters_scope": "main", "active_parameters_by_phase_b": { "main": 37 }, "field_status": {}, "audited_at": "2026-10-26T19:43:18.905509+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/169.json" }, { "id": "anthropic/claude-opus-4-20250514", "slug": "anthropic--claude-opus-4-20250514", "name": "Claude Opus 4", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude-opus", "description": "Claude hybrid reasoning model for coding, complex reasoning, and long-running agent tasks", "release_date": "2025-05-22", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 200000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.anthropic.com/news/claude-4", "source_type": "primary", "sources": [ { "label": "Claude 4 launch", "url": "https://www.anthropic.com/news/claude-4", "type": "primary" }, { "label": "Claude 4 system card", "url": "https://www.anthropic.com/claude-4-system-card", "type": "primary" }, { "label": "Exact-model interpretability study", "url": "https://transformer-circuits.pub/2025/introspection/index.html", "type": "primary" }, { "label": "Commercial terms", "url": "https://www.anthropic.com/legal/commercial-terms", "type": "primary" }, { "label": "Claude weight distribution policy", "url": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities", "type": "primary" }, { "label": "Opus 4 provider specifications", "url": "https://cloud.google.com/vertex-ai/generative-ai/docs/partner-models/claude/opus-4", "type": "docs" }, { "label": "Opus 4 PDF support", "url": "https://aws.amazon.com/about-aws/whats-new/2025/06/citations-api-pdf-claude-models-amazon-bedrock/", "type": "primary" }, { "label": "Model deprecations", "url": "https://docs.anthropic.com/en/docs/about-claude/model-deprecations", "type": "docs" }, { "label": "Anthropic pricing table with conflicting context tier", "url": "https://www-cdn.anthropic.com/files/4zrzovbb/website/5678bc2f5978e5bcd4f1fe7c14b2c72284dcf9f8.pdf", "type": "documentation" }, { "label": "Opus 4.1 model relationship", "url": "https://www.anthropic.com/claude-opus-4-1-system-card", "type": "primary" }, { "label": "Later Opus 4.5 system card", "url": "https://www.anthropic.com/document/claude-opus-4-5-system-card", "type": "primary" }, { "label": "Opus 4 continuity and lineage", "url": "https://alignment.anthropic.com/2025/sabotage-risk-report/2025_pilot_risk_report.pdf", "type": "primary" }, { "label": "Later Opus 4.6 comparison card", "url": "https://www.anthropic.com/claude-opus-4-6-system-card", "type": "primary" }, { "label": "Later Opus 5.5 comparison card", "url": "https://www.anthropic.com/claude-opus-5-5-system-card", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/anthropic/claude-opus-4-20250514.toml", "type": "catalog" }, { "label": "Alias documentation", "url": "https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions", "type": "docs" } ], "aliases": [ "anthropic/claude-opus-4-0" ], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Transformer structure is supported by interpretability research on production Opus 4; dense/MoE topology is undisclosed. No developer-published total or active count, or size-bearing base-model lineage, was established. PDF is a supported document format. The dated snapshot was retired from the Claude API on June 15, 2026; documented specifications do not establish current endpoint operation. An Anthropic pricing PDF effective May 12, 2026 also lists a 1M API tier, conflicting with the retained 200K model specification; availability of that tier was not verified.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T10:35:23.941208Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/105.json", "alias_sources": { "anthropic/claude-opus-4-0": { "url": "https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions", "note": "Alias or alternate catalog name for the same released model; only the retained ID counts as a release." } } }, { "id": "anthropic/claude-sonnet-4-20250514", "slug": "anthropic--claude-sonnet-4-20250514", "name": "Claude Sonnet 4", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude-sonnet", "description": "Claude hybrid reasoning model for coding, visual analysis, and agent workflows", "release_date": "2025-05-22", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 200000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.anthropic.com/news/claude-4", "source_type": "primary", "sources": [ { "label": "Introducing Claude 4", "url": "https://www.anthropic.com/news/claude-4", "type": "primary" }, { "label": "System Card: Claude Opus 4 & Claude Sonnet 4", "url": "https://www.anthropic.com/claude-4-system-card", "type": "primary" }, { "label": "Emergent Introspective Awareness in Large Language Models", "url": "https://transformer-circuits.pub/2025/introspection/index.html", "type": "primary" }, { "label": "Commercial Terms of Service", "url": "https://www.anthropic.com/legal/commercial-terms", "type": "primary" }, { "label": "GLM-5.3 and the spread of advanced cyber capabilities", "url": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities", "type": "primary" }, { "label": "Claude Platform release notes", "url": "https://platform.claude.com/docs/en/release-notes/overview", "type": "docs" }, { "label": "Claude Sonnet 4 - Amazon Bedrock", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-anthropic-claude-sonnet-4.html", "type": "docs" }, { "label": "Citations API and PDF support for Claude models now in Amazon Bedrock", "url": "https://aws.amazon.com/about-aws/whats-new/2025/06/citations-api-pdf-claude-models-amazon-bedrock/", "type": "primary" }, { "label": "System Card Addendum: Claude Opus 4.1", "url": "https://www-cdn.anthropic.com/9fa30625273bafdf5af82c93719d7ca606485a16.pdf", "type": "primary" }, { "label": "System Card: Claude Sonnet 4.5", "url": "https://www-cdn.anthropic.com/963373e433e489a87a10c823c52a0a013e9172dd.pdf", "type": "primary" }, { "label": "System Card: Claude Sonnet 4.6", "url": "https://www.anthropic.com/claude-sonnet-4-6-system-card", "type": "primary" }, { "label": "Model deprecations", "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "type": "docs" }, { "label": "Claude Sonnet 4 on Google Cloud", "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/partner-models/claude/sonnet-4?hl=en", "type": "docs" }, { "label": "Anthropic Transparency Hub - Claude Opus 4 and Claude Sonnet 4 summary", "url": "https://www.anthropic.com/transparency", "type": "primary" }, { "label": "Teaching Claude why", "url": "https://www.anthropic.com/research/teaching-claude-why", "type": "primary" }, { "label": "Claude Sonnet 5 System Card", "url": "https://www.anthropic.com/claude-sonnet-5-system-card", "type": "primary" }, { "label": "Claude Sonnet 5.5 System Card", "url": "https://www.anthropic.com/claude-sonnet-5-5-system-card", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/anthropic/claude-sonnet-4-20250514.toml", "type": "catalog" }, { "label": "Alias documentation", "url": "https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions", "type": "docs" } ], "aliases": [ "anthropic/claude-sonnet-4-0" ], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Production Sonnet 4 interpretability research supports broad Transformer structure; dense/MoE topology and nominal total/active parameter counts remain undisclosed. Research identifies its pretrained base but does not publish its size. PDF is a supported document input format. The standard context is 200K; the 1M beta launched August 12, 2025 and was retired on the Claude API April 30, 2026. Retained Google documentation lists 1M Preview/200K GA, but current preview operation was not verified. The dated model retired from the Claude API June 15, 2026 and reached Bedrock end of life October 14, 2026. Historical specifications do not establish current endpoint availability.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T10:47:12.387402+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/112.json", "alias_sources": { "anthropic/claude-sonnet-4-0": { "url": "https://platform.claude.com/docs/en/about-claude/models/model-ids-and-versions", "note": "Alias or alternate catalog name for the same released model; only the retained ID counts as a release." } } }, { "id": "google/gemini-2.5-flash", "slug": "google--gemini-2.5-flash", "name": "Gemini 2.5 Flash", "developer": "Google", "developer_id": "google", "family": "gemini-flash", "description": "Multimodal Gemini model with configurable thinking, a 1,048,576-token context window, and function calling.", "release_date": "2025-05-20", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "audio", "video", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-flash", "source_type": "primary", "sources": [ { "label": "Google model documentation", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.5-flash", "type": "primary" }, { "label": "Google release notes", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "primary" }, { "label": "Gemini 2.5 technical report", "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_v2_5_report.pdf", "type": "primary" }, { "label": "Gemini 2.5 Flash model card", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Flash-Model-Card.pdf", "type": "primary" }, { "label": "Google Cloud model specifications", "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/2-5-flash", "type": "primary" }, { "label": "Gemini API terms", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemini-2.5-flash.toml", "type": "catalog" }, { "label": "Gemini 2.5: Updates to our family of thinking models", "url": "https://developers.googleblog.com/en/gemini-2-5-thinking-model-updates/", "type": "primary" }, { "label": "Gemini 3 Flash model card comparisons", "url": "https://deepmind.google/models/model-cards/gemini-3-flash/", "type": "primary" }, { "label": "Gemini 4 Argon Safety & Security Report", "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Sparse MoE Transformer; Google does not disclose total or active parameter counts in the reviewed sources. First released as the May 20, 2025 preview; Google explicitly states the June 17 stable release uses the same model. The April 17 preview was an earlier revision.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T12:48:12Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/193.json" }, { "id": "google/gemini-2.5-flash-preview-05-20", "slug": "google--gemini-2.5-flash-preview-05-20", "name": "Gemini 2.5 Flash Preview 05-20", "developer": "Google", "developer_id": "google", "family": "gemini-flash", "description": "May 2025 Gemini 2.5 Flash preview with controllable thinking budgets; promoted unchanged to GA, with this preview endpoint now retired.", "release_date": "2025-05-20", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "audio", "video", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://ai.google.dev/gemini-api/docs/changelog", "source_type": "primary", "sources": [ { "label": "Gemini API release notes", "url": "https://ai.google.dev/gemini-api/docs/changelog", "type": "primary" }, { "label": "Gemini 2.5: Updates to our family of thinking models (June 17, 2025)", "url": "https://developers.googleblog.com/gemini-2-5-thinking-model-updates/", "type": "primary" }, { "label": "Gemini 2.5 technical report", "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_v2_5_report.pdf", "type": "primary" }, { "label": "Gemini 2.5 Flash model card (December 2025 revision)", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Flash-Model-Card.pdf", "type": "primary" }, { "label": "Gemini models (indexed June 2025 documentation)", "url": "https://ai.google.dev/gemini-api/docs/models/gemini?hl=sl", "type": "primary" }, { "label": "Intro to Gemini 2.5 Flash (pinned pre-GA notebook)", "url": "https://github.com/GoogleCloudPlatform/generative-ai/blob/2bb6513c6cfa06a236c14ff90d46edfc9c4a833e/gemini/getting-started/intro_gemini_2_5_flash.ipynb", "type": "primary" }, { "label": "Gemini API deprecations", "url": "https://ai.google.dev/gemini-api/docs/deprecations", "type": "primary" }, { "label": "Vertex AI release notes", "url": "https://docs.cloud.google.com/vertex-ai/generative-ai/docs/release-notes", "type": "primary" }, { "label": "Gemini API Additional Terms of Service", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemini-2.5-flash-preview-05-20.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Google identifies the May 20, 2025 preview and June 17 GA release as the same model. Sparse MoE Transformer; total and active parameter counts are undisclosed in the reviewed sources. The 1,048,576-token input limit is historical. The Gemini API preview shut down November 18, 2025; Vertex AI ended preview availability after July 15. Native-audio, TTS, image and September previews are distinct variants.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T12:53:25.362679Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/197.json" }, { "id": "upstage/solar-pro2", "slug": "upstage--solar-pro2", "name": "Solar Pro 2", "developer": "Upstage", "developer_id": "upstage", "family": "solar-pro", "description": "Upstage text model with chat and reasoning modes, tool calling, and support for English, Korean, and Japanese.", "release_date": "2025-05-20", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 30.9, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": null, "context_window": 65536, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.upstage.ai/blog/en/solar-pro-2-preview-small-powerful-now-with-reasoning", "source_type": "primary", "sources": [ { "label": "Solar Pro 2: model specifications and versions", "url": "https://console.upstage.ai/docs/models/solar-pro-2", "type": "primary" }, { "label": "Solar Pro 2 Preview: Small. Powerful. Now with reasoning.", "url": "https://www.upstage.ai/blog/en/solar-pro-2-preview-small-powerful-now-with-reasoning", "type": "primary" }, { "label": "Solar Pro 2: Fluent. Reasoning. Frontier.", "url": "https://www.upstage.ai/blog/en/solar-pro-2-launch", "type": "primary" }, { "label": "Solar Pro 2 Breaks Into Global Frontier AI", "url": "https://www.upstage.ai/news/solar-pro-2-frontier", "type": "primary" }, { "label": "Compare models", "url": "https://console.upstage.ai/docs/models/compare/generative-intelligence", "type": "primary" }, { "label": "Solar Pro2 Prompting Handbook (English)", "url": "https://github.com/UpstageAI/cookbook/blob/main/solar-pro2-prompting-handbook/%5BEN%5D%20Solar%20Pro2%20Prompting%20Handbook.ipynb", "type": "primary" }, { "label": "Solar Pro 2 final-launch language benchmark table", "url": "https://cdn.prod.website-files.com/6743d5190bb2b52f38e99ecd/686ea3743cbacdffaf647d68_solar%20pro2_table_language(en).avif", "type": "primary" }, { "label": "Solar Pro 2 final-launch frontier benchmark table", "url": "https://cdn.prod.website-files.com/6743d5190bb2b52f38e99ecd/686ea3dcc3314308ea148884_solar%20pro2_table_frontiers%20(en).avif", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/upstage/solar-pro2.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "30.9B is Upstage’s nominal total in both final-launch benchmark tables, rounded to 31B in its model documentation; active parameters and neural architecture are not separately disclosed. The release date is the May 20, 2025 public preview; general availability followed on July 10. The rolling solar-pro2 alias currently maps to solar-pro2-251215. Upstage documents retirement on October 30, 2026; specifications do not establish current endpoint operation.", "field_status": { "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T18:13:25.019659Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/516.json" }, { "id": "mistral/mistral-medium-2505", "slug": "mistral--mistral-medium-2505", "name": "Mistral Medium 3", "developer": "Mistral", "developer_id": "mistral", "family": "mistral-medium", "description": "May 2025 proprietary Mistral Medium 3 instruction model with multilingual text, image understanding, coding and native function calling", "release_date": "2025-05-07", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://mistral.ai/news/mistral-medium-3/", "source_type": "primary", "sources": [ { "label": "Mistral Medium 3 original release", "url": "https://mistral.ai/news/mistral-medium-3/", "type": "primary" }, { "label": "Alias documentation", "url": "https://docs.mistral.ai/models/mistral-medium-3-25-05", "type": "docs" }, { "label": "Original NVIDIA NIM deployment announcement", "url": "https://developer.nvidia.com/blog/build-agents-and-understand-long-docs-with-mistral-medium-3-and-nvidia-nim/", "type": "primary" }, { "label": "Mistral Magistral technical report", "url": "https://arxiv.org/html/2506.10910v1", "type": "primary" }, { "label": "Mistral Medium lifecycle and governance", "url": "https://legal.mistral.ai/ai-governance/models/mistral-medium-3/", "type": "primary" }, { "label": "IBM exact Medium and Devstral model specifications", "url": "https://www.ibm.com/docs/en/watsonx/w-and-w/2.4.x?topic=models-third-party-foundation", "type": "primary" }, { "label": "Devstral Medium 2507 exact model card", "url": "https://docs.mistral.ai/models/devstral-medium-1-0-25-07", "type": "primary" }, { "label": "Phoenix report identifies later Medium 3.1 lineage", "url": "https://arxiv.org/html/2605.10391v1", "type": "primary" }, { "label": "OpenRouter original Medium 3 hosted-route specifications", "url": "https://openrouter.ai/api/v1/models/mistralai/mistral-medium-3/endpoints", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/mistral-medium-2505.toml", "type": "catalog" } ], "aliases": [ "mistral/mistral-medium-3" ], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Original mistral-medium-2505. Total and active counts remain undisclosed: IBM's 48B text-only Devstral fine-tune does not establish retention of Medium 3's visual components; Phoenix's 123B lineage names later Medium 3.1. NVIDIA documents dense architecture and private licensing. Context 131072 is an OpenRouter route specification; Mistral advertises nominal 128k and Azure lists 128000. Native enforcement and dedicated-thinking capability remain unverified.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "reasoning": "unverified" }, "audited_at": "2026-10-27T15:38:31Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/337.json", "alias_sources": { "mistral/mistral-medium-3": { "url": "https://docs.mistral.ai/models/mistral-medium-3-25-05", "note": "Alias or alternate catalog name for the same released model; only the retained ID counts as a release." } } }, { "id": "deepseek/deepseek-r1-0528-qwen3-8b", "slug": "deepseek--deepseek-r1-0528-qwen3-8b", "name": "DeepSeek R1 0528 Qwen3 8B", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-thinking", "description": "Dense reasoning model obtained by post-training Qwen3-8B Base on chain-of-thought outputs from DeepSeek-R1-0528.", "release_date": "2025-05", "date_precision": "month", "open_weights": true, "license": "MIT", "parameters_b": 8.19073536, "active_parameters_b": 8, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepseek/deepseek-r1-0528-qwen3-8b.toml", "source_type": "catalog", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B", "type": "weights" }, { "label": "Exact checkpoint config.json", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B/blob/6e8885a6ff5c1dc5201574c8fd700323f23c25fa/config.json", "type": "primary" }, { "label": "Exact checkpoint safetensors index", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B/blob/6e8885a6ff5c1dc5201574c8fd700323f23c25fa/model.safetensors.index.json", "type": "weights" }, { "label": "Official checkpoint shard 1 tensor header", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B/resolve/6e8885a6ff5c1dc5201574c8fd700323f23c25fa/model-00001-of-000002.safetensors", "type": "weights" }, { "label": "Official checkpoint shard 2 tensor header", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B/resolve/6e8885a6ff5c1dc5201574c8fd700323f23c25fa/model-00002-of-000002.safetensors", "type": "weights" }, { "label": "Exact checkpoint tokenizer configuration and chat template", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B/blob/6e8885a6ff5c1dc5201574c8fd700323f23c25fa/tokenizer_config.json", "type": "primary" }, { "label": "DeepSeek checkpoint MIT license", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B/blob/6e8885a6ff5c1dc5201574c8fd700323f23c25fa/LICENSE", "type": "primary" }, { "label": "Underlying Qwen3-8B-Base model card", "url": "https://huggingface.co/Qwen/Qwen3-8B-Base", "type": "primary" }, { "label": "DeepSeek R1-0528 announcement", "url": "https://deepseek.com/news/r1-0528/", "type": "primary" }, { "label": "Official checkpoint repository commit history", "url": "https://huggingface.co/api/models/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B/commits/main", "type": "primary" }, { "label": "Exact checkpoint tokenizer vocabulary", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-0528-Qwen3-8B/blob/6e8885a6ff5c1dc5201574c8fd700323f23c25fa/tokenizer.json", "type": "primary" }, { "label": "Qwen3 Technical Report - activated-parameter comparison tables", "url": "https://arxiv.org/html/2505.09388v1", "type": "primary" }, { "label": "Official Qwen3-8B Base configuration", "url": "https://huggingface.co/Qwen/Qwen3-8B-Base/raw/main/config.json", "type": "primary" }, { "label": "DeepSeek-R1 revised technical report - later comparison check", "url": "https://arxiv.org/html/2501.12948v2", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepseek/deepseek-r1-0528-qwen3-8b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete checkpoint: 8,190,735,360 learned parameters. Nominal 8B active is inferred from documented Qwen3-8B post-training lineage, identical architecture and Qwen’s activated-parameter tables. Qwen3 dense architecture; 131072 configured context uses 4x YaRN from 32768 native positions. DeepSeek release declares MIT; underlying Qwen3-8B Base declares Apache-2.0. Release day is unresolved between a May 28 dated announcement and May 29 checkpoint upload, so only May 2025 is asserted. Tool-call capability of this exact distillation remains unverified despite template support; full-R1 tool benchmarks do not establish it.", "field_status": { "tool_call": "unverified" }, "audited_at": "2026-10-27T11:00:55.976004+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/170.json" }, { "id": "sarvam/sarvam-m", "slug": "sarvam--sarvam-m", "name": "Sarvam-M", "developer": "Sarvam AI", "developer_id": "sarvam", "family": "sarvam", "description": "Text-only Mistral Small derivative post-trained for English and ten Indian languages, with switchable thinking and non-thinking modes.", "release_date": "2025-05", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 23.5724032, "active_parameters_b": 24, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.sarvam.ai/blogs/sarvam-m", "source_type": "primary", "sources": [ { "label": "Sarvam-M technical launch", "url": "https://www.sarvam.ai/blogs/sarvam-m", "type": "primary" }, { "label": "Sarvam-M exact checkpoint", "url": "https://huggingface.co/sarvamai/sarvam-m/blob/01534a53c46f2788e392dbb3d994e0fa8f04d3fd/README.md", "type": "weights" }, { "label": "Sarvam-M configuration", "url": "https://huggingface.co/sarvamai/sarvam-m/blob/01534a53c46f2788e392dbb3d994e0fa8f04d3fd/config.json", "type": "primary" }, { "label": "Sarvam-M current API documentation", "url": "https://docs.sarvam.ai/api/getting-started/models/sarvam-m", "type": "primary" }, { "label": "Sthanika original MILU 2026 active-count convention", "url": "https://sthanika.ai/research/milu-2026", "type": "primary" }, { "label": "Sthanika exact Sarvam-M checkpoint configuration", "url": "https://github.com/sthanika-ai/milu-llm-evaluation/blob/main/configs/models/sarvam-m-thinkmode-0shot-json.yaml", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/sarvam/sarvam-m.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/sarvamai/sarvam-m", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Exact original checkpoint: 23.5724032B learned parameters; all 363 BF16 tensor headers reconcile with configuration and implementation. Nominal 24B active follows Sthanika AI's original MILU 2026 comparison convention for this checkpoint, not a Sarvam disclosure or an exact per-token tensor count. Checkpoint context is 32768; historical Sarvam hosting used 8192 and the API is now retired. Tool use is demonstrated as model-selected Wikipedia lookup in a custom research system; the shipped template has no standard tool-role/function-call serialization. May 2025 release is established; May 20 upload, May 23 launch and the migrated May 1 changelog do not resolve the exact first-public day.", "parameter_components_b": { "input_embeddings": 0.67108864, "decoder_layers": 22.2302208, "final_norm": 0.00000512, "output_head": 0.67108864 }, "field_status": {}, "audited_at": "2026-10-27T17:56:31.584577Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/492.json" }, { "id": "amazon/nova-premier", "slug": "amazon--nova-premier", "name": "Nova Premier", "developer": "Amazon", "developer_id": "amazon", "family": "nova", "description": "Multimodal model for complex analysis, long-context understanding, tool use, and model distillation", "release_date": "2025-04-30", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 1000000, "modalities": { "input": [ "text", "image", "video", "pdf" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://aws.amazon.com/blogs/aws/amazon-nova-premier-our-most-capable-model-for-complex-tasks-and-teacher-for-model-distillation/", "source_type": "primary", "sources": [ { "label": "Amazon Nova Premier: Our most capable model for complex tasks and teacher for model distillation", "url": "https://aws.amazon.com/blogs/aws/amazon-nova-premier-our-most-capable-model-for-complex-tasks-and-teacher-for-model-distillation/", "type": "primary" }, { "label": "Amazon Nova Premier: Technical Report and Model Card", "url": "https://cdn.amazon.science/e5/e6/ccc5378c42dca467d1abe1628ec9/amazon-nova-premier-technical-report-and-model-card.pdf", "type": "primary" }, { "label": "Amazon Nova 2: Multimodal Reasoning and Generation Models", "url": "https://cdn.amazon.science/c5/3d/84514a224666b5be6de4b43ef4aa/nova-2-0-technical-report2.pdf", "type": "primary" }, { "label": "Amazon Nova Micro, Lite, Pro, and Premier — AWS AI Service Cards", "url": "https://docs.aws.amazon.com/ai/responsible-ai/nova-micro-lite-pro/overview.html", "type": "primary" }, { "label": "What is Amazon Nova? — Version 1 model specifications", "url": "https://docs.aws.amazon.com/nova/latest/userguide/what-is-nova.html", "type": "primary" }, { "label": "Complete request schema — Amazon Nova", "url": "https://docs.aws.amazon.com/nova/latest/userguide/complete-request-schema.html", "type": "primary" }, { "label": "Document understanding — Amazon Nova Version 1", "url": "https://docs.aws.amazon.com/nova/latest/userguide/modalities-document.html", "type": "primary" }, { "label": "Iterative Training — Amazon Nova", "url": "https://docs.aws.amazon.com/nova/latest/userguide/nova-iterative-training.html", "type": "primary" }, { "label": "AWS Service Terms", "url": "https://aws.amazon.com/service-terms/", "type": "primary" }, { "label": "MASSIVE-Agents: A Benchmark for Multilingual Function-Calling in 52 Languages", "url": "https://aclanthology.org/2025.findings-emnlp.1099.pdf", "type": "primary" }, { "label": "Evaluating the Critical Risks of Amazon’s Nova Premier under the Frontier Model Safety Framework", "url": "https://cdn.amazon.science/59/0e/fd60057640a4a02bcd3072029958/evaluating-the-critical-risks-of-amazons-nova-premier-under-the-frontier-model-safety-framework.pdf", "type": "primary" }, { "label": "Customize Amazon Nova models with Amazon Bedrock fine-tuning", "url": "https://aws.amazon.com/blogs/machine-learning/customize-amazon-nova-models-with-amazon-bedrock-fine-tuning/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/amazon/nova-premier.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Transformer-based; total and active parameters remain undisclosed. Amazon AGI’s MASSIVE-Agents Table 1 lists Premier’s Params as NP; no separate active count is published. Dense/MoE topology is undisclosed. Dedicated extended-thinking support is unresolved: the current Nova v1 reasoningConfig description excludes Premier, but an exact-model Bedrock card preserved in the search index says Reasoning Supported; its live URL now redirects. General reasoning ability does not resolve the conflict. The explicit April 30, 2025 GA announcement establishes initial release, despite the indexed card claiming a later launch date. This audit verifies documented specifications, not current endpoint availability.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "reasoning": "unverified" }, "audited_at": "2026-10-27T10:07:17.017Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/082.json" }, { "id": "microsoft/phi-4-mini-reasoning", "slug": "microsoft--phi-4-mini-reasoning", "name": "Phi-4-mini-reasoning", "developer": "Microsoft", "developer_id": "microsoft", "family": "phi", "description": null, "release_date": "2025-04-30", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 3.8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://azure.microsoft.com/en-us/blog/one-year-of-phi-small-language-models-making-big-leaps-in-ai/", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://azure.microsoft.com/en-us/blog/one-year-of-phi-small-language-models-making-big-leaps-in-ai/", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/microsoft/Phi-4-mini-reasoning/blob/0e3b1e2d02ee478a3743abe3f629e9c0cb722e0a/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/microsoft/Phi-4-mini-reasoning/blob/0e3b1e2d02ee478a3743abe3f629e9c0cb722e0a/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "The April 30 developer announcement explicitly releases this checkpoint. Parameter and supported-context figures are from the model card." }, { "id": "microsoft/phi-4-reasoning", "slug": "microsoft--phi-4-reasoning", "name": "Phi-4-reasoning", "developer": "Microsoft", "developer_id": "microsoft", "family": "phi", "description": null, "release_date": "2025-04-30", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 14, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://azure.microsoft.com/en-us/blog/one-year-of-phi-small-language-models-making-big-leaps-in-ai/", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://azure.microsoft.com/en-us/blog/one-year-of-phi-small-language-models-making-big-leaps-in-ai/", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/microsoft/Phi-4-reasoning/blob/1de18ec97600877ce63dbf60c73b998da99f0195/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/microsoft/Phi-4-reasoning/blob/1de18ec97600877ce63dbf60c73b998da99f0195/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "The April 30 developer announcement explicitly releases this checkpoint. Parameter and supported-context figures are from the model card." }, { "id": "microsoft/phi-4-reasoning-plus", "slug": "microsoft--phi-4-reasoning-plus", "name": "Phi-4-reasoning-plus", "developer": "Microsoft", "developer_id": "microsoft", "family": "phi", "description": null, "release_date": "2025-04-30", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 14, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://azure.microsoft.com/en-us/blog/one-year-of-phi-small-language-models-making-big-leaps-in-ai/", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://azure.microsoft.com/en-us/blog/one-year-of-phi-small-language-models-making-big-leaps-in-ai/", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/microsoft/Phi-4-reasoning-plus/blob/69baf8528e1bcf05f475034d9e5dd32875ed125f/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/microsoft/Phi-4-reasoning-plus/blob/69baf8528e1bcf05f475034d9e5dd32875ed125f/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "The April 30 developer announcement explicitly releases this checkpoint. Parameter and supported-context figures are from the model card; Plus adds outcome-based reinforcement learning to the separately released reasoning checkpoint." }, { "id": "alibaba/qwen3-0.6b", "slug": "alibaba--qwen3-0.6b", "name": "Qwen3-0.6B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen3", "description": null, "release_date": "2025-04-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.6, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen3/", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://qwenlm.github.io/blog/qwen3/", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/Qwen/Qwen3-0.6B/blob/c1899de289a04d12100db370d81485cdf75e47ca/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/Qwen/Qwen3-0.6B/blob/c1899de289a04d12100db370d81485cdf75e47ca/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Parameter count is the model-card figure. Native context is 32,768 tokens." }, { "id": "alibaba/qwen3-0.6b-base", "slug": "alibaba--qwen3-0.6b-base", "name": "Qwen3-0.6B-Base", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen3", "description": null, "release_date": "2025-04-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.6, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen3/", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://qwenlm.github.io/blog/qwen3/", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/Qwen/Qwen3-0.6B-Base/blob/da87bfb608c14b7cf20ba1ce41287e8de496c0cd/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/Qwen/Qwen3-0.6B-Base/blob/da87bfb608c14b7cf20ba1ce41287e8de496c0cd/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Pretrained checkpoint, distinct from the post-trained Qwen3 release. The April 29 announcement explicitly releases the pre-trained counterparts; model-card parameter and native-context figures are used." }, { "id": "alibaba/qwen3-1.7b", "slug": "alibaba--qwen3-1.7b", "name": "Qwen3-1.7B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen3", "description": null, "release_date": "2025-04-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 1.7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen3/", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://qwenlm.github.io/blog/qwen3/", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/Qwen/Qwen3-1.7B/blob/70d244cc86ccca08cf5af4e1e306ecf908b1ad5e/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/Qwen/Qwen3-1.7B/blob/70d244cc86ccca08cf5af4e1e306ecf908b1ad5e/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Parameter count is the model-card figure. Native context is 32,768 tokens." }, { "id": "alibaba/qwen3-1.7b-base", "slug": "alibaba--qwen3-1.7b-base", "name": "Qwen3-1.7B-Base", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen3", "description": null, "release_date": "2025-04-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 1.7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen3/", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://qwenlm.github.io/blog/qwen3/", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/Qwen/Qwen3-1.7B-Base/blob/ea980cb0a6c2ae4b936e82123acc929f1cec04c1/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/Qwen/Qwen3-1.7B-Base/blob/ea980cb0a6c2ae4b936e82123acc929f1cec04c1/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Pretrained checkpoint, distinct from the post-trained Qwen3 release. The April 29 announcement explicitly releases the pre-trained counterparts; model-card parameter and native-context figures are used." }, { "id": "alibaba/qwen3-14b", "slug": "alibaba--qwen3-14b", "name": "Qwen3-14B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen3", "description": null, "release_date": "2025-04-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 14.8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen3/", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://qwenlm.github.io/blog/qwen3/", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/Qwen/Qwen3-14B/blob/40c069824f4251a91eefaf281ebe4c544efd3e18/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/Qwen/Qwen3-14B/blob/40c069824f4251a91eefaf281ebe4c544efd3e18/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Parameter count is the model-card figure. Native context is 32,768 tokens; the developer validates 131,072 with YaRN scaling." }, { "id": "alibaba/qwen3-14b-base", "slug": "alibaba--qwen3-14b-base", "name": "Qwen3-14B-Base", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen3", "description": null, "release_date": "2025-04-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 14.8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen3/", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://qwenlm.github.io/blog/qwen3/", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/Qwen/Qwen3-14B-Base/blob/0b0bd3732e2c374d483664439ea334928b65f304/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/Qwen/Qwen3-14B-Base/blob/0b0bd3732e2c374d483664439ea334928b65f304/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Pretrained checkpoint, distinct from the post-trained Qwen3 release. The April 29 announcement explicitly releases the pre-trained counterparts; model-card parameter and native-context figures are used." }, { "id": "alibaba/qwen3-235b-a22b", "slug": "alibaba--qwen3-235b-a22b", "name": "Qwen3 235B-A22B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Large open Qwen MoE for multilingual reasoning, coding, and tool use", "release_date": "2025-04-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 235, "active_parameters_b": 22, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-235b-a22b.toml", "source_type": "catalog", "sources": [ { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-235b-a22b.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-235B-A22B", "type": "weights" }, { "label": "Qwen3 official release announcement", "url": "https://qwenlm.github.io/blog/qwen3/", "type": "primary" }, { "label": "Qwen3-235B-A22B configuration", "url": "https://huggingface.co/Qwen/Qwen3-235B-A22B/blob/main/config.json", "type": "primary" }, { "label": "Qwen3-235B-A22B license", "url": "https://huggingface.co/Qwen/Qwen3-235B-A22B/blob/main/LICENSE", "type": "primary" }, { "label": "Qwen3 official release history", "url": "https://github.com/QwenLM/Qwen3", "type": "primary" }, { "label": "Alibaba Qwen3 announcement", "url": "https://www.alibabagroup.com/en-US/document-1853940226976645120", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Original April 2025 hybrid thinking/non-thinking checkpoint, distinct from Instruct-2507 and Thinking-2507. 235B total and 22B active are nominal developer-reported counts. Context is 32,768 tokens natively and validated to 131,072 with YaRN factor 4; default config has max_position_embeddings=40,960 and rope_scaling=null. Alibaba Model Studio enables dynamic YaRN by default. Release date uses the explicit April 29 developer announcement; repository uploads on April 28 do not establish public availability.", "field_status": {}, "audited_at": "2026-10-26T14:14:14.099065Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/024.json" }, { "id": "alibaba/qwen3-30b-a3b", "slug": "alibaba--qwen3-30b-a3b", "name": "Qwen3 30B A3B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Qwen3 mixture-of-experts text model with 30.5B total and 3.3B active parameters, switchable thinking mode, and tool calling.", "release_date": "2025-04-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 30.5, "active_parameters_b": 3.3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-30b-a3b.toml", "source_type": "catalog", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-30B-A3B", "type": "weights" }, { "label": "Qwen3 launch announcement", "url": "https://qwenlm.github.io/blog/qwen3/", "type": "primary" }, { "label": "Official model configuration", "url": "https://huggingface.co/Qwen/Qwen3-30B-A3B/raw/main/config.json", "type": "primary" }, { "label": "Official Qwen3 release history", "url": "https://github.com/QwenLM/Qwen3", "type": "primary" }, { "label": "Alibaba launch announcement", "url": "https://www.alibabagroup.com/en-US/document-1853940226976645120", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-30b-a3b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Original April 2025 hybrid-thinking checkpoint. Context 131,072 requires YaRN; native context is 32,768. Release date follows official announcement; upstream catalog lists April 28.", "field_status": {}, "audited_at": "2026-10-26T14:20:34.581236Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/027.json" }, { "id": "alibaba/qwen3-30b-a3b-base", "slug": "alibaba--qwen3-30b-a3b-base", "name": "Qwen3-30B-A3B-Base", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen3", "description": null, "release_date": "2025-04-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 30.5, "active_parameters_b": 3.3, "parameters_source": "primary_source", "architecture": "moe", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen3/", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://qwenlm.github.io/blog/qwen3/", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/Qwen/Qwen3-30B-A3B-Base/blob/1b75feb79f60b8dc6c5bc769a898c206a1c6a4f9/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/Qwen/Qwen3-30B-A3B-Base/blob/1b75feb79f60b8dc6c5bc769a898c206a1c6a4f9/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Pretrained checkpoint, distinct from the post-trained Qwen3 release. The April 29 announcement explicitly releases the pre-trained counterparts; model-card parameter and native-context figures are used." }, { "id": "alibaba/qwen3-32b", "slug": "alibaba--qwen3-32b", "name": "Qwen3 32B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Dense open Qwen model for self-hosted chat, reasoning, and coding", "release_date": "2025-04-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 32.8, "active_parameters_b": 32, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/Qwen/Qwen3-32B", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-32B", "type": "weights" }, { "label": "Qwen3 launch announcement", "url": "https://qwenlm.github.io/blog/qwen3/", "type": "announcement" }, { "label": "Qwen3-32B configuration", "url": "https://huggingface.co/Qwen/Qwen3-32B/blob/main/config.json", "type": "documentation" }, { "label": "Qwen3 release history", "url": "https://github.com/QwenLM/Qwen3", "type": "documentation" }, { "label": "Qwen3-32B license", "url": "https://huggingface.co/Qwen/Qwen3-32B/blob/main/LICENSE", "type": "license" }, { "label": "Qwen3-32B upload history", "url": "https://huggingface.co/Qwen/Qwen3-32B/commits/main", "type": "documentation" }, { "label": "Qwen3 Technical Report", "url": "https://arxiv.org/html/2505.09388v1", "type": "paper" }, { "label": "Qwen3 naming convention", "url": "https://github.com/QwenLM/Qwen3/blob/main/docs/source/getting_started/concepts.md", "type": "documentation" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-32b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Developer reports 32.8B total and nominal 32B activated parameters (technical report, Tables 13–14); 31.2B is the separate non-embedding count. Context is 32,768 natively and 131,072 with YaRN; shipped config allows 40,960 positions without scaling. Release uses the explicit April 29 announcement; April 28 checkpoint uploads do not establish public availability.", "field_status": {}, "audited_at": "2026-10-27T07:43:20.563Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/029.json" }, { "id": "alibaba/qwen3-4b", "slug": "alibaba--qwen3-4b", "name": "Qwen3-4B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen3", "description": null, "release_date": "2025-04-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 4, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen3/", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://qwenlm.github.io/blog/qwen3/", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/Qwen/Qwen3-4B/blob/1cfa9a7208912126459214e8b04321603b3df60c/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/Qwen/Qwen3-4B/blob/1cfa9a7208912126459214e8b04321603b3df60c/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Parameter count is the model-card figure. Native context is 32,768 tokens; the developer validates 131,072 with YaRN scaling." }, { "id": "alibaba/qwen3-4b-base", "slug": "alibaba--qwen3-4b-base", "name": "Qwen3-4B-Base", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen3", "description": null, "release_date": "2025-04-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 4, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen3/", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://qwenlm.github.io/blog/qwen3/", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/Qwen/Qwen3-4B-Base/blob/906bfd4b4dc7f14ee4320094d8b41684abff8539/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/Qwen/Qwen3-4B-Base/blob/906bfd4b4dc7f14ee4320094d8b41684abff8539/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Pretrained checkpoint, distinct from the post-trained Qwen3 release. The April 29 announcement explicitly releases the pre-trained counterparts; model-card parameter and native-context figures are used." }, { "id": "alibaba/qwen3-8b", "slug": "alibaba--qwen3-8b", "name": "Qwen3 8B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Dense 8.2B Qwen3 model with switchable thinking for chat, reasoning, and tool use", "release_date": "2025-04-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 8.2, "active_parameters_b": 8, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/Qwen/Qwen3-8B", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen3-8B", "type": "weights" }, { "label": "Qwen3 release announcement", "url": "https://qwenlm.github.io/blog/qwen3/", "type": "primary" }, { "label": "Qwen3 Technical Report", "url": "https://arxiv.org/html/2505.09388v1", "type": "primary" }, { "label": "Qwen3-8B configuration", "url": "https://huggingface.co/Qwen/Qwen3-8B/raw/main/config.json", "type": "primary" }, { "label": "Qwen3 Technical Report: thinking comparison (Table 17)", "url": "https://arxiv.org/html/2505.09388v1#S4.T17", "type": "primary" }, { "label": "Qwen3 Technical Report: non-thinking comparison (Table 18)", "url": "https://arxiv.org/html/2505.09388v1#S4.T18", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen3-8b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "8.2B is the developer-rounded total; the technical report publishes nominal 8B activated parameters in both thinking and non-thinking mode. The 6.95B non-embedding count is separate. Native context is 32768 tokens; 131072 requires YaRN. The model card explains the default 40960 positional allocation.", "field_status": {}, "audited_at": "2026-10-27T07:50:39.418Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/030.json" }, { "id": "alibaba/qwen3-8b-base", "slug": "alibaba--qwen3-8b-base", "name": "Qwen3-8B-Base", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen3", "description": null, "release_date": "2025-04-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 8.2, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen3/", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://qwenlm.github.io/blog/qwen3/", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/Qwen/Qwen3-8B-Base/blob/49e3418fbbbca6ecbdf9608b4d22e5a407081db4/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/Qwen/Qwen3-8B-Base/blob/49e3418fbbbca6ecbdf9608b4d22e5a407081db4/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Pretrained checkpoint, distinct from the post-trained Qwen3 release. The April 29 announcement explicitly releases the pre-trained counterparts; model-card parameter and native-context figures are used." }, { "id": "meta/llama-guard-4-12b", "slug": "meta--llama-guard-4-12b", "name": "Llama Guard 4 12B", "developer": "Meta", "developer_id": "meta", "family": "llama", "description": "Dense multimodal safety classifier pruned from Llama 4 Scout, producing safety labels for text and image prompts and responses.", "release_date": "2025-04-29", "date_precision": "day", "open_weights": true, "license": "Llama 4 Community License", "parameters_b": 12, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/meta-llama/PurpleLlama/blob/main/Llama-Guard4/12B/MODEL_CARD.md", "source_type": "primary", "sources": [ { "label": "Meta Llama Guard 4 model card", "url": "https://github.com/meta-llama/PurpleLlama/blob/main/Llama-Guard4/12B/MODEL_CARD.md", "type": "primary" }, { "label": "Meta Llama Guard 4 prompt format", "url": "https://dev.meta.ai/llama/docs/model-cards-and-prompt-formats/llama-guard-4", "type": "primary" }, { "label": "Meta model registry: sequence length", "url": "https://github.com/meta-llama/llama-models/blob/0e0b8c519242d5833d8c11bffc1232b77ad7f301/models/sku_types.py", "type": "primary" }, { "label": "Meta release announcement", "url": "https://ai.meta.com/blog/ai-defenders-program-llama-protection-tools/", "type": "primary" }, { "label": "Llama 4 license", "url": "https://github.com/meta-llama/PurpleLlama/blob/main/Llama-Guard4/12B/LICENSE", "type": "primary" }, { "label": "Llama 4 acceptable-use policy", "url": "https://github.com/meta-llama/PurpleLlama/blob/main/Llama-Guard4/12B/USE_POLICY.md", "type": "primary" }, { "label": "Official checkpoint", "url": "https://huggingface.co/meta-llama/Llama-Guard-4-12B/tree/main", "type": "weights" }, { "label": "Hugging Face", "url": "https://huggingface.co/meta-llama/Llama-Guard-4-12B", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/meta/llama-guard-4-12b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Meta reports nominal 12B total parameters for the dense multimodal model pruned from Scout; active parameters are not separately reported. Scout's active count does not transfer after routed experts and routers are removed. Meta’s model registry sets maximum sequence length to 8192; the original HF configuration was inaccessible. The classifier evaluates tool-related content, but native tool calling and a dedicated reasoning mode are not disclosed. Original release is April 29, 2025; the custom license incorporates multimodal EU restrictions with an end-user exception.", "field_status": { "active_parameters_b": "undisclosed", "reasoning": "undisclosed", "tool_call": "undisclosed" }, "audited_at": "2026-10-27T14:18:30.624127Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/280.json" }, { "id": "cognitivecomputations/dolphin-mistral-24b-venice-edition", "slug": "cognitivecomputations--dolphin-mistral-24b-venice-edition", "name": "Dolphin Mistral 24B Venice Edition", "developer": "Dolphin (Cognitive Computations) and Venice", "developer_id": "cognitivecomputations", "family": "mistral", "description": "Instruction-tuned Mistral model for conversation and creative writing, with behavior guided by configurable system prompts.", "release_date": "2025-04-28", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 23.5724032, "active_parameters_b": 24, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://venice.ai/blog/introducing-dolphin-mistral-24b-venice-edition-the-most-uncensored-ai-model-yet", "source_type": "primary", "sources": [ { "label": "Original Venice launch", "url": "https://venice.ai/blog/introducing-dolphin-mistral-24b-venice-edition-the-most-uncensored-ai-model-yet", "type": "primary" }, { "label": "Original text model card", "url": "https://huggingface.co/dphn/Dolphin-Mistral-24B-Venice-Edition/blob/853f87026d587889137c814bb56bba60f033e747/README.md", "type": "primary" }, { "label": "Original text configuration", "url": "https://huggingface.co/dphn/Dolphin-Mistral-24B-Venice-Edition/blob/853f87026d587889137c814bb56bba60f033e747/config.json", "type": "primary" }, { "label": "Original weight index", "url": "https://huggingface.co/dphn/Dolphin-Mistral-24B-Venice-Edition/blob/853f87026d587889137c814bb56bba60f033e747/model.safetensors.index.json", "type": "primary" }, { "label": "Checkpoint revision history", "url": "https://huggingface.co/api/models/dphn/Dolphin-Mistral-24B-Venice-Edition/commits/main", "type": "primary" }, { "label": "Base model nominal active count", "url": "https://docs.mistral.ai/models/mistral-small-3-0-25-01", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/cognitivecomputations/dolphin-mistral-24b-venice-edition.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "availability": "public", "notes": "Original text-only release, available on Venice April 28, 2025; its public weights were uploaded June 12. The pinned pre-v1.2 checkpoint has 23.5724032B learned parameters and a 32768-token context. The nominal 24B active language count follows the named Mistral Small 2501 base convention. Version 1.1 was retrained in July 2025; version 1.2 later adds vision and longer context and is recorded separately. Extended-thinking and tool support for the original revision are unverified.", "field_status": {}, "audited_at": "2026-10-29T11:28:27.065Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/backfill/evidence/venice-revision-review.json", "active_parameters_scope": "language_model", "parameter_components_b": { "language_model": 23.5724032 } }, { "id": "writer/palmyra-x5", "slug": "writer--palmyra-x5", "name": "Palmyra X5", "developer": "Writer", "developer_id": "writer", "family": "palmyra", "description": "Writer mixture-of-experts model with hybrid attention, text and image input, a one-million-token context window and tool calling.", "release_date": "2025-04-28", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 456, "active_parameters_b": 50, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://writer.com/blog/long-context-palmyra-x5-press-release/", "source_type": "primary", "sources": [ { "label": "Palmyra X5 release announcement, April 28, 2025", "url": "https://writer.com/blog/long-context-palmyra-x5-press-release/", "type": "primary" }, { "label": "Introducing Palmyra X5", "url": "https://writer.com/blog/long-context-palmyra-x5/", "type": "primary" }, { "label": "Writer December 2025 transparency disclosure, items 21, 35 and 41", "url": "https://crfm.stanford.edu/fmti/December-2025/company-reports/Writer_FinalReport_FMTI2025.html", "type": "primary" }, { "label": "Writer model transparency and regulation statement", "url": "https://writer.com/legal/global-ai-regulation/", "type": "primary" }, { "label": "Current Palmyra model comparison and X5 details", "url": "https://dev.writer.com/home/models.md", "type": "primary" }, { "label": "Add images to chat completions", "url": "https://dev.writer.com/home/chat-with-images.md", "type": "primary" }, { "label": "Introducing Action Agent", "url": "https://writer.com/engineering/writer-action-agent/", "type": "primary" }, { "label": "Platform services agreement, sections 1(k), 2.1 and 2.3", "url": "https://writer.com/legal/platform-services/", "type": "primary" }, { "label": "Bedrock Palmyra X5 API limits", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-palmyra-x5.html", "type": "provider" }, { "label": "Native chat completion API", "url": "https://dev.writer.com/api-reference/completion-api/chat-completion.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/writer/palmyra-x5.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Writer reports 456B total parameters. Its X5 transparency submission uses 50B active parameters in final-training compute accounting; no separate serving-phase active count is given. Counts are author disclosures, not checkpoint measurements. Native Writer documents a one-million-token context and image input; Bedrock lists 1,040,000 input tokens and text-only access. Deep-thinking mode is documented in Action Agent, while API controls vary. Released April 28, 2025; current native documentation schedules deprecation for December 14, 2026.", "active_parameters_scope": "training_compute_estimate", "active_parameters_by_phase_b": { "training_compute_estimate": 50 }, "field_status": {}, "audited_at": "2026-10-27T05:22:24.303966Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/524.json" }, { "id": "openai/o4-mini-deep-research", "slug": "openai--o4-mini-deep-research", "name": "o4-mini-deep-research", "developer": "OpenAI", "developer_id": "openai", "family": "o4", "description": "OpenAI o4-mini variant for multi-step web research and data analysis, with text and image input and cited text reports.", "release_date": "2025-04-24", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 200000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-deep-research/", "source_type": "primary", "sources": [ { "label": "o4-mini-deep-research model reference", "url": "https://developers.openai.com/api/docs/models/o4-mini-deep-research", "type": "primary" }, { "label": "Introducing deep research: April 24, 2025 update", "url": "https://openai.com/index/introducing-deep-research/", "type": "primary" }, { "label": "Deep research in the API, webhooks, and web search with o3", "url": "https://community.openai.com/t/deep-research-in-the-api-webhooks-and-web-search-with-o3/1299919/1", "type": "primary" }, { "label": "OpenAI API changelog: June 2025", "url": "https://developers.openai.com/api/docs/changelog", "type": "primary" }, { "label": "Deep research API guide", "url": "https://developers.openai.com/api/docs/guides/deep-research", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201", "type": "primary" }, { "label": "File inputs API guide", "url": "https://developers.openai.com/api/docs/guides/file-inputs", "type": "primary" }, { "label": "OpenAI API deprecations", "url": "https://developers.openai.com/api/docs/deprecations", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "o4-mini model reference", "url": "https://developers.openai.com/api/docs/models/o4-mini", "type": "primary" }, { "label": "DeepSearchQA: Bridging the Comprehensiveness Gap for Deep Research Agents", "url": "https://storage.googleapis.com/deepmind-media/DeepSearchQA/DeepSearchQA_benchmark_paper.pdf", "type": "primary" }, { "label": "Tongyi DeepResearch Technical Report, November 15, 2025", "url": "https://raw.githubusercontent.com/Alibaba-NLP/DeepResearch/main/Tech_Report.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/o4-mini-deep-research.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "First publicly released as the lightweight o4-mini research variant in ChatGPT on April 24, 2025. OpenAI's later API announcement identifies o4-mini-deep-research as that ChatGPT research offering; identical April and June checkpoints are not assumed. The API changelog gives June 24 while the contemporaneous API announcement is dated June 26. The context and tool/input specifications are last documented for o4-mini-deep-research-2025-06-26; OpenAI shut down the alias and snapshot on July 23, 2026. Tool support means web search, file search, code interpreter and specialized MCP search/fetch; custom function calling is unsupported. PDF is an API input format converted to text and page images, and tool-generated figures are not native image output. Transformer is a broad inference from exact-model caching support and the applicable pre-GPT-5.6 engineering explanation. Dense/MoE structure remains undisclosed. No supported nominal total or active parameter count was found for this variant or its named o4-mini base; later research comparisons report performance only.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T17:25:52.214524Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/458.json" }, { "id": "openai/o3", "slug": "openai--o3", "name": "o3", "developer": "OpenAI", "developer_id": "openai", "family": "o3", "description": "OpenAI o-series reasoning model with text and image input, text output, and function calling.", "release_date": "2025-04-16", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 200000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-o3-and-o4-mini/", "source_type": "primary", "sources": [ { "label": "Introducing OpenAI o3 and o4-mini", "url": "https://openai.com/index/introducing-o3-and-o4-mini/", "type": "primary" }, { "label": "o3 model reference", "url": "https://developers.openai.com/api/docs/models/o3", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "Deprecations", "url": "https://developers.openai.com/api/docs/deprecations", "type": "primary" }, { "label": "OpenAI o3 and o4-mini System Card", "url": "https://cdn.openai.com/pdf/2221c875-02dc-4789-800b-e7758f3722c1/o3-and-o4-mini-system-card.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/o3.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The o3 alias is documented as pointing to o3-2025-04-16. It is deprecated, with API removal scheduled for December 11, 2026; endpoint operation was not tested. Transformer is a broad inference from exact o3 prompt-caching support and OpenAI's applicable pre-GPT-5.6 KV-cache explanation, not a disclosed topology. Nominal total and active parameter counts remain undisclosed; the inspected sources identify no counted base for a size inference. Dense versus MoE is also undisclosed. PDF is an API input format converted to text and page images. Image generation is a hosted tool, not native o3 image output.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T17:19:40.709384Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/453.json" }, { "id": "openai/o4-mini", "slug": "openai--o4-mini", "name": "o4-mini", "developer": "OpenAI", "developer_id": "openai", "family": "o4", "description": "OpenAI o-series reasoning model with text and image input, text output, and function calling.", "release_date": "2025-04-16", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 200000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-o3-and-o4-mini/", "source_type": "primary", "sources": [ { "label": "Introducing OpenAI o3 and o4-mini", "url": "https://openai.com/index/introducing-o3-and-o4-mini/", "type": "primary" }, { "label": "o4-mini model reference", "url": "https://developers.openai.com/api/docs/models/o4-mini", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs", "type": "primary" }, { "label": "OpenAI API deprecations", "url": "https://developers.openai.com/api/docs/deprecations", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "Model Release Notes: o4-mini update, June 6, 2025", "url": "https://help.openai.com/en/articles/9624314-model-release-notes", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/o4-mini.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The o4-mini alias is documented as pointing to o4-mini-2025-04-16; OpenAI lists both as shut down on October 23, 2026. Context and capabilities are last documented API specifications, not a live endpoint test. The June 6, 2025 rollback notice does not identify an API snapshot, so it is not treated as a change to the dated API model. Transformer is a broad inference from exact prompt-caching support and OpenAI's applicable pre-GPT-5.6 engineering explanation; dense versus MoE remains undisclosed. Total and active parameter counts remain undisclosed after a fresh source review; the published approximately 100B estimate is not a developer specification. PDF is an API input format converted to text and page images. Image generation is a hosted tool, not native o4-mini output. The 256K SWE-bench evaluation setting does not replace the documented 200K API context.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T17:25:52.584970Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/457.json" }, { "id": "openai/gpt-4.1", "slug": "openai--gpt-4.1", "name": "GPT-4.1", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-4.1", "description": "Model for text and image understanding, coding, instruction following and long-context tasks.", "release_date": "2025-04-14", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 1047576, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://openai.com/index/gpt-4-1/", "source_type": "primary", "sources": [ { "label": "Introducing GPT-4.1 in the API", "url": "https://openai.com/index/gpt-4-1/", "type": "primary" }, { "label": "GPT-4.1 model", "url": "https://developers.openai.com/api/docs/models/gpt-4.1", "type": "primary" }, { "label": "GPT-4.1 Prompting Guide", "url": "https://developers.openai.com/cookbook/examples/gpt4-1_prompting_guide", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs", "type": "primary" }, { "label": "API deprecations", "url": "https://developers.openai.com/api/docs/deprecations", "type": "primary" }, { "label": "Introducing gpt-oss", "url": "https://openai.com/index/introducing-gpt-oss/", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "Training agents to self-report misbehavior", "url": "https://alignment.openai.com/self-incrimination/", "type": "primary" }, { "label": "Training Agents to Self-Report Misbehavior", "url": "https://arxiv.org/abs/2602.22303", "type": "primary" }, { "label": "OpenAI GPT-5 System Card", "url": "https://arxiv.org/abs/2601.03267", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201", "type": "primary" }, { "label": "Kimi K2: Open Agentic Intelligence", "url": "https://arxiv.org/html/2507.20534v1", "type": "primary" }, { "label": "Identifying Fine-grained Forms of Populism in Political Discourse: A Case Study on Donald Trump’s Presidential Campaigns", "url": "https://aclanthology.org/2026.eacl-long.40.pdf", "type": "primary" }, { "label": "Assessing French Readability for Adults with Low Literacy: A Global and Local Perspective", "url": "https://aclanthology.org/2025.emnlp-main.1036.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-4.1.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The gpt-4.1 alias defaults to gpt-4.1-2025-04-14, released in the API on April 14, 2025. Exact documented context is 1,047,576 tokens. PDF input is processed into text and page images. GPT-4.1 has no dedicated internal reasoning mode, but supports prompted visible planning and function calling. Transformer attention is documented; dense/MoE routing and nominal total/active parameter counts remain undisclosed after the October 27, 2026 source recheck. Deprecated October 14, 2026; API shutdown is scheduled for January 14, 2027, not already completed at review time.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T16:24:01.471000Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/405.json" }, { "id": "openai/gpt-4.1-mini", "slug": "openai--gpt-4.1-mini", "name": "GPT-4.1 mini", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-4.1", "description": "GPT-4.1 mini accepts text and images and generates text, with a 1,047,576-token context window and function calling.", "release_date": "2025-04-14", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1047576, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://openai.com/index/gpt-4-1/", "source_type": "primary", "sources": [ { "label": "GPT-4.1 Mini model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-4.1-mini.md", "type": "primary" }, { "label": "Introducing GPT-4.1 in the API", "url": "https://openai.com/index/gpt-4-1/", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "Introducing gpt-oss", "url": "https://openai.com/index/introducing-gpt-oss/", "type": "primary" }, { "label": "OpenAI deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "Training Agents to Self-Report Misbehavior", "url": "https://arxiv.org/html/2602.22303v1", "type": "primary" }, { "label": "GPT-5 System Card", "url": "https://cdn.openai.com/gpt-5-system-card.pdf", "type": "primary" }, { "label": "Incompressible Knowledge Probes: Estimating Black-Box LLM Parameter Counts via Factual Capacity (v2)", "url": "https://arxiv.org/html/2604.24827v2", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-4.1-mini.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The unpinned gpt-4.1-mini name documents gpt-4.1-mini-2025-04-14 as its default snapshot, first released April 14, 2025. The October 27, 2026 review records announced API removal on January 14, 2027; no live endpoint request was made. PDF inputs become text and page images. Total and active parameter counts remain undisclosed; effective-capacity estimates and unattributed comparison ranges do not establish nominal counts. Exact-mini architecture remains unverified.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "unverified" }, "audited_at": "2026-10-27T16:24:58.276913Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/406.json" }, { "id": "openai/gpt-4.1-nano", "slug": "openai--gpt-4.1-nano", "name": "GPT-4.1 nano", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-4.1", "description": "Non-reasoning GPT-4.1 variant with text and image input, text output, and function calling.", "release_date": "2025-04-14", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 1047576, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://openai.com/index/gpt-4-1/", "source_type": "primary", "sources": [ { "label": "Introducing GPT-4.1 in the API", "url": "https://openai.com/index/gpt-4-1/", "type": "primary" }, { "label": "GPT-4.1 nano model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-4.1-nano.md", "type": "primary" }, { "label": "API deprecations: legacy GPT model snapshots", "url": "https://developers.openai.com/api/docs/deprecations", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "Introducing gpt-oss", "url": "https://openai.com/index/introducing-gpt-oss/", "type": "primary" }, { "label": "GPT-5 System Card, August 13, 2025 revision", "url": "https://cdn.openai.com/gpt-5-system-card.pdf", "type": "primary" }, { "label": "Supervised fine-tuning", "url": "https://developers.openai.com/api/docs/guides/supervised-fine-tuning", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201", "type": "primary" }, { "label": "Fine-grained Confidence Estimation for Spurious Correctness Detection in Large Language Models", "url": "https://aclanthology.org/2025.ijcnlp-long.68.pdf", "type": "primary" }, { "label": "Directed Attention is All You Need: Profiling Style from Limited Text Data", "url": "https://aclanthology.org/2026.sigturk-1.2.pdf", "type": "primary" }, { "label": "Incompressible Knowledge Probes: Estimating Black-Box LLM Parameter Counts via Factual Capacity, v2", "url": "https://arxiv.org/pdf/2604.24827", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-4.1-nano.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Release date is the April 14, 2025 API launch. The last documented default snapshot is gpt-4.1-nano-2025-04-14; OpenAI lists both it and the alias as shut down on October 23, 2026. Specifications describe the historical model. PDF denotes API file input processed as extracted text and page images. Explicit nano prompt-caching support and OpenAI's documented pre-GPT-5.6 caching mechanism support a broad Transformer-attention inference. Complete structure and dense/MoE routing remain undisclosed. No supported total or active parameter count was found; an unsourced 7B research-table entry and effective-capacity estimates do not establish model specifications.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T16:24:52.965577Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/407.json" }, { "id": "amazon/nova-sonic", "slug": "amazon--nova-sonic", "name": "Nova Sonic", "developer": "Amazon", "developer_id": "amazon", "family": "nova-sonic", "description": "Conversational speech model with text and audio responses and tool use.", "release_date": "2025-04-08", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 300000, "modalities": { "input": [ "text", "audio" ], "output": [ "text", "audio" ] }, "reasoning": null, "tool_call": true, "source_url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-sonic-human-like-voice-conversations-for-generative-ai-applications/", "source_type": "primary", "sources": [ { "label": "Dated public launch", "url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-sonic-human-like-voice-conversations-for-generative-ai-applications/", "type": "primary" }, { "label": "Publisher service card and architecture", "url": "https://docs.aws.amazon.com/ai/responsible-ai/nova-sonic/overview.html", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:13:05.399Z", "last_seen": "2026-10-30T07:13:05.399Z", "upstream_present": false, "availability": "public", "notes": "Conversational speech model with a core transformer language model, speech encoder and renderer. The publisher does not disclose total or active parameters or dense/MoE subtype. Available through Bedrock; no downloadable exactcheckpoint was verified. Toolcalling is explicitly documented; a separate extended-thinking mode is unverified. Context300000 follows the corrected launch article. Text input covers system instructions/toolresults; arbitrary user text interaction is not asserted." }, { "id": "mistral/pixtral-large-2502", "slug": "mistral--pixtral-large-2502", "name": "Pixtral Large (25.02)", "developer": "Mistral", "developer_id": "mistral", "family": "pixtral", "description": "Mistral vision-language model for image understanding and multimodal chat", "release_date": "2025-04-08", "date_precision": "day", "open_weights": false, "license": "Mistral Research License", "parameters_b": 124, "active_parameters_b": 123, "parameters_source": "primary_source", "architecture": null, "context_window": 128000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://aws.amazon.com/blogs/aws/aws-announces-pixtral-large-25-02-model-in-amazon-bedrock-serverless/", "source_type": "primary", "sources": [ { "label": "Amazon Bedrock now offers Pixtral Large 25.02", "url": "https://aws.amazon.com/about-aws/whats-new/2025/04/amazon-bedrock-pixtral-large-25-02-mistral-ai/", "type": "primary" }, { "label": "Pixtral Large (25.02) parameters and inference", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-mistral-pixtral-large.html", "type": "primary" }, { "label": "Pixtral Large is now available in Amazon Bedrock", "url": "https://aws.amazon.com/blogs/machine-learning/pixtral-large-is-now-available-in-amazon-bedrock/", "type": "primary" }, { "label": "Pixtral Large model card", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-mistral-ai-pixtral-large.html", "type": "primary" }, { "label": "Pinned Pixtral Large 24.11 model definition", "url": "https://raw.githubusercontent.com/mistralai/platform-docs-public/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/schema/models/models/pixtral-large-24-11.ts", "type": "primary" }, { "label": "Pinned model weights table renderer", "url": "https://raw.githubusercontent.com/mistralai/platform-docs-public/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/app/%5Blocale%5D/(docs)/models/model-cards/%5Bslug%5D/components/weights-table.tsx", "type": "primary" }, { "label": "Pixtral-Large-Instruct-2411 model card and weights", "url": "https://huggingface.co/mistralai/Pixtral-Large-Instruct-2411", "type": "weights" }, { "label": "Official Mistral Hugging Face Pixtral repository inventory", "url": "https://huggingface.co/api/models?author=mistralai&search=Pixtral&limit=100", "type": "weights" }, { "label": "Mistral official changelog", "url": "https://docs.mistral.ai/resources/changelogs", "type": "primary" }, { "label": "Pixtral Large legal lifecycle", "url": "https://legal.mistral.ai/ai-governance/models/pixtral-large/", "type": "primary" }, { "label": "Initial Models.dev Pixtral Large 2502 base record", "url": "https://github.com/anomalyco/models.dev/blob/6d0cc3506e5e70085fe507f7ff31339a7ecf4028/models/mistral/pixtral-large-2502.toml", "type": "catalog" }, { "label": "Initial Models.dev Bedrock Pixtral Large 2502 provider record", "url": "https://github.com/anomalyco/models.dev/blob/f24ee000d784586f6ac010b3fd7321b0f561c06b/providers/amazon-bedrock/models/mistral.pixtral-large-2502-v1%3A0.toml", "type": "catalog" }, { "label": "Pinned official model schema index", "url": "https://raw.githubusercontent.com/mistralai/platform-docs-public/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/schema/models/models/index.ts", "type": "primary" }, { "label": "Current Pixtral Large model card", "url": "https://docs.mistral.ai/models/pixtral-large-24-11", "type": "primary" }, { "label": "Pixtral Large original release", "url": "https://mistral.ai/news/pixtral-large/", "type": "primary" }, { "label": "Kimi-VL Technical Report, Table 3", "url": "https://arxiv.org/html/2504.07491v1", "type": "primary" }, { "label": "Qwen2.5-VL Technical Report", "url": "https://arxiv.org/html/2502.13923v1", "type": "primary" }, { "label": "InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models", "url": "https://arxiv.org/html/2504.10479v1", "type": "primary" }, { "label": "Qwen3-VL Technical Report", "url": "https://arxiv.org/html/2511.21631v2", "type": "primary" }, { "label": "Pinned Mistral Large 2.0 24.07 model definition", "url": "https://raw.githubusercontent.com/mistralai/platform-docs-public/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/schema/models/models/mistral-large-2-0-24-07.ts", "type": "primary" }, { "label": "Large Enough: original Mistral Large 2 release", "url": "https://mistral.ai/news/mistral-large-2407/", "type": "primary" }, { "label": "Original Amazon and Mistral availability announcement", "url": "https://www.aboutamazon.com/news/aws/aws-mistral-ai-pixtral-large", "type": "primary" }, { "label": "Native Amazon Bedrock release history", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/bedrock-ug-doc-history.html", "type": "primary" }, { "label": "AWS announces Pixtral Large 25.02 model in Amazon Bedrock serverless", "url": "https://aws.amazon.com/blogs/aws/aws-announces-pixtral-large-25-02-model-in-amazon-bedrock-serverless/", "type": "primary" }, { "label": "Exact Pixtral Large 2502 repository search", "url": "https://huggingface.co/api/models?search=pixtral-large-2502&limit=100", "type": "primary" }, { "label": "Current Mistral model index", "url": "https://docs.mistral.ai/models", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/pixtral-large-2502.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "AWS documents a nominal 124B multimodal total (123B decoder plus 1B vision encoder). The 123B active count covers only the language decoder and is inferred from AWS’s explicit Mistral Large 2 lineage and Mistral’s published 123B active convention; whole-model active parameters remain unspecified. Provider context is 128,000 tokens with Mistral Research License. The exact 25.02 API revision launched on April 8, 2025. No public downloadable 2502 weights were found, so weight access is closed; the 2411 download is not assumed equivalent. Architecture and dedicated reasoning-mode classification remain unresolved.", "field_status": { "architecture": "unverified", "reasoning": "unverified" }, "audited_at": "2026-10-28T07:55:54.893674Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/350.json", "active_parameters_scope": "language_decoder", "parameter_components_b": { "language_decoder": 123, "vision_encoder": 1 } }, { "id": "nvidia/llama-3.1-nemotron-ultra-253b", "slug": "nvidia--llama-3.1-nemotron-ultra-253b", "name": "Llama 3.1 Nemotron Ultra 253B v1", "developer": "NVIDIA", "developer_id": "nvidia", "family": "nemotron", "description": "NVIDIA reasoning and chat model derived from Llama 3.1 405B Instruct through neural architecture search and FFN fusion, with a reasoning toggle and tool-calling support.", "release_date": "2025-04-07", "date_precision": "day", "open_weights": true, "license": "NVIDIA Open Model License + Llama 3.1 Community License", "parameters_b": 253.401268224, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/nvidia/Llama-3_1-Nemotron-Ultra-253B-v1/blob/5b47def5b8957142d82faeb44bfad9349d02840d/README.md", "source_type": "primary", "sources": [ { "label": "Llama-3.1-Nemotron-Ultra-253B-v1 model card", "url": "https://huggingface.co/nvidia/Llama-3_1-Nemotron-Ultra-253B-v1/blob/5b47def5b8957142d82faeb44bfad9349d02840d/README.md", "type": "primary" }, { "label": "Ultra 253B v1 checkpoint configuration", "url": "https://huggingface.co/nvidia/Llama-3_1-Nemotron-Ultra-253B-v1/blob/5b47def5b8957142d82faeb44bfad9349d02840d/config.json", "type": "primary" }, { "label": "Ultra 253B v1 model implementation", "url": "https://huggingface.co/nvidia/Llama-3_1-Nemotron-Ultra-253B-v1/blob/5b47def5b8957142d82faeb44bfad9349d02840d/modeling_decilm.py", "type": "primary" }, { "label": "Ultra 253B v1 BF16 checkpoint index and referenced shard headers", "url": "https://huggingface.co/nvidia/Llama-3_1-Nemotron-Ultra-253B-v1/blob/5b47def5b8957142d82faeb44bfad9349d02840d/model.safetensors.index.json", "type": "primary" }, { "label": "Llama-Nemotron: Efficient Reasoning Models, revision 5", "url": "https://arxiv.org/html/2505.00949v5", "type": "primary" }, { "label": "NVIDIA Open Models License", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/", "type": "primary" }, { "label": "Build Enterprise AI Agents with Advanced Open NVIDIA Llama Nemotron Reasoning Models", "url": "https://developer.nvidia.com/blog/build-enterprise-ai-agents-with-advanced-open-nvidia-llama-nemotron-reasoning-models/", "type": "primary" }, { "label": "Quantization-Aware Distillation for NVFP4 Inference Accuracy Recovery", "url": "https://research.nvidia.com/labs/nemotron/files/NVFP4-QAD-Report.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/nvidia/llama-3.1-nemotron-ultra-253b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Ultra v1 has exactly 253,401,268,224 learned checkpoint parameters; NVIDIA publishes the nominal size as 253B. Its dense architecture uses NAS and FFN fusion, so the Llama 3.1 405B base count does not transfer. No reliable separate active count was found. The card names NVIDIA Open Model License and Llama 3.1 Community License. This entry covers Ultra v1, not CPT, FP8 storage, or Nemotron 3 Ultra.", "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T15:54:48.843293Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/376.json" }, { "id": "meta/llama-4-maverick-17b-instruct", "slug": "meta--llama-4-maverick-17b-instruct", "name": "Llama 4 Maverick 17B Instruct", "developer": "Meta", "developer_id": "meta", "family": "llama", "description": "Meta’s instruction-tuned multimodal MoE model with nominal 400B total and 17B active parameters, 128 routed experts, text and image input, and native tool calling.", "release_date": "2025-04-05", "date_precision": "day", "open_weights": true, "license": "Llama 4 Community License", "parameters_b": 400, "active_parameters_b": 17, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://ai.meta.com/blog/llama-4-multimodal-intelligence/", "source_type": "primary", "sources": [ { "label": "Official Llama 4 model card", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama4/MODEL_CARD.md", "type": "primary" }, { "label": "Official exact Maverick Instruct context limit", "url": "https://github.com/meta-llama/llama-models/blob/0e0b8c519242d5833d8c11bffc1232b77ad7f301/models/sku_types.py", "type": "primary" }, { "label": "Official Llama 4 prompt and tool formats", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama4/prompt_format.md", "type": "primary" }, { "label": "Llama 4 launch and architecture", "url": "https://ai.meta.com/blog/llama-4-multimodal-intelligence/", "type": "announcement" }, { "label": "Hugging Face", "url": "https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/meta/llama-4-maverick-17b-instruct.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Meta reports nominal 400B total and 17B active parameters. Maverick alternates dense and MoE layers; each MoE layer uses 128 routed experts plus a shared expert and activates one routed expert per token. The standard BF16 Instruct checkpoint is distinct from its FP8 packaging and the experimental arena-chat model. Exact Instruct context is 1,048,576 tokens. Dedicated thinking-mode support is not explicitly disclosed in the reviewed native interface documentation.", "field_status": { "reasoning": "undisclosed" }, "audited_at": "2026-10-26T21:43:37.073822Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/277.json" }, { "id": "meta/llama-4-scout-17b-instruct", "slug": "meta--llama-4-scout-17b-instruct", "name": "Llama 4 Scout 17B 16E Instruct", "developer": "Meta", "developer_id": "meta", "family": "llama", "description": "Instruction-tuned multimodal MoE with 109B total and 17B active parameters, text/image input, text output, tool calling, and a 10,485,760-token context.", "release_date": "2025-04-05", "date_precision": "day", "open_weights": true, "license": "Llama 4 Community License", "parameters_b": 109, "active_parameters_b": 17, "parameters_source": "primary_source", "architecture": "moe", "context_window": 10485760, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://ai.meta.com/blog/llama-4-multimodal-intelligence/", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E-Instruct", "type": "weights" }, { "label": "Meta exact model sequence limits", "url": "https://github.com/meta-llama/llama-models/blob/main/models/sku_types.py", "type": "primary" }, { "label": "Meta Llama 4 prompt and tool formats", "url": "https://dev.meta.ai/llama/docs/model-cards-and-prompt-formats/llama4", "type": "primary" }, { "label": "Meta Llama 4 model comparison", "url": "https://dev.meta.ai/llama/docs/deployment/versioning", "type": "primary" }, { "label": "Original checkpoint public metadata", "url": "https://huggingface.co/api/models/meta-llama/Llama-4-Scout-17B-16E-Instruct", "type": "weights" }, { "label": "Llama 4 launch and architecture", "url": "https://ai.meta.com/blog/llama-4-multimodal-intelligence/", "type": "announcement" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/meta/llama-4-scout-17b-instruct.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Meta reports nominal 109B total and 17B active parameters. The exact Scout Instruct context is 10,485,760 tokens; pretrained Scout is a different 262,144-token variant. The original configuration is gated, so no precise complete tensor count or exact per-layer routing configuration is asserted. A dedicated reasoning mode is not disclosed in the reviewed primary documentation.", "field_status": { "reasoning": "undisclosed" }, "audited_at": "2026-10-26T21:44:13.844592Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/278.json" }, { "id": "deepseek/deepseek-v3-0324", "slug": "deepseek--deepseek-v3-0324", "name": "DeepSeek V3 0324", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek", "description": "March 2025 checkpoint of DeepSeek-V3 with improved reasoning and coding", "release_date": "2025-03-24", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 671, "active_parameters_b": 36.7, "parameters_source": "primary_source", "architecture": "moe", "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3-0324", "source_type": "primary", "sources": [ { "label": "Alias documentation", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3-0324", "type": "docs" }, { "label": "Exact checkpoint configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3-0324/raw/main/config.json", "type": "weights" }, { "label": "Official main-model and MTP weight guide", "url": "https://raw.githubusercontent.com/deepseek-ai/DeepSeek-V3/main/README_WEIGHTS.md", "type": "primary" }, { "label": "DeepSeek-V3 technical report", "url": "https://arxiv.org/html/2412.19437v2", "type": "primary" }, { "label": "March 24 release changelog", "url": "https://api-docs.deepseek.com/updates/#date-2025-03-24", "type": "primary" }, { "label": "Exact checkpoint MIT license", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3-0324/blob/main/LICENSE", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepseek/deepseek-v3-0324.toml", "type": "catalog" } ], "aliases": [ "deepseek/deepseek-chat-v3-0324" ], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Counts describe the main model: nominal 671B and approximately 36.7B active (37B rounded). The MTP module shares embedding and output weights; the 685B storage headline is not a unique learned total. The guide’s nominal 11.5B MTP figure also differs from tensor-shape accounting, so no complete unique total is asserted. Advertised context is 128K, represented as nominal 128000; configured positions are 163840. Release was March 24; the news announcement followed March 25.", "parameters_scope": "main", "active_parameters_scope": "main", "parameter_components_b": { "main": 671 }, "active_parameters_by_phase_b": { "main": 36.7 }, "field_status": {}, "audited_at": "2026-10-26T19:05:10.033568+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/174.json", "alias_sources": { "deepseek/deepseek-chat-v3-0324": { "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3-0324", "note": "Duplicate catalog names for the official DeepSeek-V3-0324 chat checkpoint, not separate base versus instruct releases." } } }, { "id": "nvidia/llama-3.3-nemotron-super-49b-v1", "slug": "nvidia--llama-3.3-nemotron-super-49b-v1", "name": "Llama 3.3 Nemotron Super 49B v1", "developer": "NVIDIA", "developer_id": "nvidia", "family": "nemotron", "description": "NVIDIA’s dense Llama 3.3 70B Instruct derivative, compressed through neural architecture search and post-trained for reasoning, chat and tool calling.", "release_date": "2025-03-18", "date_precision": "day", "open_weights": true, "license": "NVIDIA Open Model License + Llama 3.3 Community License", "parameters_b": 49.867145216, "active_parameters_b": 49, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/nvidia/llama-3.3-nemotron-super-49b-v1.toml", "source_type": "catalog", "sources": [ { "label": "Exact NVIDIA Super v1 model card", "url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1/blob/387156d8d6868c19f3472fa607aa9bfc4f662333/README.md", "type": "primary" }, { "label": "Pinned Super v1 configuration", "url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1/resolve/387156d8d6868c19f3472fa607aa9bfc4f662333/config.json", "type": "weights" }, { "label": "Official custom architecture implementation", "url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1/resolve/387156d8d6868c19f3472fa607aa9bfc4f662333/modeling_decilm.py", "type": "weights" }, { "label": "Official complete checkpoint tensor index", "url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1/resolve/387156d8d6868c19f3472fa607aa9bfc4f662333/model.safetensors.index.json", "type": "weights" }, { "label": "Official safetensors headers (first of 21 shards)", "url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1/resolve/387156d8d6868c19f3472fa607aa9bfc4f662333/model-00001-of-00021.safetensors", "type": "weights" }, { "label": "Llama-Nemotron technical report, September 2025 revision", "url": "https://arxiv.org/html/2505.00949v5", "type": "primary" }, { "label": "Contemporaneous NVIDIA launch announcement", "url": "https://forums.developer.nvidia.com/t/ai-reasoning-with-llama-nemotron-at-gtc25-announcements/327509", "type": "primary" }, { "label": "NVIDIA Open Model License Agreement", "url": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/", "type": "primary" }, { "label": "Llama 3.3 Community License Agreement", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_3/LICENSE", "type": "primary" }, { "label": "NVIDIA Llama Nemotron training and launch article", "url": "https://developer.nvidia.com/blog/build-enterprise-ai-agents-with-advanced-open-nvidia-llama-nemotron-reasoning-models/", "type": "primary" }, { "label": "Vultr original Nemotron Super v1.5 inference benchmark", "url": "https://docs.vultr.com/inference-cookbook/cuda/model-guides/nemotron-super-49b", "type": "primary" }, { "label": "NVIDIA Super v1.5 card: explicit v1 upgrade and post-training", "url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1_5/blob/420ba7d28211abf116b8b103ab700d92619daf98/README.md", "type": "primary" }, { "label": "Pinned NVIDIA Super v1.5 configuration compared with v1", "url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1_5/resolve/420ba7d28211abf116b8b103ab700d92619daf98/config.json", "type": "weights" }, { "label": "Pinned NVIDIA Super v1.5 implementation identical to v1", "url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1_5/resolve/420ba7d28211abf116b8b103ab700d92619daf98/modeling_decilm.py", "type": "weights" }, { "label": "NVIDIA Super v1.5 FP8 card and named lineage", "url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1_5-FP8/blob/04822723e77e036ddf2d24e83c6d469d3b009252/README.md", "type": "primary" }, { "label": "Pinned NVIDIA Super v1.5 FP8 configuration", "url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1_5-FP8/resolve/04822723e77e036ddf2d24e83c6d469d3b009252/config.json", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/nvidia/llama-3.3-nemotron-super-49b-v1.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Exact checkpoint total is 49.867145216B. Active 49B is a nominal inference from Vultr's v1.5 benchmark convention: NVIDIA identifies v1.5 as a v1 upgrade, and official configurations and implementation preserve the learned architecture. It is not a separate NVIDIA v1 active-count disclosure or exact per-token tensor census. Super v1 has 49 attention blocks and 80 variable-width FFNs; the technical report's FFN Fusion applies to Ultra.", "field_status": {}, "audited_at": "2026-10-27T15:54:28.219417+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/377.json" }, { "id": "mistral/mistral-small-3.1-24b-instruct", "slug": "mistral--mistral-small-3.1-24b-instruct", "name": "Mistral Small 3.1 24B", "developer": "Mistral", "developer_id": "mistral", "family": "mistral-small", "description": "Instruction-tuned 24B dense multimodal model for text generation, image understanding, and native function calling.", "release_date": "2025-03-17", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 24, "active_parameters_b": 24, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://mistral.ai/news/mistral-small-3-1/", "source_type": "primary", "sources": [ { "label": "Mistral Small 3.1 release announcement", "url": "https://mistral.ai/news/mistral-small-3-1/", "type": "primary" }, { "label": "Mistral-Small-3.1-24B-Instruct-2503 model card", "url": "https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503/raw/68faf511d618ef198fef186659617cfd2eb8e33a/README.md", "type": "weights" }, { "label": "Official Small 3.1 model schema and weights rows", "url": "https://raw.githubusercontent.com/mistralai/platform-docs-public/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/schema/models/models/mistral-small-3-1-25-03.ts", "type": "primary" }, { "label": "Official weights table renderer and units", "url": "https://raw.githubusercontent.com/mistralai/platform-docs-public/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/app/%5Blocale%5D/(docs)/models/model-cards/%5Bslug%5D/components/weights-table.tsx", "type": "primary" }, { "label": "Exact Instruct checkpoint Transformers configuration", "url": "https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503/raw/68faf511d618ef198fef186659617cfd2eb8e33a/config.json", "type": "weights" }, { "label": "Exact Instruct checkpoint native Mistral parameters", "url": "https://huggingface.co/mistralai/Mistral-Small-3.1-24B-Instruct-2503/raw/68faf511d618ef198fef186659617cfd2eb8e33a/params.json", "type": "weights" }, { "label": "Mistral native Transformer argument definitions", "url": "https://raw.githubusercontent.com/mistralai/mistral-inference/9eaeb91c17450e09021b6065a1d5cc69876507c8/src/mistral_inference/args.py", "type": "primary" }, { "label": "Mistral native Transformer layer implementation", "url": "https://raw.githubusercontent.com/mistralai/mistral-inference/9eaeb91c17450e09021b6065a1d5cc69876507c8/src/mistral_inference/transformer_layers.py", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/mistral-small-3.1-24b-instruct.toml", "type": "catalog" }, { "label": "Alias documentation", "url": "https://docs.mistral.ai/models/mistral-small-3-1-25-03", "type": "docs" } ], "aliases": [ "mistral/mistral-small-3-1-24b-instruct-2503" ], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Exact released checkpoint is Mistral-Small-3.1-24B-Instruct-2503. Total 24B and active 24B are separately published nominal developer figures in the official weights table, not an exact tensor census or a modality-specific active count. Native and Transformers configuration both resolve the advertised 128k context to 131072 tokens. Dedicated reasoning-mode support remains unverified; generic reasoning benchmarks do not establish it. The hosted mistral-small-2503 API was retired on 2025-11-30; this record describes the released weights.", "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-26T23:34:30.596239Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/345.json", "alias_sources": { "mistral/mistral-small-3-1-24b-instruct-2503": { "url": "https://docs.mistral.ai/models/mistral-small-3-1-25-03", "note": "Alias or alternate catalog name for the same released model; only the retained ID counts as a release." } } }, { "id": "allenai/olmo-2-0325-32b", "slug": "allenai--olmo-2-0325-32b", "name": "OLMo-2-0325-32B", "developer": "Ai2", "developer_id": "allenai", "family": "olmo", "description": null, "release_date": "2025-03-13", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 32, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://allenai.org/blog/olmo2-32B", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://allenai.org/blog/olmo2-32B", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/allenai/OLMo-2-0325-32B/blob/cc9d3cf9c7230b86ee6b84607b37db1c01e3f1ed/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/allenai/OLMo-2-0325-32B/blob/cc9d3cf9c7230b86ee6b84607b37db1c01e3f1ed/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Final pretrained 32B release. Nominal model size; the published configuration supports 4,096 tokens." }, { "id": "allenai/olmo-2-0325-32b-instruct", "slug": "allenai--olmo-2-0325-32b-instruct", "name": "OLMo-2-0325-32B-Instruct", "developer": "Ai2", "developer_id": "allenai", "family": "olmo", "description": null, "release_date": "2025-03-13", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 32, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://allenai.org/blog/olmo2-32B", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://allenai.org/blog/olmo2-32B", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/allenai/OLMo-2-0325-32B-Instruct/blob/b96024342a77a69aa0dda815c3454a671f477463/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/allenai/OLMo-2-0325-32B-Instruct/blob/b96024342a77a69aa0dda815c3454a671f477463/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Final instruction-tuned 32B release. Nominal model size; the published configuration supports 4,096 tokens." }, { "id": "cohere/command-a-03-2025", "slug": "cohere--command-a-03-2025", "name": "Command A", "developer": "Cohere", "developer_id": "cohere", "family": "command-a", "description": "Cohere text model for multilingual chat, retrieval-augmented generation, and tool-using agents.", "release_date": "2025-03-13", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-4.0 with Acceptable Use Addendum", "parameters_b": 111, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 256000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://docs.cohere.com/changelog/command-a", "source_type": "primary", "sources": [ { "label": "Cohere Command A release", "url": "https://docs.cohere.com/changelog/command-a", "type": "primary" }, { "label": "Cohere Command A model card", "url": "https://huggingface.co/CohereLabs/c4ai-command-a-03-2025/blob/main/README.md", "type": "weights" }, { "label": "Cohere weights license", "url": "https://cohere.com/cohere-labs-cc-by-nc-license", "type": "primary" }, { "label": "Command A technical report", "url": "https://cohere.com/research/papers/command-a-technical-report.pdf", "type": "primary" }, { "label": "Cohere inference research", "url": "https://cohere.com/blog/hardware-aware-dynamic-speculative-decoding", "type": "primary" }, { "label": "Cohere Command A documentation", "url": "https://docs.cohere.com/docs/command-a", "type": "primary" }, { "label": "Cohere capability definitions", "url": "https://raw.githubusercontent.com/cohere-ai/cohere-developer-experience/main/fern/pages/models/the-command-family-of-models/command-a.mdx", "type": "primary" }, { "label": "Cohere capability rendering", "url": "https://raw.githubusercontent.com/cohere-ai/cohere-developer-experience/main/fern/components/model-showcase.tsx", "type": "primary" }, { "label": "Cohere reasoning variant documentation", "url": "https://docs.cohere.com/docs/command-a-reasoning", "type": "primary" }, { "label": "Command A public checkpoint metadata", "url": "https://huggingface.co/api/models/CohereLabs/c4ai-command-a-03-2025", "type": "weights" }, { "label": "Later Cohere family comparison", "url": "https://cohere.com/blog/command-a-plus", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/cohere/command-a-03-2025.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/c4ai-command-a-03-2025", "type": "weights" }, { "label": "Alias documentation", "url": "https://docs.cohere.com/v1/docs/command-a", "type": "docs" } ], "aliases": [ "cohere/command-a" ], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "111B is Cohere's nominal total; active parameters are not separately disclosed. The weights use CC-BY-NC-4.0 with an Acceptable Use Addendum. Supported context is 256,000 tokens; the model card notes a 128K default checkpoint configuration. The reasoning flag describes dedicated thinking-mode support; ReAct tool planning is documented.", "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T12:25:53.434795+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/146.json", "alias_sources": { "cohere/command-a": { "url": "https://docs.cohere.com/v1/docs/command-a", "note": "Alias or alternate catalog name for the same released model; only the retained ID counts as a release." } } }, { "id": "google/gemma-3-12b-it", "slug": "google--gemma-3-12b-it", "name": "Gemma 3 12B IT", "developer": "Google", "developer_id": "google", "family": "gemma", "description": "Google's instruction-tuned Gemma 3 model for multilingual text generation and image understanding.", "release_date": "2025-03-12", "date_precision": "day", "open_weights": true, "license": "gemma", "parameters_b": 12.18732504, "active_parameters_b": 12, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://developers.googleblog.com/en/introducing-gemma3/", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/google/gemma-3-12b-it", "type": "weights" }, { "label": "Introducing Gemma 3: The Developer Guide", "url": "https://developers.googleblog.com/en/introducing-gemma3/", "type": "primary" }, { "label": "Gemma 3 model card", "url": "https://ai.google.dev/gemma/docs/core/model_card_3", "type": "primary" }, { "label": "Gemma 3 Technical Report", "url": "https://storage.googleapis.com/deepmind-media/gemma/Gemma3Report.pdf", "type": "primary" }, { "label": "Pinned official repository metadata", "url": "https://huggingface.co/api/models/google/gemma-3-12b-it/revision/96b6f1eccf38110c56df3a15bffe176da04bfd80", "type": "weights" }, { "label": "Google Gemma PyTorch 12B configuration", "url": "https://github.com/google/gemma_pytorch/blob/014acb7ac4563a5f77c76d7ff98f31b568c16508/gemma/config.py", "type": "primary" }, { "label": "Google SigLIP vision configuration", "url": "https://github.com/google/gemma_pytorch/blob/014acb7ac4563a5f77c76d7ff98f31b568c16508/gemma/siglip_vision/config.py", "type": "primary" }, { "label": "Gemma 3 Transformers configuration", "url": "https://github.com/huggingface/transformers/blob/v4.50.0/src/transformers/models/gemma3/configuration_gemma3.py", "type": "primary" }, { "label": "Gemma 3 Transformers implementation", "url": "https://github.com/huggingface/transformers/blob/v4.50.0/src/transformers/models/gemma3/modeling_gemma3.py", "type": "primary" }, { "label": "SigLIP Transformers implementation", "url": "https://github.com/huggingface/transformers/blob/v4.50.0/src/transformers/models/siglip/modeling_siglip.py", "type": "primary" }, { "label": "Google ADK Gemma 3 tool integration", "url": "https://github.com/google/adk-python/blob/cec52fc005a2c76963feb0a424647ca9576ebb2d/src/google/adk/models/gemma_llm.py", "type": "primary" }, { "label": "Gemma Terms of Use", "url": "https://ai.google.dev/gemma/terms", "type": "primary" }, { "label": "Gemma releases", "url": "https://ai.google.dev/gemma/docs/releases", "type": "primary" }, { "label": "Official repository file metadata", "url": "https://huggingface.co/api/models/google/gemma-3-12b-it/tree/main?recursive=true&expand=true", "type": "weights" }, { "label": "Pinned Gemma 3 checkpoint conversion implementation", "url": "https://github.com/huggingface/transformers/blob/6c5f8f4186c15f4bfbd67941a1736b2b347e0155/src/transformers/models/gemma3/convert_gemma3_weights.py", "type": "primary" }, { "label": "Qwen3 technical report — exact Gemma 3 12B IT active-size comparison", "url": "https://arxiv.org/html/2505.09388v1#S4.T18", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemma-3-12b-it.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Total comprises 11.766034176B language, 0.416866032B vision and 0.004424832B projector parameters, including 245,760 Hugging Face conversion entries. Active 12B follows the Qwen team’s nominal text-benchmark convention, not a Google per-token specification; vision participation depends on image input. Pinned metadata and implementation agree; raw headers were gated. Function calling uses prompting and output parsing. The March 12 launch conflicts with a later release log saying March 10.", "parameter_components_b": { "language_model": 11.766034176, "vision_encoder": 0.416866032, "multimodal_projector": 0.004424832 }, "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-27T13:12:53.956498+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/217.json" }, { "id": "google/gemma-3-12b-pt", "slug": "google--gemma-3-12b-pt", "name": "gemma-3-12b-pt", "developer": "Google", "developer_id": "google", "family": "gemma", "description": null, "release_date": "2025-03-12", "date_precision": "day", "open_weights": true, "license": "Gemma", "parameters_b": 12, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://developers.googleblog.com/en/introducing-gemma3/", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://developers.googleblog.com/en/introducing-gemma3/", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/google/gemma-3-12b-pt/blob/295efb63d01a7017928f273a94ebb86105c9526f/README.md", "type": "model_card" }, { "label": "Developer model card", "url": "https://ai.google.dev/gemma/docs/core/model_card_3", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Pretrained checkpoint. Nominal model size. The 1B model is text-only; larger sizes accept text and images. Weights require agreement to Google’s Gemma terms." }, { "id": "google/gemma-3-1b-it", "slug": "google--gemma-3-1b-it", "name": "gemma-3-1b-it", "developer": "Google", "developer_id": "google", "family": "gemma", "description": null, "release_date": "2025-03-12", "date_precision": "day", "open_weights": true, "license": "Gemma", "parameters_b": 1, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://developers.googleblog.com/en/introducing-gemma3/", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://developers.googleblog.com/en/introducing-gemma3/", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/google/gemma-3-1b-it/blob/dcc83ea841ab6100d6b47a070329e1ba4cf78752/README.md", "type": "model_card" }, { "label": "Developer model card", "url": "https://ai.google.dev/gemma/docs/core/model_card_3", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Instruction-tuned checkpoint. Nominal model size. The 1B model is text-only; larger sizes accept text and images. Weights require agreement to Google’s Gemma terms." }, { "id": "google/gemma-3-1b-pt", "slug": "google--gemma-3-1b-pt", "name": "gemma-3-1b-pt", "developer": "Google", "developer_id": "google", "family": "gemma", "description": null, "release_date": "2025-03-12", "date_precision": "day", "open_weights": true, "license": "Gemma", "parameters_b": 1, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://developers.googleblog.com/en/introducing-gemma3/", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://developers.googleblog.com/en/introducing-gemma3/", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/google/gemma-3-1b-pt/blob/fcf18a2a879aab110ca39f8bffbccd5d49d8eb29/README.md", "type": "model_card" }, { "label": "Developer model card", "url": "https://ai.google.dev/gemma/docs/core/model_card_3", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Pretrained checkpoint. Nominal model size. The 1B model is text-only; larger sizes accept text and images. Weights require agreement to Google’s Gemma terms." }, { "id": "google/gemma-3-27b-it", "slug": "google--gemma-3-27b-it", "name": "Gemma 3 27B IT", "developer": "Google", "developer_id": "google", "family": "gemma", "description": "Instruction-tuned Gemma 3 model for multilingual text generation and image understanding.", "release_date": "2025-03-12", "date_precision": "day", "open_weights": true, "license": "gemma", "parameters_b": 27.43240664, "active_parameters_b": 27, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://developers.googleblog.com/en/introducing-gemma3/", "source_type": "primary", "sources": [ { "label": "Gemma 3 model card", "url": "https://ai.google.dev/gemma/docs/core/model_card_3", "type": "primary" }, { "label": "Introducing Gemma 3: The Developer Guide", "url": "https://developers.googleblog.com/en/introducing-gemma3/", "type": "primary" }, { "label": "Gemma 3 Technical Report", "url": "https://arxiv.org/html/2503.19786v1", "type": "primary" }, { "label": "Hugging Face", "url": "https://huggingface.co/google/gemma-3-27b-it", "type": "weights" }, { "label": "Pinned official Gemma 3 27B IT repository metadata", "url": "https://huggingface.co/api/models/google/gemma-3-27b-it/revision/005ad3404e59d6023443cb575daa05336842228a", "type": "weights" }, { "label": "Pinned official Gemma 3 27B IT file metadata", "url": "https://huggingface.co/api/models/google/gemma-3-27b-it/tree/005ad3404e59d6023443cb575daa05336842228a?recursive=true&expand=true", "type": "weights" }, { "label": "Public copy of original Gemma 3 27B IT configuration", "url": "https://huggingface.co/second-state/gemma-3-27b-it-GGUF/raw/main/config.json", "type": "weights" }, { "label": "Transformers v4.50.0 Gemma 3 implementation", "url": "https://github.com/huggingface/transformers/blob/v4.50.0/src/transformers/models/gemma3/modeling_gemma3.py", "type": "primary" }, { "label": "Transformers v4.50.0 Gemma 3 configuration defaults", "url": "https://github.com/huggingface/transformers/blob/v4.50.0/src/transformers/models/gemma3/configuration_gemma3.py", "type": "primary" }, { "label": "Transformers v4.50.0 SigLIP implementation", "url": "https://github.com/huggingface/transformers/blob/v4.50.0/src/transformers/models/siglip/modeling_siglip.py", "type": "primary" }, { "label": "Official Google Gemma PyTorch model configurations", "url": "https://github.com/google/gemma_pytorch/blob/main/gemma/config.py", "type": "primary" }, { "label": "Gemma Terms of Use", "url": "https://ai.google.dev/gemma/terms", "type": "primary" }, { "label": "Google Gemma models for ADK agents", "url": "https://github.com/google/adk-docs/blob/main/docs/agents/models/google-gemma.md", "type": "primary" }, { "label": "Pinned Gemma 3 checkpoint conversion implementation", "url": "https://github.com/huggingface/transformers/blob/6c5f8f4186c15f4bfbd67941a1736b2b347e0155/src/transformers/models/gemma3/convert_gemma3_weights.py", "type": "primary" }, { "label": "Gemma release chronology", "url": "https://ai.google.dev/gemma/docs/releases", "type": "primary" }, { "label": "Qwen3 Table 16: nominal activated parameters", "url": "https://arxiv.org/html/2505.09388v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemma-3-27b-it.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Total includes 27,009,346,304 language, 416,866,032 vision and 6,194,304 projector parameters, with tied embeddings counted once. The language count includes 344,064 entries added during Hugging Face conversion. Pinned metadata and implementation agree; raw tensor headers were inaccessible. Active 27B follows Qwen’s nominal text-benchmark convention, not an exact per-token count or a separate Google disclosure; vision participation varies with input. Google supports function calling through prompting and output parsing. The March 12 launch announcement conflicts with a later release log saying March 10.", "parameter_components_b": { "language_model": 27.009346304, "vision_encoder": 0.416866032, "multimodal_projector": 0.006194304 }, "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-27T13:14:03.620912+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/218.json" }, { "id": "google/gemma-3-27b-pt", "slug": "google--gemma-3-27b-pt", "name": "gemma-3-27b-pt", "developer": "Google", "developer_id": "google", "family": "gemma", "description": null, "release_date": "2025-03-12", "date_precision": "day", "open_weights": true, "license": "Gemma", "parameters_b": 27, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://developers.googleblog.com/en/introducing-gemma3/", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://developers.googleblog.com/en/introducing-gemma3/", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/google/gemma-3-27b-pt/blob/9fe3c4ebc93fbadb14913801536d022054ef11cc/README.md", "type": "model_card" }, { "label": "Developer model card", "url": "https://ai.google.dev/gemma/docs/core/model_card_3", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Pretrained checkpoint. Nominal model size. The 1B model is text-only; larger sizes accept text and images. Weights require agreement to Google’s Gemma terms." }, { "id": "google/gemma-3-4b-it", "slug": "google--gemma-3-4b-it", "name": "Gemma 3 4B IT", "developer": "Google", "developer_id": "google", "family": "gemma", "description": "Google's instruction-tuned Gemma 3 model with text and image input and text output.", "release_date": "2025-03-12", "date_precision": "day", "open_weights": true, "license": "Gemma Terms of Use", "parameters_b": 4.300079472, "active_parameters_b": 4, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-3/", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/google/gemma-3-4b-it", "type": "weights" }, { "label": "Introducing Gemma 3", "url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-3/", "type": "primary" }, { "label": "Gemma 3 model card", "url": "https://ai.google.dev/gemma/docs/core/model_card_3", "type": "primary" }, { "label": "Gemma 3 Technical Report, March 12, 2025", "url": "https://storage.googleapis.com/deepmind-media/gemma/Gemma3Report.pdf", "type": "primary" }, { "label": "Exact official checkpoint metadata", "url": "https://huggingface.co/api/models/google/gemma-3-4b-it/revision/093f9f388b31de276ce2de164bdc2081324b9767", "type": "weights" }, { "label": "Official Gemma 3 Orbax to Transformers converter", "url": "https://github.com/huggingface/transformers/blob/6c5f8f4186c15f4bfbd67941a1736b2b347e0155/src/transformers/models/gemma3/convert_gemma3_weights.py", "type": "primary" }, { "label": "Transformers Gemma 3 implementation", "url": "https://github.com/huggingface/transformers/blob/6c5f8f4186c15f4bfbd67941a1736b2b347e0155/src/transformers/models/gemma3/modeling_gemma3.py", "type": "primary" }, { "label": "Google SigLIP vision implementation", "url": "https://github.com/google/gemma_pytorch/blob/main/gemma/siglip_vision/siglip_vision_model.py", "type": "primary" }, { "label": "Transformers Gemma 3 configuration", "url": "https://github.com/huggingface/transformers/blob/6c5f8f4186c15f4bfbd67941a1736b2b347e0155/src/transformers/models/gemma3/configuration_gemma3.py", "type": "primary" }, { "label": "Gemma Terms of Use", "url": "https://ai.google.dev/gemma/terms", "type": "primary" }, { "label": "Google ADK Gemma 3 function-calling adapter", "url": "https://github.com/google/adk-python/blob/main/src/google/adk/models/gemma_llm.py", "type": "primary" }, { "label": "Exact official checkpoint file history", "url": "https://huggingface.co/api/models/google/gemma-3-4b-it/tree/main?recursive=true&expand=true", "type": "weights" }, { "label": "Google Gemma 3 4B reference configuration", "url": "https://github.com/google/gemma_pytorch/blob/main/gemma/config.py", "type": "primary" }, { "label": "Gemma release chronology", "url": "https://ai.google.dev/gemma/docs/releases", "type": "primary" }, { "label": "Qwen3 Technical Report, Table 7", "url": "https://arxiv.org/html/2505.09388v1", "type": "primary" }, { "label": "Pinned Google Gemma 3 4B PT base metadata", "url": "https://huggingface.co/api/models/google/gemma-3-4b-pt/revision/cc012e0a6d0787b4adcc0fa2c4da74402494554d", "type": "weights" }, { "label": "Kimi-VL Technical Report, Figure 1", "url": "https://arxiv.org/html/2504.07491v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemma-3-4b-it.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Total 4.300079472B includes language, vision and projector weights, with tied embeddings counted once and 163,840 entries added during Hugging Face conversion. Active 4B is nominal under Qwen Team's published base-model convention, supported by Google's explicit PT-to-IT lineage and Kimi Team's approximate exact-IT comparison plot; it is not an exact per-token count or a separate Google active disclosure. Function calling uses prompting and output parsing. The March 12 launch announcement conflicts with a later release log saying March 10.", "parameter_components_b": { "vision_encoder": 0.416866032, "multimodal_projector": 0.002950272, "language_model": 3.880263168 }, "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-27T13:18:31.393Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/219.json" }, { "id": "google/gemma-3-4b-pt", "slug": "google--gemma-3-4b-pt", "name": "gemma-3-4b-pt", "developer": "Google", "developer_id": "google", "family": "gemma", "description": null, "release_date": "2025-03-12", "date_precision": "day", "open_weights": true, "license": "Gemma", "parameters_b": 4, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://developers.googleblog.com/en/introducing-gemma3/", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://developers.googleblog.com/en/introducing-gemma3/", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/google/gemma-3-4b-pt/blob/cc012e0a6d0787b4adcc0fa2c4da74402494554d/README.md", "type": "model_card" }, { "label": "Developer model card", "url": "https://ai.google.dev/gemma/docs/core/model_card_3", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Pretrained checkpoint. Nominal model size. The 1B model is text-only; larger sizes accept text and images. Weights require agreement to Google’s Gemma terms." }, { "id": "rekaai/reka-flash-3", "slug": "rekaai--reka-flash-3", "name": "Reka Flash 3", "developer": "Reka AI", "developer_id": "rekaai", "family": "reka", "description": "Dense text reasoning model with 20.9B parameters, pretrained from scratch and post-trained with supervised learning and reinforcement learning.", "release_date": "2025-03-10", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 20.90548224, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://reka.ai/news/introducing-reka-flash", "source_type": "primary", "sources": [ { "label": "Reasoning with Reka Flash 3", "url": "https://reka.ai/news/introducing-reka-flash", "type": "primary" }, { "label": "Reka Flash 3 model card at pinned release revision", "url": "https://huggingface.co/RekaAI/reka-flash-3/blob/8d251ecb02eca9f0f4aa052b29ef922628d2806f/README.md", "type": "primary" }, { "label": "Flash 3 checkpoint configuration", "url": "https://huggingface.co/RekaAI/reka-flash-3/blob/8d251ecb02eca9f0f4aa052b29ef922628d2806f/config.json", "type": "primary" }, { "label": "Flash 3 complete five-shard tensor index", "url": "https://huggingface.co/RekaAI/reka-flash-3/blob/8d251ecb02eca9f0f4aa052b29ef922628d2806f/model.safetensors.index.json", "type": "primary" }, { "label": "Current Reka Gateway Flash 3 model page", "url": "https://developer.reka.ai/models/reka-flash-3", "type": "primary" }, { "label": "Current Gateway native tool-calling support", "url": "https://developer.reka.ai/tool-calling/patterns", "type": "primary" }, { "label": "Flash 3.1 research: explicit Flash 3 update and nominal size", "url": "https://reka.ai/news/reinforcement-learning-for-reka-flash-3-1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/rekaai/reka-flash-3.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Public checkpoint revision 8d251ecb02eca9f0f4aa052b29ef922628d2806f has 20,905,482,240 learned parameters (Reka nominal 21B), including separate input and output embeddings. No separately supported active count was found in the original release or later Flash 3.1 research. Its config/tokenizer context is 32,768 tokens; current Gateway advertises 64k hosted context. Function calling is an original developer claim about the model, while the current Gateway does not return native tool_calls for this identifier. Nexus and multimodal application variants are separate.", "parameter_components_b": { "input_embedding": 0.616562688, "output_head": 0.616562688, "decoder_layers": 19.67235072, "final_rmsnorm": 0.000006144 }, "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T17:43:07.163845+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/479.json" }, { "id": "alibaba/qwq-32b", "slug": "alibaba--qwq-32b", "name": "QwQ 32B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Open reasoning model from the Qwen team for math, coding, and step-by-step problem solving", "release_date": "2025-03-05", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 32.763876352, "active_parameters_b": 32, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://qwenlm.github.io/blog/qwq-32b/", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/QwQ-32B", "type": "weights" }, { "label": "QwQ-32B: Embracing the Power of Reinforcement Learning", "url": "https://qwenlm.github.io/blog/qwq-32b/", "type": "primary" }, { "label": "Qwen3 Technical Report, Table 13", "url": "https://arxiv.org/html/2505.09388v1#S4.T13", "type": "primary" }, { "label": "QwQ-32B checkpoint configuration", "url": "https://huggingface.co/Qwen/QwQ-32B/blob/976055f8c83f394f35dbd3ab09a285a984907bd0/config.json", "type": "primary" }, { "label": "QwQ-32B published checkpoint tensor index", "url": "https://huggingface.co/Qwen/QwQ-32B/blob/976055f8c83f394f35dbd3ab09a285a984907bd0/model.safetensors.index.json", "type": "primary" }, { "label": "QwQ-32B repository and safetensors metadata", "url": "https://huggingface.co/api/models/Qwen/QwQ-32B", "type": "primary" }, { "label": "QwQ-32B official chat template", "url": "https://huggingface.co/Qwen/QwQ-32B/blob/976055f8c83f394f35dbd3ab09a285a984907bd0/tokenizer_config.json", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwq-32b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Exact BF16 checkpoint total is 32,763,876,352 parameters; the model card states 32.5B and Qwen3 Table 13 uses nominal 32B total/active. Active count is the explicitly reported nominal 32B, not an exact per-token count. Full 131,072-token context requires documented YaRN configuration. Release day is March 5 UTC; the announcement displays March 6 in UTC+08:00.", "field_status": {}, "audited_at": "2026-10-26T15:53:23.087438Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/075.json" }, { "id": "alibaba/qwq-plus", "slug": "alibaba--qwq-plus", "name": "QwQ Plus", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Qwen reasoning model for deliberate problem solving, math, and coding", "release_date": "2025-03-05", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "source_type": "primary", "sources": [ { "label": "Alibaba Cloud explicit release table", "url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "type": "primary" }, { "label": "QwQ-Plus last-documented model information", "url": "https://www.alibabacloud.com/help/en/model-studio/qwq-plus", "type": "primary" }, { "label": "Qwen3 technical report: distinct QwQ-32B comparison column", "url": "https://arxiv.org/html/2505.09388v1", "type": "primary" }, { "label": "Separate QwQ-32B checkpoint card", "url": "https://huggingface.co/Qwen/QwQ-32B", "type": "primary" }, { "label": "Alibaba Cloud announced legacy-model retirement", "url": "https://www.alibabacloud.com/en/notice/model_studio_notice_on_the_postponed_retirement_of_selected_legacy_models_7d0?_p_lc=1", "type": "primary" }, { "label": "Original QwQ-32B launch source: API example names qwq-32b", "url": "https://raw.githubusercontent.com/QwenLM/qwenlm.github.io/1b64f88aab1f8d40c87e61966ca4eb0a030cc155/content/blog/qwq-32b/index.md", "type": "primary" }, { "label": "Initial QwQ README: no Plus identity mapping", "url": "https://raw.githubusercontent.com/QwenLM/QwQ/1b0141f875dd58de25cb55d0b62b0782710f1de0/README.md", "type": "primary" }, { "label": "Original QwQ-32B model-card README", "url": "https://huggingface.co/Qwen/QwQ-32B/raw/20dea38711d9a58b751574a2e34a5fb782e75513/README.md", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=qwq-plus&limit=100", "label": "Official Qwen inventory search for the exact hosted model", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=QwQ&limit=1000", "label": "Official Qwen QwQ weight inventory", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwq-plus.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "QwQ-Plus has documented Qwen2.5 family lineage but no verified size-bearing base relation; total and active parameter counts remain undisclosed. QwQ-32B counts are not transferable from the available primary evidence. Architecture remains undisclosed. No public Plus weight release was found. Function Calling was documented for Beijing but not Singapore. Context and modalities are last-documented values from the October 10, 2026 page, which now redirects. Alibaba announced October 10, 2026 retirement; current operation was not tested. Classified as closed weights: no public release of this hosted variant was found in the developer documentation and weight inventory.", "field_status": { "license": "unverified", "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-28T08:01:10.208022+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/076.json" }, { "id": "cohere/c4ai-aya-vision-32b", "slug": "cohere--c4ai-aya-vision-32b", "name": "Aya Vision 32B", "developer": "Cohere", "developer_id": "cohere", "family": "aya-vision", "description": "Open multilingual vision model for OCR, visual reasoning, and image question answering", "release_date": "2025-03-04", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-4.0", "parameters_b": 33.1, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 16000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/cohere/c4ai-aya-vision-32b.toml", "source_type": "catalog", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/aya-vision-32b", "type": "weights" }, { "label": "Aya Vision technical report", "url": "https://arxiv.org/html/2505.08751v1", "type": "primary" }, { "label": "Aya Vision API documentation", "url": "https://docs.cohere.com/docs/aya-vision", "type": "primary" }, { "label": "Official release note", "url": "https://docs.cohere.com/changelog/aya-vision-announcement", "type": "primary" }, { "label": "Cohere Aya Vision implementation", "url": "https://raw.githubusercontent.com/huggingface/transformers/v4.49.0-AyaVision/src/transformers/models/aya_vision/modeling_aya_vision.py", "type": "primary" }, { "label": "Cohere decoder implementation", "url": "https://raw.githubusercontent.com/huggingface/transformers/v4.49.0-AyaVision/src/transformers/models/cohere/modeling_cohere.py", "type": "primary" }, { "label": "Public checkpoint metadata", "url": "https://huggingface.co/api/models/CohereLabs/aya-vision-32b", "type": "weights" }, { "label": "License and acceptable-use addendum", "url": "https://cohere.com/cohere-labs-cc-by-nc-license", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/cohere/c4ai-aya-vision-32b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Approximately 33.1B total: 32.3B language model, 0.4B vision encoder and 0.428B connector, from the developer's rounded component counts. Active parameters are not separately disclosed for either the complete model or its language backbone. The paper's patch16-512 encoder conflicts with the model card's patch14-384 description. The CC-BY-NC-4.0 release also requires Cohere Labs' acceptable-use addendum. Dedicated reasoning-mode and native tool-calling flags remain unverified.", "parameter_components_b": { "language_model": 32.3, "vision_encoder": 0.4, "vision_language_connector": 0.428 }, "field_status": { "active_parameters_b": "undisclosed", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T11:15:54.637579Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/144.json" }, { "id": "cohere/c4ai-aya-vision-8b", "slug": "cohere--c4ai-aya-vision-8b", "name": "Aya Vision 8B", "developer": "Cohere", "developer_id": "cohere", "family": "aya-vision", "description": "Multilingual vision-language model for OCR, image captioning and visual question answering across 23 languages.", "release_date": "2025-03-04", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-4.0", "parameters_b": 8.6, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 16000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/cohere/c4ai-aya-vision-8b.toml", "source_type": "catalog", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/aya-vision-8b", "type": "weights" }, { "label": "Aya Vision technical report", "url": "https://arxiv.org/html/2505.08751v1", "type": "primary" }, { "label": "Aya Vision launch", "url": "https://cohere.com/blog/aya-vision", "type": "primary" }, { "label": "Cohere model specifications", "url": "https://docs.cohere.com/docs/models", "type": "primary" }, { "label": "Cohere CC-BY-NC license and addendum", "url": "https://cohere.com/cohere-labs-cc-by-nc-license", "type": "primary" }, { "label": "Aya Vision repository metadata", "url": "https://huggingface.co/api/models/CohereLabs/aya-vision-8b", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/cohere/c4ai-aya-vision-8b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Approximately 8.6B complete parameters from the developer report's rounded components: 8B language model, 0.4B vision encoder and 0.19B connector. Active parameters are not separately disclosed. CC-BY-NC-4.0 includes Cohere's Acceptable Use Addendum. The Cohere API identifier was retired April 4, 2026.", "parameter_components_b": { "language_model": 8, "vision_encoder": 0.4, "vision_language_connector": 0.19 }, "field_status": { "active_parameters_b": "undisclosed", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T12:26:13.903333+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/145.json" }, { "id": "cohere/command-r7b-arabic-02-2025", "slug": "cohere--command-r7b-arabic-02-2025", "name": "Command R7B Arabic", "developer": "Cohere", "developer_id": "cohere", "family": "command-r", "description": "Open Command R model optimized for Arabic enterprise chat, RAG, and cultural knowledge", "release_date": "2025-02-27", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-4.0", "parameters_b": 8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://cohere.com/blog/command-r7b-arabic", "source_type": "primary", "sources": [ { "label": "Cohere Labs exact model card", "url": "https://huggingface.co/CohereLabs/c4ai-command-r7b-arabic-02-2025/blob/main/README.md", "type": "primary" }, { "label": "Exact checkpoint configuration metadata", "url": "https://huggingface.co/api/models/CohereLabs/c4ai-command-r7b-arabic-02-2025", "type": "primary" }, { "label": "Cohere2 architecture implementation", "url": "https://github.com/huggingface/transformers/blob/v4.49.0/src/transformers/models/cohere2/modeling_cohere2.py", "type": "primary" }, { "label": "Cohere Arabic launch announcement", "url": "https://cohere.com/blog/command-r7b-arabic", "type": "primary" }, { "label": "Command R7B Arabic technical report", "url": "https://arxiv.org/html/2503.14603v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/cohere/command-r7b-arabic-02-2025.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/c4ai-command-r7b-arabic-02-2025", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Approximately 8B complete parameters: 7B Transformer plus 1B embeddings, per Cohere. Active parameters are not separately disclosed. Open weights require CC-BY-NC-4.0 and Cohere Labs acceptable-use terms. The reasoning flag describes a dedicated reasoning mode; native tool planning remains supported.", "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T12:30:49.699153+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/154.json" }, { "id": "ibm/granite-3.2-2b-instruct", "slug": "ibm--granite-3.2-2b-instruct", "name": "granite-3.2-2b-instruct", "developer": "IBM", "developer_id": "ibm", "family": "granite", "description": null, "release_date": "2025-02-26", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 2, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://newsroom.ibm.com/2025-02-26-ibm-expands-granite-model-family-with-new-multi-modal-and-reasoning-ai-built-for-the-enterprise", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://newsroom.ibm.com/2025-02-26-ibm-expands-granite-model-family-with-new-multi-modal-and-reasoning-ai-built-for-the-enterprise", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/ibm-granite/granite-3.2-2b-instruct/blob/641593c3b25bec0b1efe9f0f7d7a67f7243f86a3/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/ibm-granite/granite-3.2-2b-instruct/blob/641593c3b25bec0b1efe9f0f7d7a67f7243f86a3/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Instruction model with optional chain-of-thought reasoning. Nominal model-card parameter count; the developer release date is stated explicitly." }, { "id": "ibm/granite-3.2-8b-instruct", "slug": "ibm--granite-3.2-8b-instruct", "name": "granite-3.2-8b-instruct", "developer": "IBM", "developer_id": "ibm", "family": "granite", "description": null, "release_date": "2025-02-26", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://newsroom.ibm.com/2025-02-26-ibm-expands-granite-model-family-with-new-multi-modal-and-reasoning-ai-built-for-the-enterprise", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://newsroom.ibm.com/2025-02-26-ibm-expands-granite-model-family-with-new-multi-modal-and-reasoning-ai-built-for-the-enterprise", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/ibm-granite/granite-3.2-8b-instruct/blob/610d8c6ee9c84ce51f6dfd7bc5c0215d95d49695/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/ibm-granite/granite-3.2-8b-instruct/blob/610d8c6ee9c84ce51f6dfd7bc5c0215d95d49695/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Instruction model with optional chain-of-thought reasoning. Nominal model-card parameter count; the developer release date is stated explicitly." }, { "id": "ibm/granite-vision-3.2-2b", "slug": "ibm--granite-vision-3.2-2b", "name": "granite-vision-3.2-2b", "developer": "IBM", "developer_id": "ibm", "family": "granite", "description": null, "release_date": "2025-02-26", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 2.975396928, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://newsroom.ibm.com/2025-02-26-ibm-expands-granite-model-family-with-new-multi-modal-and-reasoning-ai-built-for-the-enterprise", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://newsroom.ibm.com/2025-02-26-ibm-expands-granite-model-family-with-new-multi-modal-and-reasoning-ai-built-for-the-enterprise", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/ibm-granite/granite-vision-3.2-2b/blob/ec894191f2e9e9be3d90eab9ac6949583c46f03b/README.md", "type": "model_card" }, { "label": "Official checkpoint parameter metadata", "url": "https://huggingface.co/api/models/ibm-granite/granite-vision-3.2-2b/revision/ec894191f2e9e9be3d90eab9ac6949583c46f03b", "type": "config" }, { "label": "Official model configuration", "url": "https://huggingface.co/ibm-granite/granite-vision-3.2-2b/blob/ec894191f2e9e9be3d90eab9ac6949583c46f03b/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Vision-language model for document understanding and question answering. The nominal 2B name describes the language backbone; the total here includes the vision components in the published checkpoint metadata." }, { "id": "microsoft/phi-4-mini", "slug": "microsoft--phi-4-mini", "name": "Phi-4-mini", "developer": "Microsoft", "developer_id": "microsoft", "family": "phi", "description": "Microsoft's multilingual text instruction model, released as Phi-4-mini-instruct, with native function calling.", "release_date": "2025-02-26", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 3.83602176, "active_parameters_b": 3.8, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://huggingface.co/microsoft/Phi-4-mini-instruct", "source_type": "primary", "sources": [ { "label": "Empowering innovation: The next generation of the Phi family", "url": "https://azure.microsoft.com/en-us/blog/empowering-innovation-the-next-generation-of-the-phi-family/", "type": "primary" }, { "label": "Phi-4-mini-instruct official model card", "url": "https://huggingface.co/microsoft/Phi-4-mini-instruct/blob/cfbefacb99257ffa30c83adab238a50856ac3083/README.md", "type": "primary" }, { "label": "Phi-4-mini-instruct checkpoint configuration", "url": "https://huggingface.co/microsoft/Phi-4-mini-instruct/blob/cfbefacb99257ffa30c83adab238a50856ac3083/config.json", "type": "primary" }, { "label": "Phi-4-mini-instruct model implementation", "url": "https://huggingface.co/microsoft/Phi-4-mini-instruct/blob/cfbefacb99257ffa30c83adab238a50856ac3083/modeling_phi3.py", "type": "primary" }, { "label": "Phi-4-mini-instruct checkpoint tensor index", "url": "https://huggingface.co/microsoft/Phi-4-mini-instruct/blob/cfbefacb99257ffa30c83adab238a50856ac3083/model.safetensors.index.json", "type": "primary" }, { "label": "Phi-4-mini-instruct first shard header", "url": "https://huggingface.co/microsoft/Phi-4-mini-instruct/resolve/cfbefacb99257ffa30c83adab238a50856ac3083/model-00001-of-00002.safetensors", "type": "primary" }, { "label": "Phi-4-mini-instruct second shard header", "url": "https://huggingface.co/microsoft/Phi-4-mini-instruct/resolve/cfbefacb99257ffa30c83adab238a50856ac3083/model-00002-of-00002.safetensors", "type": "primary" }, { "label": "Phi-4-mini-instruct MIT license", "url": "https://huggingface.co/microsoft/Phi-4-mini-instruct/blob/cfbefacb99257ffa30c83adab238a50856ac3083/LICENSE", "type": "primary" }, { "label": "Phi-4-Mini Technical Report", "url": "https://arxiv.org/html/2503.01743v1", "type": "primary" }, { "label": "Phi-4-Mini-Reasoning: Exploring the Limits of Small Reasoning Language Models in Math", "url": "https://arxiv.org/html/2504.21233v1", "type": "primary" }, { "label": "Phi-4-mini-instruct repository metadata", "url": "https://huggingface.co/api/models/microsoft/Phi-4-mini-instruct", "type": "primary" }, { "label": "Qwen3 Technical Report, Table 20", "url": "https://arxiv.org/html/2505.09388v1#S4.T20", "type": "primary" }, { "label": "Sigma-MoE-Tiny Technical Report, Figure 1", "url": "https://arxiv.org/html/2512.16248v2#S0.F1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/microsoft/phi-4-mini.toml", "type": "catalog" }, { "label": "Weights", "url": "https://huggingface.co/microsoft/Phi-4-mini-instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Phi-4-mini is the launch name of Phi-4-mini-instruct. Exact checkpoint count is 3,836,021,760 learned parameters with tied input/output embedding counted once; Microsoft publishes nominal 3.8B. Active 3.8B follows Qwen Team's explicit benchmark convention, corroborated approximately by Microsoft Research's later Sigma comparison; neither is an exact per-token tensor count. Config specifies 131,072 context tokens; launch prose says 128,000 and the card says 128K. The dedicated reasoning model is a separate release.", "field_status": {}, "audited_at": "2026-10-27T15:22:45.567447Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/295.json" }, { "id": "microsoft/phi-4-multimodal-instruct", "slug": "microsoft--phi-4-multimodal-instruct", "name": "Phi-4-multimodal-instruct", "developer": "Microsoft", "developer_id": "microsoft", "family": "phi", "description": "Microsoft's multimodal instruction model for text, image, and audio understanding, with text output and modality-specific LoRA adapters.", "release_date": "2025-02-26", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 5.574460224, "active_parameters_b": 4, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "image", "audio" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://huggingface.co/microsoft/Phi-4-multimodal-instruct", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/microsoft/Phi-4-multimodal-instruct", "type": "weights" }, { "label": "Microsoft release announcement", "url": "https://azure.microsoft.com/en-us/blog/empowering-innovation-the-next-generation-of-the-phi-family/", "type": "primary" }, { "label": "Microsoft technical report", "url": "https://arxiv.org/html/2503.01743v2", "type": "primary" }, { "label": "Pinned model configuration", "url": "https://huggingface.co/microsoft/Phi-4-multimodal-instruct/blob/93f923e1a7727d1c4f446756212d9d3e8fcc5d81/config.json", "type": "primary" }, { "label": "Pinned tensor index", "url": "https://huggingface.co/microsoft/Phi-4-multimodal-instruct/blob/93f923e1a7727d1c4f446756212d9d3e8fcc5d81/model.safetensors.index.json", "type": "weights" }, { "label": "Speech encoder buffer definitions", "url": "https://huggingface.co/microsoft/Phi-4-multimodal-instruct/blob/93f923e1a7727d1c4f446756212d9d3e8fcc5d81/speech_conformer_encoder.py", "type": "primary" }, { "label": "Microsoft non-thinking model comparison", "url": "https://arxiv.org/html/2603.03975v1", "type": "primary" }, { "label": "RSCC researchers' nominal active-parameter convention (Table 5)", "url": "https://papers.nips.cc/paper_files/paper/2025/file/62867024377cac4233195949b9db0ebd-Paper-Datasets_and_Benchmarks_Track.pdf", "type": "primary" }, { "label": "RSCC authors' exact baseline checkpoint identification", "url": "https://github.com/Bili-Sakura/RSCC", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/microsoft/phi-4-multimodal-instruct.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete checkpoint: 5.574460224B learned parameters, nominally 5.6B; excludes 160 nonlearned buffer elements and includes both modality LoRAs once. Nominal 4B active follows Chen et al.'s RSCC benchmark convention, whose table lists Phi-4-Mini 4B and SigLIP (LORA) separately; exact component inclusion and per-token counts are unspecified. Context is 131072 tokens; recommended audio lengths are 40 seconds generally or 30 minutes for summarization.", "parameter_components_b": { "language_backbone": 3.83602176, "vision_encoder_projector": 0.441550016, "audio_encoder_projectors": 0.466416256, "vision_lora": 0.369098752, "speech_lora": 0.46137344 }, "field_status": {}, "audited_at": "2026-10-27T15:22:27.003381+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/296.json" }, { "id": "anthropic/claude-3-7-sonnet-20250219", "slug": "anthropic--claude-3-7-sonnet-20250219", "name": "Claude Sonnet 3.7", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude-sonnet", "description": "Anthropic's Claude Sonnet model with optional extended thinking, text and image understanding, and tool use for coding and analysis.", "release_date": "2025-02-24", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 200000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://www.anthropic.com/news/claude-3-7-sonnet", "source_type": "primary", "sources": [ { "label": "Claude 3.7 Sonnet release", "url": "https://www.anthropic.com/news/claude-3-7-sonnet", "type": "announcement" }, { "label": "Exact-model interpretability research", "url": "https://transformer-circuits.pub/2025/introspection/index.html", "type": "paper" }, { "label": "Anthropic transparency summary", "url": "https://www.anthropic.com/transparency", "type": "documentation" }, { "label": "Anthropic model deprecations", "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "type": "documentation" }, { "label": "Anthropic Bash tool compatibility", "url": "https://platform.claude.com/docs/en/agents-and-tools/tool-use/bash-tool", "type": "documentation" }, { "label": "Anthropic historical context specification", "url": "https://platform.claude.com/docs/en/build-with-claude/claude-on-amazon-bedrock-legacy", "type": "documentation" }, { "label": "Anthropic pricing PDF with conflicting 1M row", "url": "https://www-cdn.anthropic.com/files/4zrzovbb/website/5678bc2f5978e5bcd4f1fe7c14b2c72284dcf9f8.pdf", "type": "documentation" }, { "label": "Google retained partner-model specifications", "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/deprecations/partner-models", "type": "documentation" }, { "label": "AWS beta-feature compatibility", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-anthropic-claude-messages-request-response.html", "type": "documentation" }, { "label": "GLM-5.3 and the spread of advanced cyber capabilities", "url": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities", "type": "documentation" }, { "label": "Commercial Terms of Service", "url": "https://www.anthropic.com/legal/commercial-terms", "type": "documentation" }, { "label": "Citations API and PDF support for Claude models now in Amazon Bedrock", "url": "https://aws.amazon.com/about-aws/whats-new/2025/06/citations-api-pdf-claude-models-amazon-bedrock/", "type": "documentation" }, { "label": "Claude Sonnet 4.5 System Card", "url": "https://www-cdn.anthropic.com/963373e433e489a87a10c823c52a0a013e9172dd.pdf", "type": "model_card" }, { "label": "Claude 3.7 Sonnet System Card", "url": "https://www-cdn.anthropic.com/9ff93dfa8f445c932415d335c88852ef47f1201e.pdf", "type": "model_card" }, { "label": "Do reasoning models use their scratchpad like we do? Evidence from distilling paraphrases", "url": "https://alignment.anthropic.com/2025/distill-paraphrases/", "type": "paper" }, { "label": "Claude Opus 4 and Claude Sonnet 4 System Card — current official PDF", "url": "https://www.anthropic.com/claude-4-system-card", "type": "model_card" }, { "label": "Claude Haiku 4.5 System Card", "url": "https://assets.anthropic.com/m/99128ddd009bdcb/original/Claude-Haiku-4-5-System-Card.pdf", "type": "model_card" }, { "label": "Claude Sonnet 5.5 System Card", "url": "https://www.anthropic.com/claude-sonnet-5-5-system-card", "type": "model_card" }, { "label": "Claude Opus 5.5 System Card", "url": "https://www.anthropic.com/claude-opus-5-5-system-card", "type": "model_card" }, { "label": "Claude Haiku 5.5 System Card", "url": "https://www.anthropic.com/claude-haiku-5-5-system-card", "type": "model_card" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/anthropic/claude-3-7-sonnet-20250219.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The snapshot identifier contains February 19; public availability began February 24, 2025. Anthropic retired this API ID on February 19, 2026; capabilities here are historical documented specifications. Transformer structure is supported by exact-model interpretability research, but dense/MoE details and total/active parameter counts remain undisclosed. The explicitly named Claude 3.7 Sonnet base model also has no published size. Retain the documented 200K context: Anthropic's May 12, 2026 pricing PDF contains conflicting 1M API rows after retirement, without a corroborating exact-model availability specification. The 128K beta limit refers to extended output, not context. Google's retained card also conflicts with Anthropic's explicit extended-thinking support.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T10:20:04Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/092.json" }, { "id": "mistral/mistral-saba", "slug": "mistral--mistral-saba", "name": "Mistral Saba", "developer": "Mistral", "developer_id": "mistral", "family": "mistral", "description": "Mistral's 24B text model fine-tuned for Arabic and South Asian languages, including Tamil.", "release_date": "2025-02-17", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": 24, "active_parameters_b": 24, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://mistral.ai/news/mistral-saba/", "source_type": "primary", "sources": [ { "label": "Mistral Saba", "url": "https://mistral.ai/news/mistral-saba/", "type": "primary" }, { "label": "Mistral Saba v25.02 model documentation", "url": "https://docs.mistral.ai/models/mistral-saba-25-02", "type": "primary" }, { "label": "Mistral Saba 24B - GroqDocs", "url": "https://console.groq.com/docs/model/mistral-saba-24b", "type": "primary" }, { "label": "Mistral Saba model lifecycle", "url": "https://legal.mistral.ai/ai-governance/models/mistral-saba/", "type": "primary" }, { "label": "Pinned official Mistral Saba 25.02 model definition", "url": "https://raw.githubusercontent.com/mistralai/platform-docs-public/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/schema/models/models/mistral-saba-25-02.ts", "type": "primary" }, { "label": "OpenRouter Saba 2502 endpoint metadata", "url": "https://openrouter.ai/api/v1/models/mistralai/mistral-saba-2502/endpoints", "type": "primary" }, { "label": "Magistral announcement", "url": "https://mistral.ai/news/magistral/", "type": "primary" }, { "label": "PDF Inputs", "url": "https://openrouter.ai/docs/guides/overview/multimodal/pdfs", "type": "primary" }, { "label": "Mistral Docs model-type definitions", "url": "https://docs.mistral.ai/_next/static/chunks/7205-002ea58a31726078.js", "type": "primary" }, { "label": "Groq launch partner identifies Saba as based on Mistral Small 24B", "url": "https://www.linkedin.com/posts/jacobloewenstein_pumped-to-share-a-personal-milestone-2-months-activity-7300923497550086144-tmXj", "type": "primary" }, { "label": "Pinned official Mistral Small 3.0 total and active parameter fields", "url": "https://raw.githubusercontent.com/mistralai/platform-docs-public/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/schema/models/models/mistral-small-3-0-25-01.ts", "type": "primary" }, { "label": "Mistral Small 3.0 model documentation and parameter units", "url": "https://docs.mistral.ai/models/mistral-small-3-0-25-01", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/mistral-saba.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Mistral publishes a nominal 24B total. Nominal 24B active is inferred from its documented Small fine-tune lineage, the Groq launch partner's identification of Small 24B as the base, and Mistral's separate 24B active specification for Small 3/2501; Saba has no direct active-count disclosure. Groq documents a dense Transformer. Mistral gives 32k context; 32,768 is the OpenRouter Mistral-route limit, with a separate 26,214 output cap. Native input/output are text; PDF handling is gateway preprocessing. Mistral retired Saba 25.02 on 2025-09-30; provider metadata does not establish current endpoint operation. Saba-specific reasoning-mode status remains unverified.", "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-27T15:38:43.394120Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/341.json" }, { "id": "alibaba/qwen-omni-turbo", "slug": "alibaba--qwen-omni-turbo", "name": "Qwen-Omni Turbo", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Hosted Qwen multimodal model accepting text, images, audio, and video and generating streamed text and speech.", "release_date": "2025-02-14", "date_precision": "day", "open_weights": true, "license": null, "parameters_b": 7, "active_parameters_b": 7, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text", "image", "audio", "video" ], "output": [ "text", "audio" ] }, "reasoning": false, "tool_call": false, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen-omni-turbo.toml", "source_type": "catalog", "sources": [ { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen-omni-turbo.toml", "type": "catalog" }, { "label": "qwen-omni-turbo model information (Japanese documentation)", "url": "https://www.alibabacloud.com/help/ja/model-studio/qwen-omni-turbo", "type": "primary" }, { "label": "Model lifecycle and updates", "url": "https://help.aliyun.com/en/model-studio/newly-released-models", "type": "primary" }, { "label": "Qwen2.5-Omni official repository README", "url": "https://github.com/QwenLM/Qwen2.5-Omni/blob/main/README.md", "type": "primary" }, { "label": "Alibaba Cloud Model Studio model pricing", "url": "https://help.aliyun.com/en/model-studio/model-pricing", "type": "primary" }, { "label": "Qwen2.5 Omni launch announcement, March 26, 2025 revision", "url": "https://github.com/QwenLM/qwenlm.github.io/blob/989b611991a7196bd868d5a88c6caca526ca50a5/content/blog/qwen2.5-omni/index.md", "type": "primary" }, { "label": "Qwen2.5-Omni launch-day README, March 26, 2025", "url": "https://github.com/QwenLM/Qwen2.5-Omni/blob/006e089ff593a88d4b116c3454f7e2255d71a242/README.md", "type": "primary" }, { "label": "Qwen2.5-Omni Technical Report, original version", "url": "https://arxiv.org/html/2503.20215v1", "type": "primary" }, { "label": "Qwen2.5-Omni-7B official configuration", "url": "https://huggingface.co/Qwen/Qwen2.5-Omni-7B/blob/main/config.json", "type": "primary" }, { "label": "Alibaba Cloud Omni-modal overview, legacy model list", "url": "https://help.aliyun.com/en/model-studio/omni/", "type": "primary" }, { "label": "Qwen2.5-7B official language-model configuration", "url": "https://huggingface.co/Qwen/Qwen2.5-7B/blob/main/config.json", "type": "primary" }, { "label": "Qwen3 Technical Report, Qwen2.5-7B nominal activated count", "url": "https://arxiv.org/html/2505.09388v1", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=qwen-omni-turbo&limit=100", "label": "Official Qwen inventory search for the exact hosted model", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=Qwen2.5-Omni&limit=1000", "label": "Official Qwen Qwen2.5-Omni weight inventory", "type": "primary" }, { "url": "https://huggingface.co/Qwen/Qwen2.5-Omni-7B", "label": "Public Qwen2.5-Omni-7B checkpoint", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Nominal 7B total and active cover the Thinker language decoder, inferred from the Qwen2.5-Omni-7B launch, Turbo API example, matching Qwen2.5-7B text configuration and the base’s published activated count. Full multimodal total and active counts remain unspecified. Dense Thinker-Talker architecture is inherited from the named model; the alias maps to the March 26, 2025 snapshot. Open weights through the developer's explicit mapping to downloadable Qwen2.5-Omni-7B. API voice choices and serving features may differ from local inference.", "field_status": { "license": "undisclosed" }, "audited_at": "2026-10-28T08:13:49.514266+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/015.json", "parameters_scope": "thinker_decoder", "parameter_components_b": { "thinker_decoder": 7 }, "active_parameters_scope": "thinker_decoder" }, { "id": "perplexity/sonar-deep-research", "slug": "perplexity--sonar-deep-research", "name": "Sonar Deep Research", "developer": "Perplexity", "developer_id": "perplexity", "family": "sonar", "description": "Research model that performs repeated web searches and produces reports with citations.", "release_date": "2025-02-14", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://x.com/AravSrinivas/status/1894471526449385687", "source_type": "primary", "sources": [ { "label": "Introducing Perplexity Deep Research, official localized article", "url": "https://www.perplexity.ai/fr/hub/blog/introducing-perplexity-deep-research", "type": "primary" }, { "label": "Perplexity official blog archive", "url": "https://hub-prod.perplexity.ai/", "type": "primary" }, { "label": "Perplexity CEO announces Deep Research through the Sonar API", "url": "https://x.com/AravSrinivas/status/1894471526449385687", "type": "primary" }, { "label": "Perplexity API changelog", "url": "https://docs.perplexity.ai/docs/resources/changelog", "type": "primary" }, { "label": "Sonar to Agent API migration documentation", "url": "https://docs.perplexity.ai/docs/agent-api/migrate-from-sonar/overview", "type": "primary" }, { "label": "OpenRouter Sonar Deep Research endpoint specification", "url": "https://openrouter.ai/perplexity/sonar-deep-research", "type": "provider" }, { "label": "Perplexity API terms, January 23, 2026", "url": "https://www.perplexity.ai/hr/hub/legal/perplexity-api-terms-of-service", "type": "primary" }, { "label": "DRACO first-party technical report, February 12, 2026", "url": "https://r2cdn.perplexity.ai/pplx-draco.pdf", "type": "primary" }, { "label": "Retained Sonar Deep Research model documentation", "url": "https://docs.perplexity.ai/docs/legacy/sonar/models/sonar-deep-research.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/perplexity/sonar-deep-research.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Deep Research launched in the web app on February 14, 2025 and expanded to the Sonar API on February 25; OpenRouter listed it March 7. This establishes product lineage, not identical checkpoints. Total/active parameters and exact neural architecture are undisclosed; basic Sonar, Reasoning Pro and later Opus web-agent specifications do not establish this model’s counts. The developer specifies nominal 128K context and the provider 128,000 tokens. Tool use means built-in searches; the provider does not support custom function calls. Sonar support ended September 27, 2026: synchronous/streaming calls migrate to Agent API requests and legacy asynchronous requests are unsupported.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T17:38:53.879577+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/467.json" }, { "id": "google/gemini-2.0-flash-lite", "slug": "google--gemini-2.0-flash-lite", "name": "Gemini 2.0 Flash-Lite", "developer": "Google", "developer_id": "google", "family": "gemini-flash-lite", "description": "Cost-efficient Gemini 2.0 model for high-volume, low-latency multimodal input and text output.", "release_date": "2025-02-05", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "audio", "video", "pdf" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://developers.googleblog.com/en/gemini-2-family-expands/", "source_type": "primary", "sources": [ { "label": "Gemini 2.0 Flash-Lite model card", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Lite-Model-Card.pdf", "type": "primary" }, { "label": "Gemini 2.0 Flash-Lite API specifications", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.0-flash-lite", "type": "primary" }, { "label": "Gemini API service terms", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Gemini 2.X technical report", "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_v2_5_report.pdf", "type": "primary" }, { "label": "Gemini 2.0 Flash-Lite Cloud specifications", "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/2-0-flash-lite?hl=es", "type": "primary" }, { "label": "Gemini 2.0 Flash-Lite public preview", "url": "https://developers.googleblog.com/en/gemini-2-family-expands/", "type": "announcement" }, { "label": "Gemini 2.0 Flash-Lite general availability", "url": "https://developers.googleblog.com/en/start-building-with-the-gemini-2-0-flash-family/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemini-2.0-flash-lite.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "First public preview: February 5, 2025; GA followed February 25. Sparse MoE Transformer; total and active parameter counts are undisclosed. Google documents shutdown on June 1, 2026; capabilities describe the last documented specification.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T12:47:00.915Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/191.json" }, { "id": "openai/o3-deep-research", "slug": "openai--o3-deep-research", "name": "o3-deep-research", "developer": "OpenAI", "developer_id": "openai", "family": "o3", "description": "OpenAI o3 variant trained for multi-step web research and data analysis, with text and image input and cited text reports.", "release_date": "2025-02-02", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 200000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-deep-research/", "source_type": "primary", "sources": [ { "label": "o3-deep-research model reference", "url": "https://developers.openai.com/api/docs/models/o3-deep-research", "type": "primary" }, { "label": "Introducing deep research", "url": "https://openai.com/index/introducing-deep-research/", "type": "primary" }, { "label": "Deep research in the API, webhooks, and web search with o3", "url": "https://community.openai.com/t/deep-research-in-the-api-webhooks-and-web-search-with-o3/1299919/1", "type": "primary" }, { "label": "Deep Research System Card, February 25, 2025", "url": "https://cdn.openai.com/deep-research-system-card.pdf", "type": "primary" }, { "label": "OpenAI API changelog", "url": "https://developers.openai.com/api/docs/changelog", "type": "primary" }, { "label": "Deep research API guide", "url": "https://developers.openai.com/api/docs/guides/deep-research", "type": "primary" }, { "label": "OpenAI API deprecations", "url": "https://developers.openai.com/api/docs/deprecations", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201", "type": "primary" }, { "label": "ChatGPT Agent System Card, July 17, 2025", "url": "https://cdn.openai.com/pdf/839e66fc-602c-48bf-81d3-b21eacc3459d/chatgpt_agent_system_card.pdf", "type": "primary" }, { "label": "OpenAI o3 and o4-mini System Card, April 16, 2025", "url": "https://cdn.openai.com/pdf/2221c875-02dc-4789-800b-e7758f3722c1/o3-and-o4-mini-system-card.pdf", "type": "primary" }, { "label": "OpenAI GPT-5 System Card", "url": "https://arxiv.org/abs/2601.03267", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "File inputs API guide", "url": "https://developers.openai.com/api/docs/guides/file-inputs", "type": "primary" }, { "label": "Introduction to the Deep Research API", "url": "https://developers.openai.com/cookbook/examples/deep_research_api/introduction_to_deep_research_api", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/o3-deep-research.toml", "type": "catalog" }, { "label": "o3 model reference, Markdown", "url": "https://developers.openai.com/api/docs/models/o3.md", "type": "primary" }, { "label": "DeepSearchQA: Bridging the Comprehensiveness Gap for Deep Research Agents, December 11, 2025", "url": "https://storage.googleapis.com/deepmind-media/DeepSearchQA/DeepSearchQA_benchmark_paper.pdf", "type": "primary" }, { "label": "Tongyi DeepResearch Technical Report, November 15, 2025", "url": "https://raw.githubusercontent.com/Alibaba-NLP/DeepResearch/main/Tech_Report.pdf", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The release date is the February 2, 2025 ChatGPT debut of the o3 research variant. OpenAI’s later API announcement identifies o3-deep-research as the same ChatGPT research offering. The June 26 API snapshot is a later identifier; checkpoint identity across those dates is not established. The current API changelog lists June 24, while the contemporaneous API announcement is dated June 26. The 200,000-token context and text/image input specifications describe the last documented API snapshot, o3-deep-research-2025-06-26, not necessarily the original ChatGPT deployment. OpenAI shut down the alias and dated API snapshot on July 23, 2026. Tool support covers web search, file search, code interpreter and MCP search/fetch. Custom function calling and arbitrary MCP tool interfaces are unsupported. PDF processing and Python-generated figures are service/tool capabilities, not additional native modalities. Transformer is a broad inference from exact-model prompt-caching support and OpenAI’s applicable engineering explanation. Dense/MoE structure and nominal total/active parameter counts remain undisclosed. The documented early-o3 lineage cannot yield a nominal size because no supported o3 count is published. The separate o3-mini chain-of-thought summarizer is excluded from this model’s specifications.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T17:22:36.081674Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/454.json" }, { "id": "perplexity/sonar-reasoning-pro", "slug": "perplexity--sonar-reasoning-pro", "name": "Sonar Reasoning Pro", "developer": "Perplexity", "developer_id": "perplexity", "family": "sonar", "description": "Legacy Sonar reasoning model for multi-step questions, with web retrieval and cited text answers.", "release_date": "2025-02", "date_precision": "month", "open_weights": false, "license": "Proprietary", "parameters_b": 671, "active_parameters_b": 37, "parameters_source": "primary_source", "architecture": "moe", "context_window": 128000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://web.archive.org/web/20250205102225id_/https://docs.perplexity.ai/guides/model-cards", "source_type": "primary", "sources": [ { "label": "Sonar Reasoning Pro — legacy model reference", "url": "https://docs.perplexity.ai/docs/legacy/sonar/models/sonar-reasoning-pro.md", "type": "primary" }, { "label": "Sonar model range — legacy reference", "url": "https://docs.perplexity.ai/docs/legacy/sonar/models.md", "type": "primary" }, { "label": "Improved Sonar Models: Industry Leading Performance at Lower Costs", "url": "https://hub-prod.perplexity.ai/hub/blog/new-sonar-search-modes-outperform-openai-in-cost-and-performance", "type": "primary" }, { "label": "API changelog", "url": "https://docs.perplexity.ai/docs/resources/changelog.md", "type": "primary" }, { "label": "Migrate from Sonar — overview", "url": "https://docs.perplexity.ai/docs/agent-api/migrate-from-sonar/overview.md", "type": "primary" }, { "label": "Perplexity API Terms of Service", "url": "https://www.perplexity.ai/en-GB/hub/legal/perplexity-api-terms-of-service", "type": "primary" }, { "label": "Media & Attachments — legacy Sonar API", "url": "https://docs.perplexity.ai/docs/legacy/sonar/media.md", "type": "primary" }, { "label": "Efficient and Portable Mixture-of-Experts Communication", "url": "https://research.perplexity.ai/articles/efficient-and-portable-mixture-of-experts-communication", "type": "primary" }, { "label": "Lower Latency and Higher Throughput with Multi-node DeepSeek Deployment", "url": "https://research.perplexity.ai/articles/lower-latency-and-higher-throughput-with-multi-node-deepseek-deployment", "type": "primary" }, { "label": "RL Training For Math Reasoning", "url": "https://research.perplexity.ai/articles/rl-training-for-math-reasoning", "type": "primary" }, { "label": "Accelerating Sonar Through Speculation", "url": "https://research.perplexity.ai/articles/accelerating-sonar-through-speculation", "type": "primary" }, { "label": "Sonar Now Supports Image Upload", "url": "https://community.perplexity.ai/t/sonar-now-supports-image-upload/4735.json", "type": "primary" }, { "label": "Perplexity Sonar Dominates New Search Arena Evaluation", "url": "https://hub-prod.perplexity.ai/hub/blog/perplexity-sonar-dominates-new-search-arena-evolution", "type": "primary" }, { "label": "Perplexity Models — April 5, 2025 archived developer documentation", "url": "https://web.archive.org/web/20250405124746id_/https://docs.perplexity.ai/guides/model-cards", "type": "primary" }, { "label": "DeepSeek-R1 official model card — nominal total and activated parameters", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1", "type": "primary" }, { "label": "Sonar Reasoning Pro — August 1, 2025 archived exact-model documentation", "url": "https://web.archive.org/web/20250801011459id_/https://docs.perplexity.ai/getting-started/models/models/sonar-reasoning-pro", "type": "primary" }, { "label": "Perplexity original Sonar Reasoning launch (non-Pro)", "url": "https://x.com/perplexity_ai/status/1884409454675759211", "type": "primary" }, { "label": "Perplexity supported models — January 30, 2025 archive", "url": "https://web.archive.org/web/20250130094226id_/https://docs.perplexity.ai/guides/model-cards", "type": "primary" }, { "label": "Perplexity supported models — February 5, 2025 archive", "url": "https://web.archive.org/web/20250205102225id_/https://docs.perplexity.ai/guides/model-cards", "type": "primary" }, { "label": "Perplexity API changelog — February 4, 2025 archive", "url": "https://web.archive.org/web/20250204215932id_/https://docs.perplexity.ai/changelog/changelog", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/perplexity/sonar-reasoning-pro.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Release month: February 2025. The original publisher model table has no Reasoning Pro on January 30 and adds it by February 5; a January 31 native discussion reproduces the pre-Pro list. NanoGPT announces the new Pro release on February 3, and a February 4 native API discussion reports its new deployment. This combined chronology supports the month; the exact first day is not established. The January 28/29 sibling launch, February 5 archive capture and March 7 OpenRouter listing are not used as the release day. The 128K context and image-input restrictions describe the legacy service. Sonar Chat Completions support ended September 27, 2026; compatibility requests are gradually reformulated through Agent API, so successful requests do not verify the original backend. Fresh October 27 parameter re-audit recovered an April 5, 2025 archive of Perplexity’s own model overview explicitly identifying sonar-reasoning-pro as powered by DeepSeek R1. DeepSeek’s official card and Perplexity’s deployment research establish the nominal R1 specifications of 671B total, 37B activated per token and MoE architecture. Retain those as a documented lineage inference for the historical language backbone, together with the proposed component and active scopes; no reviewed primary source discloses structural changes defeating the inference. Exact fine-tuned checkpoint, later backend revisions and auxiliary search, routing, image-processing or speculative-decoding components remain unspecified and are not counted. Missing exact deployed tensor sizes do not invalidate the nominal backbone specification. Tool invocation remains unverified; built-in retrieval alone does not establish model-selected tools, and Sonar Pro’s Pro Search capabilities are not inherited.", "field_status": { "tool_call": "unverified" }, "audited_at": "2026-10-28T08:14:51.132210Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/469.json", "parameters_scope": "language_backbone", "active_parameters_scope": "language_backbone", "parameter_components_b": { "language_backbone": 671 } }, { "id": "openai/o3-mini", "slug": "openai--o3-mini", "name": "o3-mini", "developer": "OpenAI", "developer_id": "openai", "family": "o3", "description": "Text-only reasoning model with adjustable reasoning effort, function calling, and structured outputs.", "release_date": "2025-01-31", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 200000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/openai-o3-mini/", "source_type": "primary", "sources": [ { "label": "o3-mini release", "url": "https://openai.com/index/openai-o3-mini/", "type": "announcement" }, { "label": "o3-mini API reference", "url": "https://developers.openai.com/api/docs/models/o3-mini.md", "type": "documentation" }, { "label": "o3-mini system card", "url": "https://cdn.openai.com/o3-mini-system-card-feb10.pdf", "type": "paper" }, { "label": "GPT-OSS model card comparisons", "url": "https://cdn.openai.com/pdf/419b6906-9da6-406c-a19d-1bb078ac7637/oai_gpt-oss_model_card.pdf", "type": "paper" }, { "label": "Deliberative alignment", "url": "https://arxiv.org/pdf/2412.16339", "type": "paper" }, { "label": "Persona Features Control Emergent Misalignment", "url": "https://arxiv.org/pdf/2506.19823", "type": "paper" }, { "label": "Prompt Caching 201: pre-GPT-5.6 engineering", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201.md", "type": "documentation" }, { "label": "API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "documentation" }, { "label": "API changelog", "url": "https://developers.openai.com/api/docs/changelog.md", "type": "documentation" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "license" }, { "label": "Qwen3 original comparison: o3-mini counts unreported", "url": "https://arxiv.org/html/2505.09388v1", "type": "paper" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/o3-mini.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Released January 31, 2025 after a December 2024 preview. The o3-mini-2025-01-31 snapshot and alias were shut down on October 23, 2026. Last documented limits: 200,000 context tokens and 100,000 output tokens. Transformer is a broad inference from exact-model caching support and applicable OpenAI engineering documentation; detailed topology is undisclosed. Nominal total and active parameter counts remain undisclosed; Qwen3 Table 13 leaves both o3-mini count cells blank. Low, medium and high are reasoning-effort settings.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T17:22:06.109076Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/455.json" }, { "id": "mistral/mistral-small-24b-instruct-2501", "slug": "mistral--mistral-small-24b-instruct-2501", "name": "Mistral Small 3", "developer": "Mistral", "developer_id": "mistral", "family": "mistral-small", "description": "Instruction-tuned 24B text model with a 32,768-token context window, native function calling, and Apache-2.0 weights.", "release_date": "2025-01-30", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 24, "active_parameters_b": 24, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://mistral.ai/news/mistral-small-3/", "source_type": "primary", "sources": [ { "label": "Mistral Small 3 release announcement", "url": "https://mistral.ai/news/mistral-small-3/", "type": "primary" }, { "label": "Mistral-Small-24B-Instruct-2501 model card", "url": "https://huggingface.co/mistralai/Mistral-Small-24B-Instruct-2501", "type": "weights" }, { "label": "Mistral Small 2501 Transformers configuration", "url": "https://huggingface.co/mistralai/Mistral-Small-24B-Instruct-2501/raw/main/config.json", "type": "weights" }, { "label": "Mistral Small 2501 native parameters", "url": "https://huggingface.co/mistralai/Mistral-Small-24B-Instruct-2501/blob/main/params.json", "type": "weights" }, { "label": "Mistral Small 2501 checkpoint tensor index", "url": "https://huggingface.co/mistralai/Mistral-Small-24B-Instruct-2501/raw/main/model.safetensors.index.json", "type": "weights" }, { "label": "Pinned official Mistral Small 3.0 v25.01 documentation data", "url": "https://github.com/mistralai/platform-docs-public/blob/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/schema/models/models/mistral-small-3-0-25-01.ts", "type": "primary" }, { "label": "Pinned official weights-table unit labels", "url": "https://github.com/mistralai/platform-docs-public/blob/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/app/%5Blocale%5D/(docs)/models/model-cards/%5Bslug%5D/components/weights-table.tsx", "type": "primary" }, { "label": "Mistral Small 3.0 v25.01 model documentation", "url": "https://docs.mistral.ai/models/mistral-small-3-0-25-01", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/mistral-small-24b-instruct-2501.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "January 2025 Mistral Small 3 Instruct checkpoint. Total and active parameter counts are nominal disclosures of 24B each in separate official documentation fields. Native max_seq_len is 32768. Native dedicated reasoning support remains unverified; general reasoning skill is not treated as an explicit reasoning-mode flag.", "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-26T23:29:13.010465Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/342.json" }, { "id": "alibaba/qwen2-5-vl-72b-instruct", "slug": "alibaba--qwen2-5-vl-72b-instruct", "name": "Qwen2.5-VL 72B Instruct", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Qwen vision-language model for visual reasoning, documents, and agent tasks", "release_date": "2025-01-26", "date_precision": "day", "open_weights": true, "license": "qwen", "parameters_b": 73.410777344, "active_parameters_b": 72, "parameters_source": "primary_source", "architecture": "dense", "context_window": 128000, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen2-5-vl-72b-instruct.toml", "source_type": "catalog", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen2.5-VL-72B-Instruct", "type": "weights" }, { "label": "Qwen2.5-VL release announcement", "url": "https://qwenlm.github.io/blog/qwen2.5-vl/", "type": "primary" }, { "label": "Exact checkpoint model configuration", "url": "https://huggingface.co/Qwen/Qwen2.5-VL-72B-Instruct/blob/89c86200743eec961a297729e7990e8f2ddbc4c5/config.json", "type": "primary" }, { "label": "Qwen License Agreement", "url": "https://huggingface.co/Qwen/Qwen2.5-VL-72B-Instruct/blob/89c86200743eec961a297729e7990e8f2ddbc4c5/LICENSE", "type": "primary" }, { "label": "Checkpoint index and referenced Safetensors headers", "url": "https://huggingface.co/Qwen/Qwen2.5-VL-72B-Instruct/blob/89c86200743eec961a297729e7990e8f2ddbc4c5/model.safetensors.index.json", "type": "weights" }, { "label": "Exact checkpoint tokenizer configuration", "url": "https://huggingface.co/Qwen/Qwen2.5-VL-72B-Instruct/blob/89c86200743eec961a297729e7990e8f2ddbc4c5/tokenizer_config.json", "type": "primary" }, { "label": "Qwen2.5-VL Technical Report", "url": "https://arxiv.org/html/2502.13923v1", "type": "primary" }, { "label": "Qwen2.5-VL configuration implementation", "url": "https://raw.githubusercontent.com/huggingface/transformers/v4.51.3/src/transformers/models/qwen2_5_vl/configuration_qwen2_5_vl.py", "type": "primary" }, { "label": "Qwen2.5-VL rotary embedding implementation", "url": "https://raw.githubusercontent.com/huggingface/transformers/v4.51.3/src/transformers/models/qwen2_5_vl/modeling_qwen2_5_vl.py", "type": "primary" }, { "label": "Transformers tokenizer truncation and padding documentation", "url": "https://huggingface.co/docs/transformers/v4.51.3/en/main_classes/tokenizer", "type": "primary" }, { "label": "Live checkpoint repository and Safetensors metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2.5-VL-72B-Instruct", "type": "weights" }, { "label": "Qwen3 Technical Report: Qwen2.5-72B activated-parameter specification", "url": "https://arxiv.org/html/2505.09388v1", "type": "primary" }, { "label": "Qwen2.5-72B base language-model configuration", "url": "https://huggingface.co/Qwen/Qwen2.5-72B/blob/efba10c8e54e91e0d9570ab5f7b51a958474d4cb/config.json", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen2-5-vl-72b-instruct.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "72B is the nominal label; the complete BF16 checkpoint has 73.410777344B parameters, including 0.704573696B in the visual encoder and merger. Active 72B is nominal and scoped to the Qwen2.5 language backbone: Qwen publishes 72B activated for that base, which the VL model explicitly uses. It excludes the visual encoder and merger; a full multimodal active count is unspecified. Context 128000 follows the shipped model positional configuration; tokenizer length 131072 is a preprocessing setting. The card describes 32768 and optional YaRN, and the report documents 32768 training sequences, so configured context does not establish long-text quality. A dedicated reasoning-mode boolean remains unverified.", "parameter_components_b": { "language_model": 72.706203648, "vision_encoder_and_merger": 0.704573696 }, "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-27T07:44:30.041Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/020.json", "active_parameters_scope": "language_model" }, { "id": "perplexity/sonar", "slug": "perplexity--sonar", "name": "Sonar", "developer": "Perplexity", "developer_id": "perplexity", "family": "sonar", "description": "Perplexity search model that generates answers with web citations and accepts text, image and document inputs.", "release_date": "2025-01-21", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": 70, "active_parameters_b": 70, "parameters_source": "primary_source", "architecture": "transformer", "context_window": 128000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": false, "tool_call": null, "source_url": "https://x.com/perplexity_ai/status/1881779310840984043", "source_type": "primary", "sources": [ { "label": "Official Sonar launch", "url": "https://x.com/perplexity_ai/status/1881779310840984043", "type": "primary" }, { "label": "Official launch timestamp and text", "url": "https://cdn.syndication.twimg.com/tweet-result?id=1881779310840984043&lang=en&token=0", "type": "primary" }, { "label": "February 2025 Sonar backbone disclosure", "url": "https://www.perplexity.ai/en-GB/hub/blog/meet-new-sonar", "type": "primary" }, { "label": "Sonar speculative decoding research", "url": "https://research.perplexity.ai/articles/accelerating-sonar-through-speculation", "type": "primary" }, { "label": "Legacy Sonar specifications", "url": "https://docs.perplexity.ai/docs/legacy/sonar/models/sonar.md", "type": "primary" }, { "label": "Sonar API capability changelog", "url": "https://docs.perplexity.ai/docs/resources/changelog.md", "type": "primary" }, { "label": "Perplexity API terms", "url": "https://www.perplexity.ai/en-GB/hub/legal/perplexity-api-terms-of-service", "type": "primary" }, { "label": "Sonar migration status", "url": "https://docs.perplexity.ai/docs/agent-api/migrate-from-sonar/overview.md", "type": "primary" }, { "label": "Microsoft original Llama 3.3 active-parameter comparison", "url": "https://www.microsoft.com/en-us/research/wp-content/uploads/2025/01/WEF-2025_Leave-Behind_Phi-4.pdf", "type": "primary" }, { "label": "Alem original Llama 3.3 active-parameter comparison", "url": "https://arxiv.org/html/2606.08340v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/perplexity/sonar.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The February 2025 Sonar language backbone has nominal 70B total parameters through explicit Llama 3.3 70B Instruct further-training lineage. Approximately 70B active is inherited from Microsoft Research and Tessera et al. comparison plots for that base, not directly disclosed by Perplexity. Both counts exclude the separate 1B speculative draft and undisclosed visual/service components. A later legal page names Llama 3.1; later backend identity remains unresolved. Image/PDF are API inputs, and context/non-reasoning are retained legacy specifications. Sonar Chat Completions support ended September 27, 2026; requests are gradually reformulated into Agent API calls. Current model listing does not prove unchanged weights or tools.", "parameters_scope": "language_backbone", "parameter_components_b": { "language_backbone": 70 }, "field_status": { "tool_call": "unverified" }, "audited_at": "2026-10-27T17:34:12.382936+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/466.json", "active_parameters_scope": "language_backbone" }, { "id": "perplexity/sonar-pro", "slug": "perplexity--sonar-pro", "name": "Sonar Pro", "developer": "Perplexity", "developer_id": "perplexity", "family": "sonar", "description": "Search model for cited answers, with image and document input and optional Pro Search reasoning and web tools.", "release_date": "2025-01-21", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 200000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://docs.perplexity.ai/docs/legacy/sonar/models/sonar-pro.md", "source_type": "primary", "sources": [ { "label": "Sonar Pro — retained Sonar model documentation", "url": "https://docs.perplexity.ai/docs/legacy/sonar/models/sonar-pro.md", "type": "primary" }, { "label": "Perplexity blog index — Introducing the Sonar Pro API", "url": "https://hub-prod.perplexity.ai/", "type": "primary" }, { "label": "Perplexity launch announcement — official X post 1881779310840984043", "url": "https://cdn.syndication.twimg.com/tweet-result?id=1881779310840984043&lang=en&token=0", "type": "primary" }, { "label": "Introducing the Sonar Pro API — official Dutch translation", "url": "https://www.perplexity.ai/nl/hub/blog/introductie-van-de-sonar-pro-api", "type": "primary" }, { "label": "Perplexity API Terms of Service", "url": "https://www.perplexity.ai/en-GB/hub/legal/perplexity-api-terms-of-service", "type": "primary" }, { "label": "Sonar Media & Attachments", "url": "https://docs.perplexity.ai/docs/legacy/sonar/media.md", "type": "primary" }, { "label": "Pro Search for Sonar Pro — quickstart", "url": "https://docs.perplexity.ai/docs/legacy/sonar/pro-search/quickstart.md", "type": "primary" }, { "label": "Pro Search built-in tool capabilities", "url": "https://docs.perplexity.ai/docs/legacy/sonar/pro-search/tools.md", "type": "primary" }, { "label": "Perplexity API changelog", "url": "https://docs.perplexity.ai/docs/resources/changelog.md", "type": "primary" }, { "label": "Sonar to Agent API migration overview", "url": "https://docs.perplexity.ai/docs/agent-api/migrate-from-sonar/overview.md", "type": "primary" }, { "label": "Agent API vs Sonar benchmarks", "url": "https://docs.perplexity.ai/docs/agent-api/migrate-from-sonar/benchmarks.md", "type": "primary" }, { "label": "Meet new Sonar", "url": "https://www.perplexity.ai/en-GB/hub/blog/meet-new-sonar", "type": "primary" }, { "label": "Introducing the Sonar Pro API — original English URL", "url": "https://www.perplexity.ai/hub/blog/introducing-the-sonar-pro-api", "type": "primary" }, { "label": "Evaluating Web Retrieval-Assisted Large Language Models With and Without Whitelisting for Evidence-Based Neurology", "url": "https://www.jmir.org/2025/1/e79379/", "type": "secondary" }, { "label": "OPENXRD: a comprehensive benchmark framework for LLM/MLLM XRD question answering", "url": "https://pubs.rsc.org/en/content/articlehtml/2026/dd/d5dd00519a", "type": "secondary" }, { "label": "InfiAgent: An Infinite-Horizon Framework for General-Purpose Autonomous Agents", "url": "https://aclanthology.org/2026.findings-acl.1787.pdf", "type": "secondary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/perplexity/sonar-pro.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Sonar Pro was publicly released on January 21, 2025. The standard fast mode is non-reasoning; the reasoning and tool-use flags refer to optional streamed Pro Search, documented from November 2025, where the model selects built-in web search and URL fetching. Custom functions cannot be registered. PDF input reflects API document handling, while returned image/video URLs are retrieved assets. Total parameters remain unverified; active parameters and network architecture are undisclosed. Published 70B claims lack a traced Sonar Pro lineage source; the documented Llama 3.3 70B model is Sonar. Sonar Chat Completions support ended September 27, 2026; synchronous and streaming requests are gradually translated to Agent API calls. Specifications here describe the retained Sonar Pro documentation, not verified current inference on the original model.", "field_status": { "parameters_b": "unverified", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T17:39:28.874426Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/468.json" }, { "id": "deepseek/deepseek-r1", "slug": "deepseek--deepseek-r1", "name": "DeepSeek-R1", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-thinking", "description": "Original DeepSeek-R1 open-weight reasoning model, post-trained from DeepSeek-V3-Base for mathematics, coding and reasoning.", "release_date": "2025-01-20", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 682.636487424, "active_parameters_b": 37, "parameters_source": "primary_source", "architecture": "moe", "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": false, "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1", "source_type": "primary", "sources": [ { "label": "DeepSeek API release history", "url": "https://api-docs.deepseek.com/updates/", "type": "primary" }, { "label": "DeepSeek-R1 revised technical report", "url": "https://arxiv.org/html/2501.12948v2", "type": "primary" }, { "label": "Original R1 configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1/blob/56d4cbbb4d29f4355bab4b9a39ccb717a14ad5ad/config.json", "type": "primary" }, { "label": "DeepSeek architecture weight-sharing guide", "url": "https://github.com/deepseek-ai/DeepSeek-V3/blob/main/README_WEIGHTS.md", "type": "primary" }, { "label": "Original R1 checkpoint tensor index", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1/blob/56d4cbbb4d29f4355bab4b9a39ccb717a14ad5ad/model.safetensors.index.json", "type": "weights" }, { "label": "Original R1 release announcement", "url": "https://api-docs.deepseek.com/news/news250120/", "type": "primary" }, { "label": "Original R1 model.norm.weight comparison", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1/resolve/56d4cbbb4d29f4355bab4b9a39ccb717a14ad5ad/model-00160-of-000163.safetensors", "type": "weights" }, { "label": "Original R1 model.layers.61.shared_head.norm.weight comparison", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1/resolve/56d4cbbb4d29f4355bab4b9a39ccb717a14ad5ad/model-00163-of-000163.safetensors", "type": "weights" }, { "label": "Hugging Face", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepseek/deepseek-r1.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Original January 2025 R1 checkpoint, distinct from R1-0528 and the changing deepseek-reasoner API alias. Complete unique logical count is 682.636487424B: 671.0264192B main plus 11.610068224B MTP. FP8 scales and documented shared embedding/output-head copies are excluded. The MTP final norm differs from the main norm and is retained. Nominal 37B active refers to main inference. Advertised context is 128K; configured positions are 163840. Native tool calling is unsupported for this original version.", "active_parameters_scope": "main", "parameter_components_b": { "main": 671.0264192, "mtp_unique": 11.610068224 }, "active_parameters_by_phase_b": { "main": 37 }, "field_status": {}, "audited_at": "2026-10-26T19:15:59.946749+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/168.json" }, { "id": "deepseek/deepseek-r1-distill-llama-70b", "slug": "deepseek--deepseek-r1-distill-llama-70b", "name": "DeepSeek-R1-Distill-Llama-70B", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-r1", "description": null, "release_date": "2025-01-20", "date_precision": "day", "open_weights": true, "license": "MIT; Llama 3.3 Community License", "parameters_b": 70, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-70B", "source_type": "primary", "sources": [ { "label": "Original dated release includes all six distilled checkpoint downloads", "url": "https://github.com/deepseek-ai/DeepSeek-R1/blob/23807ced51627276434655dd9f27725354818974/README.md", "type": "announcement" }, { "label": "Publisher model card and license terms", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-70B/blob/b1c0b44b4369b597ad119a196caf79a9c40e141e/README.md", "type": "model_card" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-70B/blob/b1c0b44b4369b597ad119a196caf79a9c40e141e/config.json", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T11:22:26.610Z", "last_seen": "2026-10-29T11:22:26.610Z", "upstream_present": false, "availability": "public", "notes": "DeepSeek company release distilled through fine-tuning an external Qwen/Llama base using R1-generated examples. The original January 20 release explicitly includes this checkpoint. Parameter count is nominal; context follows the exact checkpoint configuration." }, { "id": "deepseek/deepseek-r1-distill-llama-8b", "slug": "deepseek--deepseek-r1-distill-llama-8b", "name": "DeepSeek-R1-Distill-Llama-8B", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-r1", "description": null, "release_date": "2025-01-20", "date_precision": "day", "open_weights": true, "license": "MIT; Llama 3.1 Community License", "parameters_b": 8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-8B", "source_type": "primary", "sources": [ { "label": "Original dated release includes all six distilled checkpoint downloads", "url": "https://github.com/deepseek-ai/DeepSeek-R1/blob/23807ced51627276434655dd9f27725354818974/README.md", "type": "announcement" }, { "label": "Publisher model card and license terms", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-8B/blob/6a6f4aa4197940add57724a7707d069478df56b1/README.md", "type": "model_card" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-8B/blob/6a6f4aa4197940add57724a7707d069478df56b1/config.json", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T11:22:26.610Z", "last_seen": "2026-10-29T11:22:26.610Z", "upstream_present": false, "availability": "public", "notes": "DeepSeek company release distilled through fine-tuning an external Qwen/Llama base using R1-generated examples. The original January 20 release explicitly includes this checkpoint. Parameter count is nominal; context follows the exact checkpoint configuration." }, { "id": "deepseek/deepseek-r1-distill-qwen-1.5b", "slug": "deepseek--deepseek-r1-distill-qwen-1.5b", "name": "DeepSeek-R1-Distill-Qwen-1.5B", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-r1", "description": null, "release_date": "2025-01-20", "date_precision": "day", "open_weights": true, "license": "MIT; Apache-2.0 base", "parameters_b": 1.5, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B", "source_type": "primary", "sources": [ { "label": "Original dated release includes all six distilled checkpoint downloads", "url": "https://github.com/deepseek-ai/DeepSeek-R1/blob/23807ced51627276434655dd9f27725354818974/README.md", "type": "announcement" }, { "label": "Publisher model card and license terms", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B/blob/ad9f0ae0864d7fbcd1cd905e3c6c5b069cc8b562/README.md", "type": "model_card" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B/blob/ad9f0ae0864d7fbcd1cd905e3c6c5b069cc8b562/config.json", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T11:22:26.610Z", "last_seen": "2026-10-29T11:22:26.610Z", "upstream_present": false, "availability": "public", "notes": "DeepSeek company release distilled through fine-tuning an external Qwen/Llama base using R1-generated examples. The original January 20 release explicitly includes this checkpoint. Parameter count is nominal; context follows the exact checkpoint configuration." }, { "id": "deepseek/deepseek-r1-distill-qwen-14b", "slug": "deepseek--deepseek-r1-distill-qwen-14b", "name": "DeepSeek-R1-Distill-Qwen-14B", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-r1", "description": null, "release_date": "2025-01-20", "date_precision": "day", "open_weights": true, "license": "MIT; Apache-2.0 base", "parameters_b": 14, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B", "source_type": "primary", "sources": [ { "label": "Original dated release includes all six distilled checkpoint downloads", "url": "https://github.com/deepseek-ai/DeepSeek-R1/blob/23807ced51627276434655dd9f27725354818974/README.md", "type": "announcement" }, { "label": "Publisher model card and license terms", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B/blob/1df8507178afcc1bef68cd8c393f61a886323761/README.md", "type": "model_card" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B/blob/1df8507178afcc1bef68cd8c393f61a886323761/config.json", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T11:22:26.610Z", "last_seen": "2026-10-29T11:22:26.610Z", "upstream_present": false, "availability": "public", "notes": "DeepSeek company release distilled through fine-tuning an external Qwen/Llama base using R1-generated examples. The original January 20 release explicitly includes this checkpoint. Parameter count is nominal; context follows the exact checkpoint configuration." }, { "id": "deepseek/deepseek-r1-distill-qwen-32b", "slug": "deepseek--deepseek-r1-distill-qwen-32b", "name": "DeepSeek-R1-Distill-Qwen-32B", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-thinking", "description": "Dense text reasoning model fine-tuned from Qwen2.5-32B using about 800,000 samples curated with DeepSeek-R1.", "release_date": "2025-01-20", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 32.763876352, "active_parameters_b": 32, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B", "source_type": "primary", "sources": [ { "label": "Official distilled-model card", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B/blob/711ad2ea6aa40cfca18895e8aca02ab92df1a746/README.md", "type": "primary" }, { "label": "Dated DeepSeek release announcement", "url": "https://deepseek.com/news/deepseek-r1/", "type": "primary" }, { "label": "Exact checkpoint configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B/raw/711ad2ea6aa40cfca18895e8aca02ab92df1a746/config.json", "type": "primary" }, { "label": "Complete checkpoint tensor index", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B/raw/711ad2ea6aa40cfca18895e8aca02ab92df1a746/model.safetensors.index.json", "type": "weights" }, { "label": "Released checkpoint MIT license", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B/raw/711ad2ea6aa40cfca18895e8aca02ab92df1a746/LICENSE", "type": "primary" }, { "label": "Underlying Qwen2.5-32B Apache-2.0 license", "url": "https://huggingface.co/Qwen/Qwen2.5-32B/blob/main/LICENSE", "type": "primary" }, { "label": "Exact tokenizer and chat template", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B/raw/711ad2ea6aa40cfca18895e8aca02ab92df1a746/tokenizer_config.json", "type": "primary" }, { "label": "Original DeepSeek-R1 technical report", "url": "https://arxiv.org/html/2501.12948v1#S2.SS4", "type": "primary" }, { "label": "Revised DeepSeek-R1 technical report", "url": "https://arxiv.org/html/2501.12948v2", "type": "primary" }, { "label": "Qwen3 Technical Report: base and distill activated-parameter comparison tables", "url": "https://arxiv.org/html/2505.09388v1", "type": "primary" }, { "label": "Official Qwen2.5-32B Base configuration", "url": "https://huggingface.co/Qwen/Qwen2.5-32B/raw/main/config.json", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepseek/deepseek-r1-distill-qwen-32b.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "January 2025 R1 distill of Qwen2.5-32B. Complete BF16 checkpoint: 32,763,876,352 learned parameters; untied input/output embeddings are both included. Nominal 32B active is inferred from the documented Qwen2.5-32B Base fine-tuning lineage and matching configuration, using Qwen’s activated-parameter convention. MIT release derives from an Apache-2.0 base. Context 131,072 is the configured position limit; the official example/evaluation uses 32,768. Tool-call formatting exists, but exact-variant tool-use capability remains unverified.", "field_status": { "tool_call": "unverified" }, "audited_at": "2026-10-27T11:00:33.541819+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/171.json" }, { "id": "deepseek/deepseek-r1-distill-qwen-7b", "slug": "deepseek--deepseek-r1-distill-qwen-7b", "name": "DeepSeek-R1-Distill-Qwen-7B", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-r1", "description": null, "release_date": "2025-01-20", "date_precision": "day", "open_weights": true, "license": "MIT; Apache-2.0 base", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B", "source_type": "primary", "sources": [ { "label": "Original dated release includes all six distilled checkpoint downloads", "url": "https://github.com/deepseek-ai/DeepSeek-R1/blob/23807ced51627276434655dd9f27725354818974/README.md", "type": "announcement" }, { "label": "Publisher model card and license terms", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B/blob/916b56a44061fd5cd7d6a8fb632557ed4f724f60/README.md", "type": "model_card" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B/blob/916b56a44061fd5cd7d6a8fb632557ed4f724f60/config.json", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T11:22:26.610Z", "last_seen": "2026-10-29T11:22:26.610Z", "upstream_present": false, "availability": "public", "notes": "DeepSeek company release distilled through fine-tuning an external Qwen/Llama base using R1-generated examples. The original January 20 release explicitly includes this checkpoint. Parameter count is nominal; context follows the exact checkpoint configuration." }, { "id": "deepseek/deepseek-r1-zero", "slug": "deepseek--deepseek-r1-zero", "name": "DeepSeek-R1-Zero", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-r1", "description": null, "release_date": "2025-01-20", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 671, "active_parameters_b": 37, "parameters_source": "primary_source", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://api-docs.deepseek.com/news/news250120/", "source_type": "primary", "sources": [ { "label": "Developer release announcement", "url": "https://api-docs.deepseek.com/news/news250120/", "type": "announcement" }, { "label": "Official model card and weights", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Zero/blob/72234287cbc67dbf474d911359ae32b61a2fdc7e/README.md", "type": "model_card" }, { "label": "Developer release repository", "url": "https://github.com/deepseek-ai/DeepSeek-R1", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-R1-Zero/blob/72234287cbc67dbf474d911359ae32b61a2fdc7e/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:07:38.377Z", "last_seen": "2026-10-29T10:07:38.377Z", "upstream_present": false, "availability": "public", "notes": "Distinct reinforcement-learning checkpoint without the preliminary supervised fine-tuning used by R1. The publisher’s nominal 671B parameters and 37B active refer to the main model. The configuration also includes one multi-token-prediction layer; a complete unique checkpoint total has not been independently counted. Supported context is the publisher’s nominal 128K.", "parameters_scope": "main", "active_parameters_scope": "main", "parameter_components_b": { "main": 671 } }, { "id": "deepseek/deepseek-reasoner", "slug": "deepseek--deepseek-reasoner", "name": "DeepSeek Reasoner", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-thinking", "description": "Legacy DeepSeek reasoning API alias, first released with R1 and last explicitly mapped to DeepSeek-V4-Flash preview in thinking mode; announced for retirement in July 2026.", "release_date": "2025-01-20", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 284, "active_parameters_b": 13, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://api-docs.deepseek.com/news/news250120/", "source_type": "primary", "sources": [ { "label": "Original reasoner API release", "url": "https://api-docs.deepseek.com/news/news250120/", "type": "primary" }, { "label": "DeepSeek alias history and function calling", "url": "https://api-docs.deepseek.com/updates/", "type": "primary" }, { "label": "V4 preview mapping and retirement announcement", "url": "https://api-docs.deepseek.com/news/news260424/", "type": "primary" }, { "label": "Last documented backing weights: V4-Flash preview", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash", "type": "weights" }, { "label": "DeepSeek-V4 technical report", "url": "https://arxiv.org/html/2606.19348v1", "type": "primary" }, { "label": "V4-Flash weight license", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash/blob/main/LICENSE", "type": "primary" }, { "label": "V3.2 thinking in tool use", "url": "https://api-docs.deepseek.com/news/news251201/", "type": "primary" }, { "label": "Current API model identifiers", "url": "https://api-docs.deepseek.com/quick_start/pricing/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepseek/deepseek-reasoner.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Specifications describe the last explicit April 24, 2026 V4-Flash preview mapping. The January 20, 2025 release date is the alias debut with R1, not the backing preview release. DeepSeek announced inaccessibility after July 24, 2026, 15:59 UTC; no authenticated endpoint test was performed. Later V4.1 compatibility mappings name other identifiers. Counts are developer nominal model parameters, not storage totals; MIT/open weights refer to the backing checkpoint.", "field_status": {}, "audited_at": "2026-10-26T19:02:44.030210+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/172.json" }, { "id": "minimax/MiniMax-Text-01", "slug": "minimax--minimax-text-01", "name": "MiniMax-Text-01", "developer": "MiniMax", "developer_id": "minimax", "family": "minimax", "description": "Text-only MoE model combining Lightning Attention and softmax attention, with native function calling and long-context inference.", "release_date": "2025-01-15", "date_precision": "day", "open_weights": true, "license": "MiniMax Model License", "parameters_b": 456, "active_parameters_b": 45.9, "parameters_source": "primary_source", "architecture": "moe", "context_window": 4000000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://www.minimax.cn/news/minimax-01-%E7%B3%BB%E5%88%97", "source_type": "primary", "sources": [ { "label": "MiniMax-01 release announcement", "url": "https://www.minimax.cn/news/minimax-01-%E7%B3%BB%E5%88%97", "type": "primary" }, { "label": "MiniMax-Text-01 official model card", "url": "https://huggingface.co/MiniMaxAI/MiniMax-Text-01/blob/a7351bf2bee0e1253919d349f1ad304e6dac13e9/README.md", "type": "primary" }, { "label": "MiniMax-01: Scaling Foundation Models with Lightning Attention", "url": "https://arxiv.org/html/2501.08313v1", "type": "primary" }, { "label": "MiniMax-Text-01 pinned configuration", "url": "https://huggingface.co/MiniMaxAI/MiniMax-Text-01/blob/a7351bf2bee0e1253919d349f1ad304e6dac13e9/config.json", "type": "weights" }, { "label": "MiniMax Model License", "url": "https://huggingface.co/MiniMaxAI/MiniMax-Text-01/blob/a7351bf2bee0e1253919d349f1ad304e6dac13e9/LICENSE-MODEL", "type": "weights" }, { "label": "MiniMax-Text-01 Function Call Guide", "url": "https://huggingface.co/MiniMaxAI/MiniMax-Text-01/blob/a7351bf2bee0e1253919d349f1ad304e6dac13e9/MiniMax-Text-01_Function_Call_Guide.md", "type": "weights" }, { "label": "MiniMax-Text-01 chat template", "url": "https://huggingface.co/MiniMaxAI/MiniMax-Text-01/blob/a7351bf2bee0e1253919d349f1ad304e6dac13e9/tokenizer_config.json", "type": "weights" }, { "label": "Official checkpoint repository metadata", "url": "https://huggingface.co/api/models/MiniMaxAI/MiniMax-Text-01", "type": "weights" }, { "label": "MiniMax-Text-01 checkpoint index", "url": "https://huggingface.co/MiniMaxAI/MiniMax-Text-01/blob/a7351bf2bee0e1253919d349f1ad304e6dac13e9/model.safetensors.index.json", "type": "weights" }, { "label": "MiniMax-M1 technical report", "url": "https://arxiv.org/html/2506.13585v1", "type": "primary" }, { "label": "Current MiniMax Model Invocation documentation", "url": "https://platform.minimax.io/docs/guides/text-generation", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/minimax/MiniMax-Text-01.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "456B total and 45.9B active are developer nominal figures, not an exact tensor census. Text-01 combines Lightning Attention with softmax attention and top-2-of-32 expert routing. Reported 4M inference is extrapolation after 1M training, demonstrated on retrieval; the current native API limit is not verified. Configuration/tokenizer maxima are higher but do not establish supported context. Dedicated thinking-mode status is unverified. Text-01 weights use the custom MiniMax Model License; the visual model is the separate VL-01.", "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-26T22:42:24.204980Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/310.json" }, { "id": "allenai/olmo-2-1124-13b-instruct", "slug": "allenai--olmo-2-1124-13b-instruct", "name": "OLMo-2-1124-13B-Instruct", "developer": "Ai2", "developer_id": "allenai", "family": "OLMo 2", "description": null, "release_date": "2025-01-03", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 13.7161984, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/allenai/OLMo-2-1124-13B-Instruct", "source_type": "primary", "sources": [ { "label": "Developer model card", "url": "https://huggingface.co/allenai/OLMo-2-1124-13B-Instruct", "type": "model_card" }, { "label": "Official release record", "url": "https://docs.allenai.org/release_notes/olmo-release-notes", "type": "announcement" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/allenai/OLMo-2-1124-13B-Instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "Ai2’s January 3, 2025 model-card update explicitly releases newly trained post-training models under the original repository names. The original November 26, 2024 weights are preserved separately with the -preview suffix. The 1124 name refers to the base generation, not this revised instruction checkpoint’s release day." }, { "id": "allenai/olmo-2-1124-7b-instruct", "slug": "allenai--olmo-2-1124-7b-instruct", "name": "OLMo-2-1124-7B-Instruct", "developer": "Ai2", "developer_id": "allenai", "family": "OLMo 2", "description": null, "release_date": "2025-01-03", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 7.298617344, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/allenai/OLMo-2-1124-7B-Instruct", "source_type": "primary", "sources": [ { "label": "Developer model card", "url": "https://huggingface.co/allenai/OLMo-2-1124-7B-Instruct", "type": "model_card" }, { "label": "Official release record", "url": "https://docs.allenai.org/release_notes/olmo-release-notes", "type": "announcement" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/allenai/OLMo-2-1124-7B-Instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "Ai2’s January 3, 2025 model-card update explicitly releases newly trained post-training models under the original repository names. The original November 26, 2024 weights are preserved separately with the -preview suffix. The 1124 name refers to the base generation, not this revised instruction checkpoint’s release day." }, { "id": "deepseek/deepseek-v3", "slug": "deepseek--deepseek-v3", "name": "DeepSeek-V3", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek", "description": "Open DeepSeek MoE chat model for coding, math, and general reasoning", "release_date": "2024-12-26", "date_precision": "day", "open_weights": true, "license": "DeepSeek Model License", "parameters_b": 682.636487424, "active_parameters_b": 36.7, "parameters_source": "primary_source", "architecture": "moe", "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3", "type": "weights" }, { "label": "Weight structure and active count", "url": "https://raw.githubusercontent.com/deepseek-ai/DeepSeek-V3/main/README_WEIGHTS.md", "type": "primary" }, { "label": "Original V3 configuration", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3/raw/e815299b0bcbac849fa540c768ef21845365c9eb/config.json", "type": "weights" }, { "label": "Technical report", "url": "https://arxiv.org/html/2412.19437v2", "type": "primary" }, { "label": "Dated official release changelog", "url": "https://api-docs.deepseek.com/updates/", "type": "primary" }, { "label": "Original model license", "url": "https://raw.githubusercontent.com/deepseek-ai/DeepSeek-V3/main/LICENSE-MODEL", "type": "primary" }, { "label": "Original chat and tool template", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3/raw/e815299b0bcbac849fa540c768ef21845365c9eb/tokenizer_config.json", "type": "weights" }, { "label": "Checkpoint tensor index", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3/resolve/e815299b0bcbac849fa540c768ef21845365c9eb/model.safetensors.index.json", "type": "weights" }, { "label": "Original V3 release announcement", "url": "https://api-docs.deepseek.com/news/news1226/", "type": "primary" }, { "label": "model.norm.weight bounded comparison", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3/resolve/e815299b0bcbac849fa540c768ef21845365c9eb/model-00160-of-000163.safetensors", "type": "weights" }, { "label": "model.layers.61.shared_head.norm.weight bounded comparison", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3/resolve/e815299b0bcbac849fa540c768ef21845365c9eb/model-00163-of-000163.safetensors", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepseek/deepseek-v3.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete count: 671.0264192B main + 11.610068224B unique MTP; shared embedding/head counted once and FP8 scales excluded. The MTP norm differs from the main norm. Nominal 36.7B active applies to the main model. Advertised context is 128K; config positions are 163840. The detailed weight guide includes embedding and output head in its main active count.", "parameter_components_b": { "main": 671.0264192, "mtp_unique": 11.610068224 }, "active_parameters_scope": "main", "field_status": {}, "audited_at": "2026-10-26T19:45:19.732401+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/173.json" }, { "id": "ibm/granite-3.1-1b-a400m-base", "slug": "ibm--granite-3.1-1b-a400m-base", "name": "granite-3.1-1b-a400m-base", "developer": "IBM", "developer_id": "ibm", "family": "Granite 3.1", "description": null, "release_date": "2024-12-18", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 1.33462528, "active_parameters_b": 0.4, "parameters_source": "checkpoint_metadata", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.ibm.com/new/announcements/ibm-granite-3-1-powerful-performance-long-context-and-more", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://www.ibm.com/new/announcements/ibm-granite-3-1-powerful-performance-long-context-and-more", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/ibm-granite/granite-3.1-1b-a400m-base", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/ibm-granite/granite-3.1-1b-a400m-base", "type": "weights" }, { "label": "Developer archive dates Granite 3.1 release December 18", "url": "https://www.ibm.com/new/announcements/ibm-granite-3-0-open-state-of-the-art-enterprise-models", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "ibm/granite-3.1-1b-a400m-instruct", "slug": "ibm--granite-3.1-1b-a400m-instruct", "name": "granite-3.1-1b-a400m-instruct", "developer": "IBM", "developer_id": "ibm", "family": "Granite 3.1", "description": null, "release_date": "2024-12-18", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 1.334628352, "active_parameters_b": 0.4, "parameters_source": "checkpoint_metadata", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.ibm.com/new/announcements/ibm-granite-3-1-powerful-performance-long-context-and-more", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://www.ibm.com/new/announcements/ibm-granite-3-1-powerful-performance-long-context-and-more", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/ibm-granite/granite-3.1-1b-a400m-instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/ibm-granite/granite-3.1-1b-a400m-instruct", "type": "weights" }, { "label": "Developer archive dates Granite 3.1 release December 18", "url": "https://www.ibm.com/new/announcements/ibm-granite-3-0-open-state-of-the-art-enterprise-models", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "ibm/granite-3.1-2b-base", "slug": "ibm--granite-3.1-2b-base", "name": "granite-3.1-2b-base", "developer": "IBM", "developer_id": "ibm", "family": "Granite 3.1", "description": null, "release_date": "2024-12-18", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 2.533525504, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.ibm.com/new/announcements/ibm-granite-3-1-powerful-performance-long-context-and-more", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://www.ibm.com/new/announcements/ibm-granite-3-1-powerful-performance-long-context-and-more", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/ibm-granite/granite-3.1-2b-base", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/ibm-granite/granite-3.1-2b-base", "type": "weights" }, { "label": "Developer archive dates Granite 3.1 release December 18", "url": "https://www.ibm.com/new/announcements/ibm-granite-3-0-open-state-of-the-art-enterprise-models", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "ibm/granite-3.1-2b-instruct", "slug": "ibm--granite-3.1-2b-instruct", "name": "granite-3.1-2b-instruct", "developer": "IBM", "developer_id": "ibm", "family": "Granite 3.1", "description": null, "release_date": "2024-12-18", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 2.533531648, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.ibm.com/new/announcements/ibm-granite-3-1-powerful-performance-long-context-and-more", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://www.ibm.com/new/announcements/ibm-granite-3-1-powerful-performance-long-context-and-more", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/ibm-granite/granite-3.1-2b-instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/ibm-granite/granite-3.1-2b-instruct", "type": "weights" }, { "label": "Developer archive dates Granite 3.1 release December 18", "url": "https://www.ibm.com/new/announcements/ibm-granite-3-0-open-state-of-the-art-enterprise-models", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "ibm/granite-3.1-3b-a800m-base", "slug": "ibm--granite-3.1-3b-a800m-base", "name": "granite-3.1-3b-a800m-base", "developer": "IBM", "developer_id": "ibm", "family": "Granite 3.1", "description": null, "release_date": "2024-12-18", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 3.298788864, "active_parameters_b": 0.8, "parameters_source": "checkpoint_metadata", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.ibm.com/new/announcements/ibm-granite-3-1-powerful-performance-long-context-and-more", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://www.ibm.com/new/announcements/ibm-granite-3-1-powerful-performance-long-context-and-more", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/ibm-granite/granite-3.1-3b-a800m-base", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/ibm-granite/granite-3.1-3b-a800m-base", "type": "weights" }, { "label": "Developer archive dates Granite 3.1 release December 18", "url": "https://www.ibm.com/new/announcements/ibm-granite-3-0-open-state-of-the-art-enterprise-models", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "ibm/granite-3.1-3b-a800m-instruct", "slug": "ibm--granite-3.1-3b-a800m-instruct", "name": "granite-3.1-3b-a800m-instruct", "developer": "IBM", "developer_id": "ibm", "family": "Granite 3.1", "description": null, "release_date": "2024-12-18", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 3.298793472, "active_parameters_b": 0.8, "parameters_source": "checkpoint_metadata", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.ibm.com/new/announcements/ibm-granite-3-1-powerful-performance-long-context-and-more", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://www.ibm.com/new/announcements/ibm-granite-3-1-powerful-performance-long-context-and-more", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/ibm-granite/granite-3.1-3b-a800m-instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/ibm-granite/granite-3.1-3b-a800m-instruct", "type": "weights" }, { "label": "Developer archive dates Granite 3.1 release December 18", "url": "https://www.ibm.com/new/announcements/ibm-granite-3-0-open-state-of-the-art-enterprise-models", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "ibm/granite-3.1-8b-base", "slug": "ibm--granite-3.1-8b-base", "name": "granite-3.1-8b-base", "developer": "IBM", "developer_id": "ibm", "family": "Granite 3.1", "description": null, "release_date": "2024-12-18", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 8.170835968, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.ibm.com/new/announcements/ibm-granite-3-1-powerful-performance-long-context-and-more", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://www.ibm.com/new/announcements/ibm-granite-3-1-powerful-performance-long-context-and-more", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/ibm-granite/granite-3.1-8b-base", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/ibm-granite/granite-3.1-8b-base", "type": "weights" }, { "label": "Developer archive dates Granite 3.1 release December 18", "url": "https://www.ibm.com/new/announcements/ibm-granite-3-0-open-state-of-the-art-enterprise-models", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "ibm/granite-3.1-8b-instruct", "slug": "ibm--granite-3.1-8b-instruct", "name": "granite-3.1-8b-instruct", "developer": "IBM", "developer_id": "ibm", "family": "Granite 3.1", "description": null, "release_date": "2024-12-18", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 8.170848256, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.ibm.com/new/announcements/ibm-granite-3-1-powerful-performance-long-context-and-more", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://www.ibm.com/new/announcements/ibm-granite-3-1-powerful-performance-long-context-and-more", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/ibm-granite/granite-3.1-8b-instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/ibm-granite/granite-3.1-8b-instruct", "type": "weights" }, { "label": "Developer archive dates Granite 3.1 release December 18", "url": "https://www.ibm.com/new/announcements/ibm-granite-3-0-open-state-of-the-art-enterprise-models", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "cohere/command-r7b-12-2024", "slug": "cohere--command-r7b-12-2024", "name": "Command R7B", "developer": "Cohere", "developer_id": "cohere", "family": "command-r", "description": "Cohere multilingual text-generation model for long-context chat, retrieval-augmented generation, and multistep tool use.", "release_date": "2024-12-13", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-4.0", "parameters_b": 8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://docs.cohere.com/changelog/command-r-7b", "source_type": "primary", "sources": [ { "label": "Cohere release announcement", "url": "https://docs.cohere.com/changelog/command-r-7b", "type": "primary" }, { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/c4ai-command-r7b-12-2024", "type": "weights" }, { "label": "Exact checkpoint parameter and architecture metadata", "url": "https://huggingface.co/api/models/CohereLabs/c4ai-command-r7b-12-2024", "type": "primary" }, { "label": "Cohere2 dense implementation", "url": "https://raw.githubusercontent.com/huggingface/transformers/v4.55.0/src/transformers/models/cohere2/modeling_cohere2.py", "type": "primary" }, { "label": "Cohere model specifications", "url": "https://docs.cohere.com/docs/command-r7b", "type": "primary" }, { "label": "Cohere native capabilities source", "url": "https://raw.githubusercontent.com/cohere-ai/cohere-developer-experience/main/fern/pages/models/the-command-family-of-models/command-r7b.mdx", "type": "primary" }, { "label": "Cohere capability renderer", "url": "https://raw.githubusercontent.com/cohere-ai/cohere-developer-experience/main/fern/components/model-showcase.tsx", "type": "primary" }, { "label": "Cohere weight license", "url": "https://cohere.com/cohere-labs-cc-by-nc-license", "type": "primary" }, { "label": "Aya Vision report: exact backbone and LLM component size", "url": "https://arxiv.org/html/2505.08751v1", "type": "primary" }, { "label": "Cohere Arabic card: parameter scope corroboration", "url": "https://huggingface.co/CohereLabs/c4ai-command-r7b-arabic-02-2025", "type": "primary" }, { "label": "Arabic checkpoint matching metadata aggregate", "url": "https://huggingface.co/api/models/CohereLabs/c4ai-command-r7b-arabic-02-2025", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/cohere/command-r7b-12-2024.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Approximately 8B including embeddings. The 7B Transformer plus 1B embedding interpretation is inferred from exact checkpoint metadata and later Cohere backbone/component disclosures. Active parameters are not separately disclosed. Weights use CC-BY-NC-4.0 with an acceptable-use addendum. The reasoning flag denotes a dedicated reasoning capability; tool planning is supported.", "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T12:30:27.460357+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/153.json" }, { "id": "microsoft/phi-4", "slug": "microsoft--phi-4", "name": "phi-4", "developer": "Microsoft", "developer_id": "microsoft", "family": "Phi-4", "description": null, "release_date": "2024-12-12", "date_precision": "day", "open_weights": true, "license": "mit", "parameters_b": 14.6595072, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/microsoft/phi-4", "source_type": "primary", "sources": [ { "label": "Developer model card", "url": "https://huggingface.co/microsoft/phi-4", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/microsoft/phi-4", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "First external release was in December 2024; downloadable weights were made available later." }, { "id": "google/gemini-2.0-flash", "slug": "google--gemini-2.0-flash", "name": "Gemini 2.0 Flash", "developer": "Google", "developer_id": "google", "family": "gemini-flash", "description": "Gemini 2.0 model with multimodal input, text output and native tool use.", "release_date": "2024-12-11", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 1048576, "modalities": { "input": [ "text", "image", "audio", "video", "pdf" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "source_type": "primary", "sources": [ { "label": "Gemini 2.0 Flash model card", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-0-Flash-Model-Card.pdf", "type": "primary" }, { "label": "Gemini 2.0 public experimental launch", "url": "https://blog.google/innovation-and-ai/models-and-research/google-deepmind/google-gemini-ai-update-december-2024/", "type": "primary" }, { "label": "Gemini 2.0 Flash general availability", "url": "https://developers.googleblog.com/en/gemini-2-family-expands/", "type": "primary" }, { "label": "Gemini 2.0 Flash API specifications", "url": "https://ai.google.dev/gemini-api/docs/models/gemini-2.0-flash", "type": "primary" }, { "label": "Gemini 2.X technical report", "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_v2_5_report.pdf", "type": "primary" }, { "label": "Gemini 2.5 Flash comparison model card", "url": "https://storage.googleapis.com/deepmind-media/Model-Cards/Gemini-2-5-Flash-Model-Card.pdf", "type": "primary" }, { "label": "Gemini API terms", "url": "https://ai.google.dev/gemini-api/terms", "type": "primary" }, { "label": "Google Cloud Gemini 2.0 Flash specifications", "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/2-0-flash?hl=de", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemini-2.0-flash.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Sparse MoE Transformer; total and active parameter counts are undisclosed. Release date is the public experimental debut; stable API GA followed February 5, 2025. Specifications describe standard text-output Flash, distinct from Flash Thinking and image-generation variants. Google documents shutdown on June 1, 2026.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T12:43:16Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/190.json" }, { "id": "deepseek/deepseek-v2.5-1210", "slug": "deepseek--deepseek-v2.5-1210", "name": "DeepSeek-V2.5-1210", "developer": "DeepSeek", "developer_id": "deepseek", "family": "DeepSeek-V2.5", "description": null, "release_date": "2024-12-10", "date_precision": "day", "open_weights": true, "license": "DEEPSEEK LICENSE AGREEMENT", "parameters_b": 235.74143488, "active_parameters_b": 21, "parameters_source": "checkpoint_metadata", "architecture": "moe", "context_window": 163840, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.deepseek.com/news/v2-5-final/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://www.deepseek.com/news/v2-5-final/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V2.5-1210", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/deepseek-ai/DeepSeek-V2.5-1210", "type": "weights" }, { "label": "Model license", "url": "https://github.com/deepseek-ai/DeepSeek-V2/blob/main/LICENSE-MODEL", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "lg-ai/exaone-3.5-2.4b-instruct", "slug": "lg-ai--exaone-3.5-2.4b-instruct", "name": "EXAONE-3.5-2.4B-Instruct", "developer": "LG AI Research", "developer_id": "lg-ai", "family": "EXAONE 3.5", "description": null, "release_date": "2024-12-09", "date_precision": "day", "open_weights": true, "license": "EXAONE AI Model License Agreement 1.1 - NC", "parameters_b": 2.40532736, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/LG-AI-EXAONE/EXAONE-3.5", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/LG-AI-EXAONE/EXAONE-3.5", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/LGAI-EXAONE/EXAONE-3.5-2.4B-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/LGAI-EXAONE/EXAONE-3.5-2.4B-Instruct", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/LGAI-EXAONE/EXAONE-3.5-2.4B-Instruct/blob/main/LICENSE", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "lg-ai/exaone-3.5-32b-instruct", "slug": "lg-ai--exaone-3.5-32b-instruct", "name": "EXAONE-3.5-32B-Instruct", "developer": "LG AI Research", "developer_id": "lg-ai", "family": "EXAONE 3.5", "description": null, "release_date": "2024-12-09", "date_precision": "day", "open_weights": true, "license": "EXAONE AI Model License Agreement 1.1 - NC", "parameters_b": 32.0032, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/LG-AI-EXAONE/EXAONE-3.5", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/LG-AI-EXAONE/EXAONE-3.5", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/LGAI-EXAONE/EXAONE-3.5-32B-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/LGAI-EXAONE/EXAONE-3.5-32B-Instruct", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/LGAI-EXAONE/EXAONE-3.5-32B-Instruct/blob/main/LICENSE", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "lg-ai/exaone-3.5-7.8b-instruct", "slug": "lg-ai--exaone-3.5-7.8b-instruct", "name": "EXAONE-3.5-7.8B-Instruct", "developer": "LG AI Research", "developer_id": "lg-ai", "family": "EXAONE 3.5", "description": null, "release_date": "2024-12-09", "date_precision": "day", "open_weights": true, "license": "EXAONE AI Model License Agreement 1.1 - NC", "parameters_b": 7.818448896, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/LG-AI-EXAONE/EXAONE-3.5", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/LG-AI-EXAONE/EXAONE-3.5", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/LGAI-EXAONE/EXAONE-3.5-7.8B-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/LGAI-EXAONE/EXAONE-3.5-7.8B-Instruct", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/LGAI-EXAONE/EXAONE-3.5-7.8B-Instruct/blob/main/LICENSE", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "meta/llama-3.3-70b-instruct", "slug": "meta--llama-3.3-70b-instruct", "name": "Llama-3.3-70B-Instruct", "developer": "Meta", "developer_id": "meta", "family": "llama", "description": "Meta's 70B instruction-tuned Llama 3.3 model for multilingual text chat, code generation and tool calling.", "release_date": "2024-12-06", "date_precision": "day", "open_weights": true, "license": "Llama 3.3 Community License Agreement", "parameters_b": 70, "active_parameters_b": 70, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_3/MODEL_CARD.md", "source_type": "primary", "sources": [ { "label": "Llama 3.3 official model card", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_3/MODEL_CARD.md", "type": "primary" }, { "label": "Llama 3.3 Community License Agreement", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_3/LICENSE", "type": "primary" }, { "label": "Meta model identifiers and context limits", "url": "https://github.com/meta-llama/llama-models/blob/main/models/sku_types.py", "type": "primary" }, { "label": "Meta Llama 3.3 70B Instruct architecture registration", "url": "https://github.com/meta-llama/llama-models/blob/main/models/sku_list.py", "type": "primary" }, { "label": "Meta Llama Transformer implementation", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3/model.py", "type": "primary" }, { "label": "Llama 3.3 prompt and tool-calling formats", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_3/prompt_format.md", "type": "primary" }, { "label": "Llama 3.3 developer documentation", "url": "https://dev.meta.ai/llama/docs/model-cards-and-prompt-formats/llama3_3", "type": "primary" }, { "label": "Hugging Face", "url": "https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct", "type": "weights" }, { "label": "Llama models repository overview", "url": "https://github.com/meta-llama/llama-models/blob/main/README.md", "type": "primary" }, { "label": "Phi-4 January 2025 research brief, page 1 active-parameter comparison", "url": "https://www.microsoft.com/en-us/research/wp-content/uploads/2025/01/WEF-2025_Leave-Behind_Phi-4.pdf", "type": "primary" }, { "label": "Benchmarking Open-Ended Multi-Agent Coordination in Language Agents, Table 1 and Figure 11", "url": "https://arxiv.org/html/2606.08340v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/meta/llama-3.3-70b-instruct.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Meta publishes a nominal 70B total. Approximately 70B active follows Microsoft Research's plotted comparison, corroborated by Tessera et al.; exact per-token component scope is unspecified. Dense Transformer with GQA; context is 131072 tokens. Native custom and built-in tool calling are documented; dedicated thinking mode remains undisclosed. Exact card and license date release to December 6, 2024; the overview conflicts with December 4.", "field_status": { "reasoning": "undisclosed" }, "audited_at": "2026-10-27T14:15:59.677877+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/276.json" }, { "id": "openai/o1", "slug": "openai--o1", "name": "o1", "developer": "OpenAI", "developer_id": "openai", "family": "o1", "description": "Reasoning model for complex tasks with text and image input and support for function calling.", "release_date": "2024-12-05", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 200000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-chatgpt-pro/", "source_type": "primary", "sources": [ { "label": "o1 API model reference", "url": "https://developers.openai.com/api/docs/models/o1.md", "type": "primary" }, { "label": "Introducing ChatGPT Pro — December 5, 2024", "url": "https://openai.com/index/introducing-chatgpt-pro/", "type": "primary" }, { "label": "OpenAI o1 and new tools for developers", "url": "https://openai.com/index/o1-and-new-tools-for-developers/", "type": "primary" }, { "label": "OpenAI API changelog", "url": "https://developers.openai.com/api/docs/changelog.md", "type": "primary" }, { "label": "OpenAI o1 System Card", "url": "https://cdn.openai.com/o1-system-card-20241205.pdf", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201.md", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "OpenAI API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning", "url": "https://arxiv.org/html/2501.12948v1", "type": "primary" }, { "label": "Qwen3 Technical Report", "url": "https://arxiv.org/html/2505.09388v1", "type": "primary" }, { "label": "Can Large Reasoning Models Reason about Spatial Relations?", "url": "https://cgi.di.uoa.gr/~koubarak/publications/2025/Sigspatial_GeoAI_paper.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/o1.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Full o1 first released in ChatGPT on December 5, 2024. The API default o1-2024-12-17 is a later post-trained revision with function calling and vision; its last-documented context is 200,000 tokens with 100,000 maximum output. OpenAI shut down o1 and that API snapshot on October 23, 2026. PDF input is converted to text and page images. Broad Transformer architecture is inferred from exact-model caching support and the applicable pre-GPT-5.6 engineering explanation; detailed topology remains undisclosed. Total and active parameter counts remain undisclosed; the documented December post-training lineage supplies no nominal size.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T17:18:51.333924Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/451.json" }, { "id": "openai/o1-pro", "slug": "openai--o1-pro", "name": "o1-pro", "developer": "OpenAI", "developer_id": "openai", "family": "o1", "description": "Higher-compute version of o1 for reasoning tasks, with text and image input.", "release_date": "2024-12-05", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 200000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://openai.com/index/introducing-chatgpt-pro/", "source_type": "primary", "sources": [ { "label": "Introducing ChatGPT Pro", "url": "https://openai.com/index/introducing-chatgpt-pro/", "type": "primary" }, { "label": "o1-pro model reference", "url": "https://developers.openai.com/api/docs/models/o1-pro", "type": "primary" }, { "label": "OpenAI API changelog: March 19, 2025", "url": "https://developers.openai.com/api/docs/changelog", "type": "primary" }, { "label": "Model Release Notes: o3-pro, June 10, 2025", "url": "https://help.openai.com/en/articles/9624314-model-release-notes", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/pdf-files", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "OpenAI API deprecations", "url": "https://developers.openai.com/api/docs/deprecations", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/o1-pro.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "First publicly available as o1 pro mode in ChatGPT Pro on December 5, 2024. The distinct API identifier o1-pro-2025-03-19 launched on March 19, 2025; identical checkpoints across product releases are not assumed. A retrospective OpenAI usage-paper timeline instead says December 1; the contemporaneous launch is preferred. The 200,000-token context and tool/input support are last documented API specifications for the March 19 snapshot. PDF inputs are converted to text and page images. OpenAI lists o1-pro and its snapshot as shut down on October 23, 2026; o3-pro replaced it in the ChatGPT model picker on June 10, 2025. Total and active parameters remain undisclosed. OpenAI identifies o1-pro as a higher-compute version of o1, but the checked sources provide no nominal o1 count to inherit. Architecture remains unverified: higher inference compute, family lineage, and generic cache explanations do not establish exact o1-pro internals.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "unverified" }, "audited_at": "2026-10-27T17:18:39.181816Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/452.json" }, { "id": "amazon/nova-lite", "slug": "amazon--nova-lite", "name": "Nova Lite", "developer": "Amazon", "developer_id": "amazon", "family": "nova", "description": "Nova model with text, image, video and PDF input, text output, a 300K-token context and tool calling.", "release_date": "2024-12-03", "date_precision": "day", "open_weights": false, "license": "AWS Service Terms", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 300000, "modalities": { "input": [ "text", "image", "video", "pdf" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/amazon/nova-lite.toml", "source_type": "catalog", "sources": [ { "label": "Introducing Amazon Nova: A New Generation of Foundation Models", "url": "https://press.aboutamazon.com/2024/12/introducing-amazon-nova-a-new-generation-of-foundation-models", "type": "primary" }, { "label": "Nova Lite - Amazon Bedrock", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-lite.html", "type": "primary" }, { "label": "What is Amazon Nova? - Version 1", "url": "https://docs.aws.amazon.com/nova/latest/userguide/what-is-nova.html", "type": "primary" }, { "label": "The Amazon Nova Family of Models: Technical Report and Model Card", "url": "https://arxiv.org/html/2506.12103v1", "type": "primary" }, { "label": "Amazon Nova Micro, Lite, Pro, and Premier - AWS AI Service Cards", "url": "https://docs.aws.amazon.com/ai/responsible-ai/nova-micro-lite-pro/overview.html", "type": "primary" }, { "label": "Customize Amazon Nova in Amazon SageMaker AI", "url": "https://aws.amazon.com/about-aws/whats-new/2025/07/customize-amazon-nova-sagemaker-ai/", "type": "primary" }, { "label": "Available foundation models - Amazon SageMaker AI", "url": "https://docs.aws.amazon.com/sagemaker/latest/dg/jumpstart-foundation-models-latest.html", "type": "primary" }, { "label": "Iterative Training - Amazon Nova", "url": "https://docs.aws.amazon.com/nova/latest/userguide/nova-iterative-training.html", "type": "primary" }, { "label": "Migrate from Amazon Nova 1 to Amazon Nova 2 on Amazon Bedrock", "url": "https://aws.amazon.com/blogs/machine-learning/migrate-from-amazon-nova-1-to-amazon-nova-2-on-amazon-bedrock/", "type": "primary" }, { "label": "Complete request schema - Amazon Nova", "url": "https://docs.aws.amazon.com/nova/latest/userguide/complete-request-schema.html", "type": "primary" }, { "label": "Amazon Nova Premier: Technical Report and Model Card", "url": "https://cdn.amazon.science/e5/e6/ccc5378c42dca467d1abe1628ec9/amazon-nova-premier-technical-report-and-model-card.pdf", "type": "primary" }, { "label": "Amazon Nova 2: Multimodal Reasoning and Generation Models", "url": "https://cdn.amazon.science/c5/3d/84514a224666b5be6de4b43ef4aa/nova-2-0-technical-report2.pdf", "type": "primary" }, { "label": "MASSIVE-Agents: A Benchmark for Multilingual Function-Calling in 52 Languages", "url": "https://aclanthology.org/2025.findings-emnlp.1099.pdf", "type": "primary" }, { "label": "PHLoRA: data-free Post-hoc Low-Rank Adapter extraction from full-rank checkpoint", "url": "https://cdn.amazon.science/50/3b/6b14413b42b7a50f6c97ea63f0cd/post-hoc-adapter-compression-extracting-low-rank-updates-from-full-rank-finetuning-er-cr.pdf", "type": "primary" }, { "label": "Fine-tune Nova 1.0 - Amazon Nova", "url": "https://docs.aws.amazon.com/nova/latest/userguide/nova-fine-tune-1.html", "type": "primary" }, { "label": "Continued pre-training (CPT) on Nova 2.0 on SageMaker HyperPod", "url": "https://docs.aws.amazon.com/nova/latest/nova2-userguide/nova-cpt-2.html", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/amazon/nova-lite.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Total and active parameter counts remain undisclosed; later Amazon comparison tables provide no numeric size. Transformer architecture is confirmed; dense versus MoE topology is unknown. Public GA was December 3, 2024, despite the current Bedrock card listing December 5. Dedicated reasoning remains unresolved: the AWS migration table says Nova 1 Lite has no extended thinking, while the v1 API schema documents reasoningConfig for Lite. AWS classifies Nova as proprietary; the audit records inconsistent SageMaker table grouping.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "reasoning": "unverified" }, "audited_at": "2026-10-27T10:03:58.373963+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/078.json" }, { "id": "amazon/nova-lite-v1", "slug": "amazon--nova-lite-v1", "name": "Nova Lite 1.0", "developer": "Amazon", "developer_id": "amazon", "family": "nova", "description": "Nova model with text, image, video and PDF input, text output, a 300K-token context and tool calling.", "release_date": "2024-12-03", "date_precision": "day", "open_weights": false, "license": "AWS Service Terms", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 300000, "modalities": { "input": [ "text", "image", "video", "pdf" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://press.aboutamazon.com/2024/12/introducing-amazon-nova-a-new-generation-of-foundation-models", "source_type": "primary", "sources": [ { "label": "Amazon Nova launch and public availability", "url": "https://press.aboutamazon.com/2024/12/introducing-amazon-nova-a-new-generation-of-foundation-models", "type": "primary" }, { "label": "Nova Lite Bedrock model card", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-lite.html", "type": "primary" }, { "label": "Amazon Nova technical report", "url": "https://cdn.amazon.science/96/7d/0d3e59514abf8fdcfafcdc574300/nova-tech-report-20250317-0810.pdf", "type": "primary" }, { "label": "Nova model service card", "url": "https://docs.aws.amazon.com/ai/responsible-ai/nova-micro-lite-pro/overview.html", "type": "primary" }, { "label": "Nova proprietary customization announcement", "url": "https://aws.amazon.com/about-aws/whats-new/2025/07/customize-amazon-nova-sagemaker-ai/", "type": "primary" }, { "label": "Nova Lite license in SageMaker inventory", "url": "https://docs.aws.amazon.com/sagemaker/latest/dg/jumpstart-foundation-models-latest.html", "type": "primary" }, { "label": "Nova 1 to Nova 2 feature comparison", "url": "https://aws.amazon.com/blogs/machine-learning/migrate-from-amazon-nova-1-to-amazon-nova-2-on-amazon-bedrock/", "type": "primary" }, { "label": "Nova proprietary checkpoint handling", "url": "https://docs.aws.amazon.com/nova/latest/userguide/nova-iterative-training.html", "type": "primary" }, { "label": "Nova request schema reasoning conflict", "url": "https://docs.aws.amazon.com/nova/latest/userguide/complete-request-schema.html", "type": "primary" }, { "label": "Nova Version 1 Converse API", "url": "https://docs.aws.amazon.com/nova/latest/userguide/using-converse-api.html", "type": "primary" }, { "label": "Amazon Nova Version 1 model and document specifications", "url": "https://docs.aws.amazon.com/nova/latest/userguide/what-is-nova.html", "type": "primary" }, { "label": "MASSIVE-Agents: A Benchmark for Multilingual Function-Calling in 52 Languages", "url": "https://cdn.amazon.science/91/54/8aa885d04ccbb58f9b16da17f632/scipub-approval152129-35517880-massiveagents-a-benchmark-for-multilingual-functioncalling-in-52-languages.pdf", "type": "primary" }, { "label": "PHLoRA: data-free Post-hoc Low-Rank Adapter extraction from full-rank checkpoint", "url": "https://cdn.amazon.science/50/3b/6b14413b42b7a50f6c97ea63f0cd/post-hoc-adapter-compression-extracting-low-rank-updates-from-full-rank-finetuning-er-cr.pdf", "type": "primary" }, { "label": "Continued pre-training (CPT) on Nova 2.0 on SageMaker HyperPod", "url": "https://docs.aws.amazon.com/nova/latest/nova2-userguide/nova-cpt-2.html", "type": "primary" }, { "label": "Fine-tune Nova 1.0 - Amazon Nova", "url": "https://docs.aws.amazon.com/nova/latest/userguide/nova-fine-tune-1.html", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/amazon/nova-lite-v1.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Public GA was announced December 3, 2024; the later Bedrock card lists December 5. Transformer-based is disclosed; total/active counts and dense/MoE topology remain undisclosed. Amazon AGI lists Lite parameters as NP; Nova 2 Lite is explicitly a larger model, so its size cannot be transferred to Lite v1. AWS Service Terms are proprietary use terms; Nova-specific documentation identifies proprietary weights held in AWS-managed escrow despite the JumpStart inventory's broad Open-Weight heading. Reasoning mode is unresolved: the Nova 1 Lite migration table says no extended thinking, while the current shared Nova request schema assigns reasoningConfig to Lite. PDF input is supported through the Converse API.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "reasoning": "unverified" }, "audited_at": "2026-10-27T10:04:06.081Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/079.json" }, { "id": "amazon/nova-micro", "slug": "amazon--nova-micro", "name": "Nova Micro", "developer": "Amazon", "developer_id": "amazon", "family": "nova", "description": "Text-only Nova model for summarization, translation and classification, with a 128K-token context and tool calling.", "release_date": "2024-12-03", "date_precision": "day", "open_weights": false, "license": "AWS Service Terms", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://press.aboutamazon.com/2024/12/introducing-amazon-nova-a-new-generation-of-foundation-models", "source_type": "primary", "sources": [ { "label": "Introducing Amazon Nova: A New Generation of Foundation Models", "url": "https://press.aboutamazon.com/2024/12/introducing-amazon-nova-a-new-generation-of-foundation-models", "type": "primary" }, { "label": "The Amazon Nova Family of Models: Technical Report and Model Card (March 2025 revision)", "url": "https://cdn.amazon.science/96/7d/0d3e59514abf8fdcfafcdc574300/nova-tech-report-20250317-0810.pdf", "type": "primary" }, { "label": "Amazon Nova Micro, Lite, Pro, and Premier - AWS AI Service Cards", "url": "https://docs.aws.amazon.com/ai/responsible-ai/nova-micro-lite-pro/overview.html", "type": "primary" }, { "label": "Nova Micro - Amazon Bedrock", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-micro.html", "type": "primary" }, { "label": "Amazon Nova Premier: Technical Report and Model Card", "url": "https://cdn.amazon.science/e5/e6/ccc5378c42dca467d1abe1628ec9/amazon-nova-premier-technical-report-and-model-card.pdf", "type": "primary" }, { "label": "Amazon Nova 2: Multimodal Reasoning and Generation Models", "url": "https://cdn.amazon.science/c5/3d/84514a224666b5be6de4b43ef4aa/nova-2-0-technical-report2.pdf", "type": "primary" }, { "label": "MASSIVE-Agents: A Benchmark for Multilingual Function-Calling in 52 Languages", "url": "https://cdn.amazon.science/91/54/8aa885d04ccbb58f9b16da17f632/scipub-approval152129-35517880-massiveagents-a-benchmark-for-multilingual-functioncalling-in-52-languages.pdf", "type": "primary" }, { "label": "Complete request schema - Amazon Nova", "url": "https://docs.aws.amazon.com/nova/latest/userguide/complete-request-schema.html", "type": "primary" }, { "label": "Available foundation models - Amazon SageMaker AI", "url": "https://docs.aws.amazon.com/sagemaker/latest/dg/jumpstart-foundation-models-latest.html", "type": "primary" }, { "label": "Iterative Training - Amazon Nova", "url": "https://docs.aws.amazon.com/nova/latest/userguide/nova-iterative-training.html", "type": "primary" }, { "label": "AWS Service Terms", "url": "https://aws.amazon.com/service-terms/", "type": "primary" }, { "label": "Customize Amazon Nova in Amazon SageMaker AI using Direct Preference Optimization", "url": "https://aws.amazon.com/blogs/machine-learning/customize-amazon-nova-in-amazon-sagemaker-ai-using-direct-preference-optimization/", "type": "primary" }, { "label": "Amazon Nova", "url": "https://aws.amazon.com/nova/", "type": "primary" }, { "label": "MASSIVE-Agents: A Benchmark for Multilingual Function-Calling in 52 Languages (EMNLP 2025 archival version)", "url": "https://aclanthology.org/2025.findings-emnlp.1099.pdf", "type": "primary" }, { "label": "PHLoRA: data-free Post-hoc Low-Rank Adapter extraction from full-rank checkpoint (workshop version)", "url": "https://cdn.amazon.science/50/3b/6b14413b42b7a50f6c97ea63f0cd/post-hoc-adapter-compression-extracting-low-rank-updates-from-full-rank-finetuning-er-cr.pdf", "type": "primary" }, { "label": "PHLoRA: Data-Free Post-Hoc Low-Rank Adapter Extraction from Full-Rank Checkpoint (December 2025 archival version)", "url": "https://aclanthology.org/2025.findings-ijcnlp.125.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/amazon/nova-micro.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Total and active parameter counts remain undisclosed; Amazon-authored MASSIVE-Agents lists Micro Params as NP, and PHLoRA reports adapter ranks rather than model size. Transformer is confirmed, while dense versus MoE is unknown. Release uses Amazon’s December 3, 2024 GA announcement despite a December 5 date on the current Bedrock card. Micro remains proprietary under AWS Service Terms: customized weights stay in Amazon-controlled escrow; the JumpStart open-weight table heading conflicts with explicit weight-access documentation. Reasoning=false refers to a dedicated reasoning mode, not inability to perform reasoning tasks.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T10:05:51.895133+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/080.json" }, { "id": "amazon/nova-micro-v1", "slug": "amazon--nova-micro-v1", "name": "Nova Micro 1.0", "developer": "Amazon", "developer_id": "amazon", "family": "nova", "description": "Text-only Nova model for summarization, translation and classification, with a 128K-token context and tool calling.", "release_date": "2024-12-03", "date_precision": "day", "open_weights": false, "license": "AWS Service Terms", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leading-price-performance/", "source_type": "primary", "sources": [ { "label": "Introducing Amazon Nova foundation models: Frontier intelligence and industry leading price performance", "url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leading-price-performance/", "type": "primary" }, { "label": "What is Amazon Nova? (Version 1)", "url": "https://docs.aws.amazon.com/nova/latest/userguide/what-is-nova.html", "type": "primary" }, { "label": "Nova Micro - Amazon Bedrock", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-micro.html", "type": "primary" }, { "label": "The Amazon Nova Family of Models: Technical Report and Model Card", "url": "https://cdn.amazon.science/96/7d/0d3e59514abf8fdcfafcdc574300/nova-tech-report-20250317-0810.pdf", "type": "primary" }, { "label": "Amazon Nova Premier: Technical Report and Model Card", "url": "https://cdn.amazon.science/e5/e6/ccc5378c42dca467d1abe1628ec9/amazon-nova-premier-technical-report-and-model-card.pdf", "type": "primary" }, { "label": "Amazon Nova 2: Multimodal Reasoning and Generation Models", "url": "https://cdn.amazon.science/c5/3d/84514a224666b5be6de4b43ef4aa/nova-2-0-technical-report2.pdf", "type": "primary" }, { "label": "AWS AI Service Cards - Amazon Nova Micro, Lite, Pro, and Premier", "url": "https://docs.aws.amazon.com/ai/responsible-ai/nova-micro-lite-pro/overview.html", "type": "primary" }, { "label": "Iterative Training - Amazon Nova", "url": "https://docs.aws.amazon.com/nova/latest/userguide/nova-iterative-training.html", "type": "primary" }, { "label": "Available foundation models - Amazon SageMaker AI", "url": "https://docs.aws.amazon.com/sagemaker/latest/dg/jumpstart-foundation-models-latest.html", "type": "primary" }, { "label": "Complete request schema - Amazon Nova", "url": "https://docs.aws.amazon.com/nova/latest/userguide/complete-request-schema.html", "type": "primary" }, { "label": "MASSIVE-Agents: A Benchmark for Multilingual Function-Calling in 52 Languages", "url": "https://aclanthology.org/2025.findings-emnlp.1099.pdf", "type": "primary" }, { "label": "PHLoRA: data-free Post-hoc Low-Rank Adapter extraction from full-rank checkpoint (workshop version)", "url": "https://cdn.amazon.science/50/3b/6b14413b42b7a50f6c97ea63f0cd/post-hoc-adapter-compression-extracting-low-rank-updates-from-full-rank-finetuning-er-cr.pdf", "type": "primary" }, { "label": "PHLoRA: Data-Free Post-Hoc Low-Rank Adapter Extraction from Full-Rank Checkpoint (December 2025 archival version)", "url": "https://aclanthology.org/2025.findings-ijcnlp.125.pdf", "type": "primary" }, { "label": "Fine-tune Nova 1.0 - Amazon Nova", "url": "https://docs.aws.amazon.com/nova/latest/userguide/nova-fine-tune-1.html", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/amazon/nova-micro-v1.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The Dec 3, 2024 launch announcement explicitly states availability; the current Bedrock card instead lists Dec 5, 2024, an unexplained date conflict. Transformer backbone is documented; total and active parameters remain undisclosed. Amazon AGI’s MASSIVE-Agents table lists Micro parameters as NP; PHLoRA reports adapter ranks only. No size-disclosed base lineage is established. Nova weights are proprietary and trained checkpoints remain in AWS-managed escrow. Micro accepts text only, with no native document input or configurable thinking mode. Endpoint operation was not tested.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T10:07:12.388693Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/081.json" }, { "id": "amazon/nova-pro", "slug": "amazon--nova-pro", "name": "Nova Pro", "developer": "Amazon", "developer_id": "amazon", "family": "nova", "description": "First-generation Amazon Nova multimodal model for document and video analysis, code generation, and tool-using applications.", "release_date": "2024-12-03", "date_precision": "day", "open_weights": false, "license": "AWS Service Terms", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 300000, "modalities": { "input": [ "text", "image", "video", "pdf" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leading-price-performance/", "source_type": "primary", "sources": [ { "label": "AWS Nova launch (December 3, 2024)", "url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leading-price-performance/", "type": "primary" }, { "label": "Amazon Nova technical report", "url": "https://cdn.amazon.science/96/7d/0d3e59514abf8fdcfafcdc574300/nova-tech-report-20250317-0810.pdf", "type": "primary" }, { "label": "AWS Nova service card", "url": "https://docs.aws.amazon.com/ai/responsible-ai/nova-micro-lite-pro/overview.html", "type": "primary" }, { "label": "Nova Pro terms in SageMaker catalog", "url": "https://docs.aws.amazon.com/sagemaker/latest/dg/jumpstart-foundation-models-latest.html", "type": "primary" }, { "label": "Nova proprietary checkpoint handling", "url": "https://docs.aws.amazon.com/nova/latest/userguide/nova-iterative-training.html", "type": "primary" }, { "label": "Version 1 request schema", "url": "https://docs.aws.amazon.com/nova/latest/userguide/complete-request-schema.html", "type": "primary" }, { "label": "Nova 1 to Nova 2 migration comparison", "url": "https://aws.amazon.com/blogs/machine-learning/migrate-from-amazon-nova-1-to-amazon-nova-2-on-amazon-bedrock/", "type": "primary" }, { "label": "Version 1 extended-thinking examples", "url": "https://docs.aws.amazon.com/nova/latest/userguide/extended-thinking.html", "type": "primary" }, { "label": "Amazon Nova Premier: Technical Report and Model Card", "url": "https://cdn.amazon.science/e5/e6/ccc5378c42dca467d1abe1628ec9/amazon-nova-premier-technical-report-and-model-card.pdf", "type": "primary" }, { "label": "Amazon Nova 2: Multimodal Reasoning and Generation Models", "url": "https://cdn.amazon.science/c5/3d/84514a224666b5be6de4b43ef4aa/nova-2-0-technical-report2.pdf", "type": "primary" }, { "label": "MASSIVE-Agents: A Benchmark for Multilingual Function-Calling in 52 Languages (EMNLP 2025 archival version)", "url": "https://aclanthology.org/2025.findings-emnlp.1099.pdf", "type": "primary" }, { "label": "PHLoRA: Data-Free Post-Hoc Low-Rank Adapter Extraction from Full-Rank Checkpoint", "url": "https://aclanthology.org/2025.findings-ijcnlp.125.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/amazon/nova-pro.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Original endpoint: amazon.nova-pro-v1:0. Transformer-based; total and active parameters remain undisclosed. Amazon AGI’s MASSIVE-Agents table lists Pro parameters as NP; PHLoRA reports adapter ranks only. No size-disclosed base lineage is established. Dense/MoE structure is undisclosed. Extended-thinking support is unresolved: current Version 1 schema names Pro as supporting reasoningConfig, but AWS's March 2026 migration table says Nova 1 Pro does not; the Version 1 examples use Lite 1.5. PDF input uses Converse. Public launch was December 3, 2024; later official listings show conflicting December 2/5 dates. Nova weights are proprietary despite broad JumpStart open-weight tab placement.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "reasoning": "unverified" }, "audited_at": "2026-10-27T10:08:46Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/083.json" }, { "id": "amazon/nova-pro-v1", "slug": "amazon--nova-pro-v1", "name": "Nova Pro 1.0", "developer": "Amazon", "developer_id": "amazon", "family": "nova", "description": "Amazon Nova Pro 1.0 is a multimodal Transformer with a 300K-token context window, document and video inputs, and tool calling.", "release_date": "2024-12-03", "date_precision": "day", "open_weights": false, "license": "AWS Service Terms", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 300000, "modalities": { "input": [ "text", "image", "video", "pdf" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://press.aboutamazon.com/2024/12/introducing-amazon-nova-a-new-generation-of-foundation-models", "source_type": "primary", "sources": [ { "label": "Amazon Nova general-availability announcement", "url": "https://press.aboutamazon.com/2024/12/introducing-amazon-nova-a-new-generation-of-foundation-models", "type": "primary" }, { "label": "Amazon Nova family technical report", "url": "https://arxiv.org/html/2506.12103v1", "type": "primary" }, { "label": "Nova Pro Bedrock model card", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-amazon-nova-pro.html", "type": "primary" }, { "label": "AWS Nova 1 to Nova 2 migration table", "url": "https://aws.amazon.com/blogs/machine-learning/migrate-from-amazon-nova-1-to-amazon-nova-2-on-amazon-bedrock/", "type": "primary" }, { "label": "Nova shared request schema (conflicting reasoning guidance)", "url": "https://docs.aws.amazon.com/nova/latest/userguide/complete-request-schema.html", "type": "primary" }, { "label": "Nova proprietary weight and checkpoint policy", "url": "https://docs.aws.amazon.com/nova/latest/userguide/nova-iterative-training.html", "type": "primary" }, { "label": "AWS SageMaker Nova Pro license listing", "url": "https://docs.aws.amazon.com/sagemaker/latest/dg/jumpstart-foundation-models-latest.html", "type": "primary" }, { "label": "AWS Service Terms", "url": "https://aws.amazon.com/service-terms/", "type": "primary" }, { "label": "Nova version 1 tool calling", "url": "https://docs.aws.amazon.com/nova/latest/userguide/tool-use.html", "type": "primary" }, { "label": "What is Amazon Nova? — Version 1 model specifications", "url": "https://docs.aws.amazon.com/nova/latest/userguide/what-is-nova.html", "type": "primary" }, { "label": "Amazon Nova Premier: Technical Report and Model Card", "url": "https://cdn.amazon.science/e5/e6/ccc5378c42dca467d1abe1628ec9/amazon-nova-premier-technical-report-and-model-card.pdf", "type": "primary" }, { "label": "Amazon Nova 2: Multimodal Reasoning and Generation Models", "url": "https://cdn.amazon.science/c5/3d/84514a224666b5be6de4b43ef4aa/nova-2-0-technical-report2.pdf", "type": "primary" }, { "label": "MASSIVE-Agents: A Benchmark for Multilingual Function-Calling in 52 Languages", "url": "https://aclanthology.org/2025.findings-emnlp.1099.pdf", "type": "primary" }, { "label": "PHLoRA: Data-Free Post-Hoc Low-Rank Adapter Extraction from Full-Rank Checkpoint", "url": "https://aclanthology.org/2025.findings-ijcnlp.125.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/amazon/nova-pro-v1.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Total and active parameters remain undisclosed; Amazon AGI’s MASSIVE-Agents table marks Nova Pro Params as NP. The reported 90B estimate is secondary and benchmark-based. Dense/MoE structure is unspecified. Amazon announced general availability on December 3, 2024; the Bedrock card's December 5 date conflicts. Reasoning-mode support remains unresolved: AWS's Nova 1 Pro migration table says no extended thinking, while its current shared schema claims reasoningConfig for Nova Pro without an exact generation ID. Dedicated Nova training documentation explicitly identifies proprietary weights and AWS-managed escrow storage despite the broad JumpStart table's inconsistent Open-Weight heading.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "reasoning": "unverified" }, "audited_at": "2026-10-27T10:10:05.508Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/084.json" }, { "id": "alibaba/qwen-mt-plus", "slug": "alibaba--qwen-mt-plus", "name": "Qwen MT Plus", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Flagship Qwen translation model with terminology control and support for 92 languages", "release_date": "2024-12", "date_precision": "month", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": false, "source_url": "https://developer.aliyun.com/article/1647733", "source_type": "primary", "sources": [ { "label": "Alibaba exact-model documentation", "url": "https://help.aliyun.com/en/model-studio/qwen-mt-plus", "type": "primary" }, { "label": "Alibaba translation guide", "url": "https://www.alibabacloud.com/help/en/model-studio/machine-translation", "type": "primary" }, { "label": "Alibaba model lifecycle and regional dates", "url": "https://help.aliyun.com/zh/model-studio/newly-released-models", "type": "primary" }, { "label": "Alibaba capability table", "url": "https://help.aliyun.com/zh/model-studio/text-generation-model", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen-mt-plus.toml", "type": "catalog" }, { "label": "Alibaba Chinese exact-model documentation", "url": "https://help.aliyun.com/zh/model-studio/qwen-mt-plus", "type": "primary" }, { "label": "Alibaba Cloud Bailian product newsletter: December 2024 new model releases", "url": "https://developer.aliyun.com/article/1647733", "type": "primary" }, { "label": "Alibaba Cloud model release history: Qwen-MT 2025 upgrades", "url": "https://www.alibabacloud.com/help/tc/model-studio/newly-released-models", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=qwen-mt-plus&limit=100", "label": "Official Qwen inventory search for the exact hosted model", "type": "primary" }, { "url": "https://huggingface.co/api/models?author=Qwen&search=Qwen-MT&limit=1000", "label": "Official Qwen Qwen-MT weight inventory", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Total and active parameter counts are undisclosed. Qwen3 lineage is documented without a size-specific base; Plus dense/MoE architecture remains unestablished. The official December 2024 product newsletter lists Qwen-MT Plus as newly launched, establishing 2024-12 at month precision. July 22 (Beijing) and July 24 (international/global), 2025 describe the later Qwen3 upgrade. Classified as closed weights: no public release of this hosted variant was found in the developer documentation and weight inventory.", "field_status": { "license": "undisclosed", "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-28T08:01:10.202601+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/014.json" }, { "id": "aion-labs/aion-rp-llama-3.1-8b", "slug": "aion-labs--aion-rp-llama-3.1-8b", "name": "Aion-RP 1.0 (8B)", "developer": "Aion Labs", "developer_id": "aion-labs", "family": "llama", "description": "Llama 3.1 8B fine-tune for roleplaying and creative writing.", "release_date": "2024-11-30", "date_precision": "day", "open_weights": true, "license": null, "parameters_b": 8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": false, "source_url": "https://www.aionlabs.ai/docs/models/", "source_type": "primary", "sources": [ { "label": "Aion Labs model documentation", "url": "https://www.aionlabs.ai/docs/models/", "type": "docs" }, { "label": "Aion Labs models API", "url": "https://api.aionlabs.ai/v1/models", "type": "docs" }, { "label": "Fine-tune model card (pinned revision)", "url": "https://huggingface.co/aion-labs/Aion-RP-Llama-3.1-8B/raw/8ae617bc85c5d6698c3323863c42098d1998f7bb/README.md", "type": "model_card" }, { "label": "Fine-tune configuration (pinned revision)", "url": "https://huggingface.co/aion-labs/Aion-RP-Llama-3.1-8B/raw/8ae617bc85c5d6698c3323863c42098d1998f7bb/config.json", "type": "weights" }, { "label": "Fine-tune weight index (pinned revision)", "url": "https://huggingface.co/aion-labs/Aion-RP-Llama-3.1-8B/raw/8ae617bc85c5d6698c3323863c42098d1998f7bb/model.safetensors.index.json", "type": "weights" }, { "label": "Base-model Llama 3.1 license", "url": "https://huggingface.co/meta-llama/Llama-3.1-8B/blob/main/LICENSE", "type": "license" }, { "label": "Meta Llama 3.1 base-model card", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/MODEL_CARD.md", "type": "model_card" }, { "label": "Fine-tune repository parameter metadata", "url": "https://huggingface.co/api/models/aion-labs/Aion-RP-Llama-3.1-8B", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/aion-labs/aion-rp-llama-3.1-8b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Nominal 8B is developer-published and supported by explicit Llama 3.1 8B lineage; repository metadata reports 8,030,261,248 checkpoint parameters. Active parameters are not separately specified. API context is 32,768 tokens; the downloadable fine-tune config supports 131,072 positions. Developer reports release on 2024-11-30, while the repository was created on 2024-11-09. Fine-tune metadata declares Apache-2.0, but its Llama 3.1 base has a separate community license; the complete licensing position is unresolved.", "field_status": { "license": "unverified", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T07:31:05.945Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/009.json" }, { "id": "allenai/olmo-2-1124-13b", "slug": "allenai--olmo-2-1124-13b", "name": "OLMo-2-1124-13B", "developer": "Ai2", "developer_id": "allenai", "family": "OLMo 2", "description": null, "release_date": "2024-11-26", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 13.7161984, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://docs.allenai.org/release_notes/olmo-release-notes", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://docs.allenai.org/release_notes/olmo-release-notes", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/allenai/OLMo-2-1124-13B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/allenai/OLMo-2-1124-13B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "allenai/olmo-2-1124-13b-instruct-preview", "slug": "allenai--olmo-2-1124-13b-instruct-preview", "name": "OLMo-2-1124-13B-Instruct Preview", "developer": "Ai2", "developer_id": "allenai", "family": "OLMo 2", "description": null, "release_date": "2024-11-26", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 13.7161984, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://allenai.org/blog/olmo2", "source_type": "primary", "sources": [ { "label": "Original November release", "url": "https://allenai.org/blog/olmo2", "type": "announcement" }, { "label": "Publisher retraining and historical-checkpoint notice", "url": "https://huggingface.co/allenai/OLMo-2-1124-13B-Instruct", "type": "model_card" }, { "label": "Original released checkpoint", "url": "https://huggingface.co/allenai/OLMo-2-1124-13B-Instruct-preview", "type": "model_card" }, { "label": "Preview checkpoint configuration", "url": "https://huggingface.co/allenai/OLMo-2-1124-13B-Instruct-preview/blob/main/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:36:18.421Z", "last_seen": "2026-10-29T10:36:18.421Z", "upstream_present": false, "availability": "public", "notes": "Original November 2024 instruction checkpoint. Ai2 later retrained the instruction models and moved the original weights to the -preview repository; this record preserves that independently trained release." }, { "id": "allenai/olmo-2-1124-7b", "slug": "allenai--olmo-2-1124-7b", "name": "OLMo-2-1124-7B", "developer": "Ai2", "developer_id": "allenai", "family": "OLMo 2", "description": null, "release_date": "2024-11-26", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 7.298617344, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://docs.allenai.org/release_notes/olmo-release-notes", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://docs.allenai.org/release_notes/olmo-release-notes", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/allenai/OLMo-2-1124-7B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/allenai/OLMo-2-1124-7B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "allenai/olmo-2-1124-7b-instruct-preview", "slug": "allenai--olmo-2-1124-7b-instruct-preview", "name": "OLMo-2-1124-7B-Instruct Preview", "developer": "Ai2", "developer_id": "allenai", "family": "OLMo 2", "description": null, "release_date": "2024-11-26", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 7.298617344, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://allenai.org/blog/olmo2", "source_type": "primary", "sources": [ { "label": "Original November release", "url": "https://allenai.org/blog/olmo2", "type": "announcement" }, { "label": "Publisher retraining and historical-checkpoint notice", "url": "https://huggingface.co/allenai/OLMo-2-1124-7B-Instruct", "type": "model_card" }, { "label": "Original released checkpoint", "url": "https://huggingface.co/allenai/OLMo-2-1124-7B-Instruct-preview", "type": "model_card" }, { "label": "Preview checkpoint configuration", "url": "https://huggingface.co/allenai/OLMo-2-1124-7B-Instruct-preview/blob/main/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:35:23.625Z", "last_seen": "2026-10-29T10:35:23.625Z", "upstream_present": false, "availability": "public", "notes": "Original November 2024 instruction checkpoint. Ai2 later retrained the instruction models and moved the original weights to the -preview repository; this record preserves that independently trained release." }, { "id": "openai/gpt-4o-2024-11-20", "slug": "openai--gpt-4o-2024-11-20", "name": "GPT-4o (2024-11-20)", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-4o", "description": "November 2024 GPT-4o snapshot with text and image input, text output, a 128,000-token context window, and function calling.", "release_date": "2024-11-20", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 128000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://developers.openai.com/api/docs/changelog", "source_type": "primary", "sources": [ { "label": "OpenAI API changelog: November 20, 2024", "url": "https://developers.openai.com/api/docs/changelog", "type": "primary" }, { "label": "GPT-4o model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-4o", "type": "primary" }, { "label": "Foundry Models sold by Azure: GPT-4o capabilities", "url": "https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure", "type": "primary" }, { "label": "Introducing GPT-4.1 in the API: comparison appendix", "url": "https://openai.com/index/gpt-4-1/", "type": "primary" }, { "label": "Reasoning best practices", "url": "https://developers.openai.com/api/docs/guides/reasoning-best-practices", "type": "primary" }, { "label": "API deprecations: GPT-4 series models", "url": "https://developers.openai.com/api/docs/deprecations", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "Introducing gpt-oss", "url": "https://openai.com/index/introducing-gpt-oss/", "type": "primary" }, { "label": "Toward understanding and preventing misalignment generalization", "url": "https://cdn.openai.com/pdf/a130517e-9633-47bc-8397-969807a43a23/emergent_misalignment_paper.pdf", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs", "type": "primary" }, { "label": "Create a model response", "url": "https://developers.openai.com/api/reference/resources/responses/methods/create", "type": "primary" }, { "label": "Qwen3 Technical Report, Table 12", "url": "https://arxiv.org/html/2505.09388v1", "type": "primary" }, { "label": "Reasoning through Exploration: A Reinforcement Learning Framework for Robust Function Calling, Table 1", "url": "https://arxiv.org/html/2508.05118v4", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-4o-2024-11-20.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "This is the fixed November 20, 2024 API snapshot; the undated alias defaults to August 6. Scheduled API shutdown is January 14, 2027; no live endpoint test was performed. PDF inputs are processed as extracted text and page images. Exact-snapshot dedicated reasoning support and architecture remain unverified. Total and active parameter counts remain undisclosed: the Qwen3 comparison explicitly leaves both unreported for this snapshot.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "unverified", "reasoning": "unverified" }, "audited_at": "2026-10-27T16:28:06.222086Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/410.json" }, { "id": "mistral/mistral-large-2411", "slug": "mistral--mistral-large-2411", "name": "Mistral Large 2.1", "developer": "Mistral", "developer_id": "mistral", "family": "mistral-large", "description": "November 2024 Mistral Large 2.1 instruction model with multilingual text, code generation and native function calling", "release_date": "2024-11-18", "date_precision": "day", "open_weights": true, "license": "Mistral Research License", "parameters_b": 123, "active_parameters_b": 123, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://docs.mistral.ai/models/mistral-large-2-1-24-11", "source_type": "primary", "sources": [ { "label": "Mistral Large 2.1 official documentation and weights table", "url": "https://docs.mistral.ai/models/mistral-large-2-1-24-11", "type": "primary" }, { "label": "Mistral Large 2.1 official documentation, Markdown", "url": "https://docs.mistral.ai/models/mistral-large-2-1-24-11.md", "type": "primary" }, { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Mistral-Large-Instruct-2411", "type": "weights" }, { "label": "Official Instruct-2411 Transformers configuration", "url": "https://huggingface.co/mistralai/Mistral-Large-Instruct-2411/raw/ba78820945ae22361b0274cf0ae6d696c967c1a4/config.json", "type": "primary" }, { "label": "Official Instruct-2411 native configuration", "url": "https://huggingface.co/mistralai/Mistral-Large-Instruct-2411/raw/ba78820945ae22361b0274cf0ae6d696c967c1a4/params.json", "type": "primary" }, { "label": "November 18, 2024 Mistral Large 24.11 announcement", "url": "https://mistral.ai/news/pixtral-large/", "type": "primary" }, { "label": "Mistral official release changelog", "url": "https://docs.mistral.ai/resources/changelogs", "type": "primary" }, { "label": "Mistral Research License", "url": "https://mistral.ai/licenses/MRL-0.1.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/mistral-large-2411.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Mistral directly discloses nominal 123B total and 123B active parameters in the exact Instruct-2411 weights row. Both native and Transformers configurations specify 131072 context tokens. Dedicated thinking support remains unverified; general reasoning claims do not establish that capability. The released weights use the restricted Mistral Research License.", "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-26T23:15:12Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/334.json" }, { "id": "alibaba/qwen2.5-coder-0.5b", "slug": "alibaba--qwen2.5-coder-0.5b", "name": "Qwen2.5-Coder-0.5B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Qwen2.5-Coder base language model for code generation and fill-in-the-middle completion, released with open weights under Apache 2.0.", "release_date": "2024-11-12", "date_precision": "day", "open_weights": true, "license": "Apache 2.0", "parameters_b": 0.49, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen2.5-coder-0.5b.toml", "source_type": "catalog", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen2.5-Coder-0.5B", "type": "weights" }, { "label": "Qwen2.5-Coder-0.5B configuration", "url": "https://huggingface.co/Qwen/Qwen2.5-Coder-0.5B/blob/main/config.json", "type": "weights" }, { "label": "Qwen2.5-Coder Series: Powerful, Diverse, Practical.", "url": "https://qwenlm.github.io/blog/qwen2.5-coder-family/", "type": "primary" }, { "label": "Qwen2.5-Coder Technical Report, revision 3", "url": "https://arxiv.org/html/2409.12186v3", "type": "paper" }, { "label": "Qwen2.5-Coder-0.5B tokenizer configuration", "url": "https://huggingface.co/Qwen/Qwen2.5-Coder-0.5B/resolve/main/tokenizer_config.json", "type": "weights" }, { "label": "Qwen2.5-Coder-0.5B released files", "url": "https://huggingface.co/Qwen/Qwen2.5-Coder-0.5B/tree/main", "type": "weights" }, { "label": "Hugging Face checkpoint metadata for Qwen2.5-Coder-0.5B", "url": "https://huggingface.co/api/models/Qwen/Qwen2.5-Coder-0.5B", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen2.5-coder-0.5b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "This is the pretrained base checkpoint, not the Instruct variant. Developer-rounded total is 0.49B; current Safetensors metadata counts 494,032,768 parameters. The 0.36B non-embedding figure is not an active-parameter count. Dense active parameters are not separately defined. Tool-formatting support in the tokenizer does not establish tool-calling competence of the base weights; dedicated reasoning-mode support also remains unverified.", "field_status": { "active_parameters_b": "not_applicable", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-26T14:12:32.087098Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/022.json" }, { "id": "alibaba/qwen2.5-coder-32b-instruct", "slug": "alibaba--qwen2.5-coder-32b-instruct", "name": "Qwen2.5-Coder-32B-Instruct", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Open-weight instruction-tuned Qwen coding model for code generation, repair, and code reasoning.", "release_date": "2024-11-12", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 32.763876352, "active_parameters_b": 32, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen2.5-coder-32b-instruct.toml", "source_type": "catalog", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen2.5-Coder-32B-Instruct", "type": "weights" }, { "label": "Official model configuration", "url": "https://huggingface.co/Qwen/Qwen2.5-Coder-32B-Instruct/blob/main/config.json", "type": "primary" }, { "label": "Qwen release announcement", "url": "https://qwenlm.github.io/blog/qwen2.5-coder-family/", "type": "primary" }, { "label": "Official checkpoint tensor metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2.5-Coder-32B-Instruct?expand[]=safetensors&expand[]=sha&expand[]=createdAt", "type": "primary" }, { "label": "Official checkpoint index", "url": "https://huggingface.co/Qwen/Qwen2.5-Coder-32B-Instruct/resolve/main/model.safetensors.index.json", "type": "primary" }, { "label": "Official license", "url": "https://huggingface.co/Qwen/Qwen2.5-Coder-32B-Instruct/blob/main/LICENSE", "type": "primary" }, { "label": "Official tool-call chat template", "url": "https://huggingface.co/Qwen/Qwen2.5-Coder-32B-Instruct/blob/main/tokenizer_config.json", "type": "primary" }, { "label": "Qwen2.5-Coder technical report", "url": "https://arxiv.org/html/2409.12186v3", "type": "primary" }, { "label": "Qwen3 Technical Report, version 1", "url": "https://arxiv.org/html/2505.09388v1", "type": "primary" }, { "label": "Qwen2.5-32B base configuration", "url": "https://huggingface.co/Qwen/Qwen2.5-32B/raw/main/config.json", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen2.5-coder-32b-instruct.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Official checkpoint metadata reports 32,763,876,352 parameters; the card and release table publish 32.5B nominal total (32B in the model name). Nominal active 32B is inferred from the documented Qwen2.5 architecture lineage and matching parameter dimensions, using the base model’s developer-published activated count. The separate 31.0B figure excludes embeddings. Context is 32,768 tokens by default, extended to 131,072 with the documented YaRN factor-4 configuration. Dedicated reasoning-mode status remains unverified.", "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-27T07:51:46.870461+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/023.json" }, { "id": "anthropic/claude-3-5-haiku-20241022", "slug": "anthropic--claude-3-5-haiku-20241022", "name": "Claude Haiku 3.5", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude-haiku", "description": "Claude Haiku model released in November 2024 with a 200K-token context window and tool calling.", "release_date": "2024-11-04", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 200000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://platform.claude.com/docs/en/release-notes/overview", "source_type": "primary", "sources": [ { "label": "Anthropic API availability and vision release notes", "url": "https://platform.claude.com/docs/en/release-notes/overview", "type": "documentation" }, { "label": "Claude 3.5 Haiku announcement", "url": "https://www.anthropic.com/news/3-5-models-and-computer-use", "type": "announcement" }, { "label": "Claude 3.5 Haiku model-card addendum", "url": "https://assets.anthropic.com/m/1cd9d098ac3e6467/original/Claude-3-Model-Card-OctoberAddendum.pdf", "type": "model_card" }, { "label": "Claude 3.5 Haiku documented provider specifications", "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/deprecations/partner-models", "type": "documentation" }, { "label": "Anthropic research proprietary-model classification", "url": "https://alignment.anthropic.com/2025/unsupervised-elicitation/paper.pdf", "type": "paper" }, { "label": "Haiku 4.5 card comparing Haiku 3.5", "url": "https://assets.anthropic.com/m/99128ddd009bdcb/original/Claude-Haiku-4-5-System-Card.pdf", "type": "model_card" }, { "label": "On the Biology of a Large Language Model", "url": "https://www.transformer-circuits.pub/2025/attribution-graphs/biology.html", "type": "primary" }, { "label": "Anthropic later pricing table with conflicting context tier", "url": "https://www-cdn.anthropic.com/files/4zrzovbb/website/5678bc2f5978e5bcd4f1fe7c14b2c72284dcf9f8.pdf", "type": "documentation" }, { "label": "Anthropic circuit-tracing methods distinguishing replacement features", "url": "https://www.transformer-circuits.pub/2025/attribution-graphs/methods.html", "type": "paper" }, { "label": "Claude 3.5 Haiku availability", "url": "https://cloud.google.com/blog/products/ai-machine-learning/upgraded-claude-3-5-sonnet-with-computer-use-on-vertex-ai", "type": "announcement" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/anthropic/claude-3-5-haiku-20241022.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Announced October 22, 2024; Anthropic's release notes confirm public API availability November 4. Image input was added February 24, 2025, so modalities describe later documented capability. No extended-thinking mode. Transformer architecture is documented by Anthropic interpretability research; nominal total/active counts and dense/MoE routing remain undisclosed. The study's 30M features belong to a replacement transcoder. Claude API retirement was February 19, 2026; retained capability documentation does not establish current endpoint availability. A May 12, 2026 pricing PDF lists a conflicting 1M API tier after retirement; retained exact-model specifications support 200K.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T10:16:35.510538Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/089.json" }, { "id": "huggingface/smollm2-1.7b", "slug": "huggingface--smollm2-1.7b", "name": "SmolLM2-1.7B", "developer": "Hugging Face", "developer_id": "huggingface", "family": "SmolLM2", "description": null, "release_date": "2024-10-31", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 1.711376384, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.linkedin.com/posts/loubna-ben-allal-238690152_introducing-smollm2-the-new-best-and-open-activity-7257822657004425216-8NjH", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://www.linkedin.com/posts/loubna-ben-allal-238690152_introducing-smollm2-the-new-best-and-open-activity-7257822657004425216-8NjH", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/HuggingFaceTB/SmolLM2-1.7B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/HuggingFaceTB/SmolLM2-1.7B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer launch post links the available three model sizes; its datePublished metadata is 2024-10-31T18:36:05Z. Later media coverage dates are not used as the first release date." }, { "id": "huggingface/smollm2-1.7b-instruct", "slug": "huggingface--smollm2-1.7b-instruct", "name": "SmolLM2-1.7B-Instruct", "developer": "Hugging Face", "developer_id": "huggingface", "family": "SmolLM2", "description": null, "release_date": "2024-10-31", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 1.711376384, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.linkedin.com/posts/loubna-ben-allal-238690152_introducing-smollm2-the-new-best-and-open-activity-7257822657004425216-8NjH", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://www.linkedin.com/posts/loubna-ben-allal-238690152_introducing-smollm2-the-new-best-and-open-activity-7257822657004425216-8NjH", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/HuggingFaceTB/SmolLM2-1.7B-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/HuggingFaceTB/SmolLM2-1.7B-Instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer launch post links the available three model sizes; its datePublished metadata is 2024-10-31T18:36:05Z. Later media coverage dates are not used as the first release date." }, { "id": "huggingface/smollm2-135m", "slug": "huggingface--smollm2-135m", "name": "SmolLM2-135M", "developer": "Hugging Face", "developer_id": "huggingface", "family": "SmolLM2", "description": null, "release_date": "2024-10-31", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 0.134515008, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.linkedin.com/posts/loubna-ben-allal-238690152_introducing-smollm2-the-new-best-and-open-activity-7257822657004425216-8NjH", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://www.linkedin.com/posts/loubna-ben-allal-238690152_introducing-smollm2-the-new-best-and-open-activity-7257822657004425216-8NjH", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/HuggingFaceTB/SmolLM2-135M", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/HuggingFaceTB/SmolLM2-135M", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer launch post links the available three model sizes; its datePublished metadata is 2024-10-31T18:36:05Z. Later media coverage dates are not used as the first release date." }, { "id": "huggingface/smollm2-135m-instruct", "slug": "huggingface--smollm2-135m-instruct", "name": "SmolLM2-135M-Instruct", "developer": "Hugging Face", "developer_id": "huggingface", "family": "SmolLM2", "description": null, "release_date": "2024-10-31", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 0.134515008, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.linkedin.com/posts/loubna-ben-allal-238690152_introducing-smollm2-the-new-best-and-open-activity-7257822657004425216-8NjH", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://www.linkedin.com/posts/loubna-ben-allal-238690152_introducing-smollm2-the-new-best-and-open-activity-7257822657004425216-8NjH", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/HuggingFaceTB/SmolLM2-135M-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/HuggingFaceTB/SmolLM2-135M-Instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer launch post links the available three model sizes; its datePublished metadata is 2024-10-31T18:36:05Z. Later media coverage dates are not used as the first release date." }, { "id": "huggingface/smollm2-360m", "slug": "huggingface--smollm2-360m", "name": "SmolLM2-360M", "developer": "Hugging Face", "developer_id": "huggingface", "family": "SmolLM2", "description": null, "release_date": "2024-10-31", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 0.36182112, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.linkedin.com/posts/loubna-ben-allal-238690152_introducing-smollm2-the-new-best-and-open-activity-7257822657004425216-8NjH", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://www.linkedin.com/posts/loubna-ben-allal-238690152_introducing-smollm2-the-new-best-and-open-activity-7257822657004425216-8NjH", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/HuggingFaceTB/SmolLM2-360M", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/HuggingFaceTB/SmolLM2-360M", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer launch post links the available three model sizes; its datePublished metadata is 2024-10-31T18:36:05Z. Later media coverage dates are not used as the first release date." }, { "id": "huggingface/smollm2-360m-instruct", "slug": "huggingface--smollm2-360m-instruct", "name": "SmolLM2-360M-Instruct", "developer": "Hugging Face", "developer_id": "huggingface", "family": "SmolLM2", "description": null, "release_date": "2024-10-31", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 0.36182112, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.linkedin.com/posts/loubna-ben-allal-238690152_introducing-smollm2-the-new-best-and-open-activity-7257822657004425216-8NjH", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://www.linkedin.com/posts/loubna-ben-allal-238690152_introducing-smollm2-the-new-best-and-open-activity-7257822657004425216-8NjH", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/HuggingFaceTB/SmolLM2-360M-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/HuggingFaceTB/SmolLM2-360M-Instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer launch post links the available three model sizes; its datePublished metadata is 2024-10-31T18:36:05Z. Later media coverage dates are not used as the first release date." }, { "id": "cohere/c4ai-aya-expanse-32b", "slug": "cohere--c4ai-aya-expanse-32b", "name": "Aya Expanse 32B", "developer": "Cohere", "developer_id": "cohere", "family": "aya-expanse", "description": "Open-weight multilingual text model supporting 23 languages.", "release_date": "2024-10-24", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-4.0", "parameters_b": 32, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": false, "source_url": "https://huggingface.co/CohereLabs/aya-expanse-32b", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/aya-expanse-32b", "type": "weights" }, { "label": "Aya Expanse: Connecting our world", "url": "https://cohere.com/blog/aya-expanse-connecting-our-world", "type": "primary" }, { "label": "Aya Expanse: Combining Research Breakthroughs for a New Multilingual Frontier", "url": "https://arxiv.org/html/2412.04261v1", "type": "primary" }, { "label": "Public Aya Expanse 32B repository metadata", "url": "https://huggingface.co/api/models/CohereLabs/aya-expanse-32b", "type": "primary" }, { "label": "Cohere model implementation in Transformers 4.46.2", "url": "https://github.com/huggingface/transformers/blob/v4.46.2/src/transformers/models/cohere/modeling_cohere.py", "type": "primary" }, { "label": "An Overview of Cohere models", "url": "https://docs.cohere.com/docs/models", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/cohere/c4ai-aya-expanse-32b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "32B is Cohere's nominal total; active parameters are not separately disclosed. Context is advertised as 128K. The released template provides ordinary text chat without dedicated thinking or native tool-call formatting. The CC-BY-NC-4.0 release includes Cohere Labs' Acceptable Use Addendum.", "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T11:14:51.345340Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/142.json" }, { "id": "cohere/c4ai-aya-expanse-8b", "slug": "cohere--c4ai-aya-expanse-8b", "name": "Aya Expanse 8B", "developer": "Cohere", "developer_id": "cohere", "family": "aya-expanse", "description": "Open-weight multilingual instruction model for text generation in 23 languages.", "release_date": "2024-10-24", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-4.0", "parameters_b": 8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": false, "source_url": "https://huggingface.co/CohereLabs/aya-expanse-8b", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/aya-expanse-8b", "type": "weights" }, { "label": "Aya Expanse: Connecting our world", "url": "https://cohere.com/blog/aya-expanse-connecting-our-world", "type": "primary" }, { "label": "A Deepdive into Aya Expanse: Advancing the Frontier of Multilinguality", "url": "https://huggingface.co/blog/aya-expanse", "type": "primary" }, { "label": "Aya Expanse: Combining Research Breakthroughs for a New Multilingual Frontier", "url": "https://arxiv.org/html/2412.04261v1", "type": "primary" }, { "label": "Official checkpoint public metadata and chat template", "url": "https://huggingface.co/api/models/CohereLabs/aya-expanse-8b", "type": "primary" }, { "label": "CohereForCausalLM implementation, Transformers v4.45.2", "url": "https://raw.githubusercontent.com/huggingface/transformers/v4.45.2/src/transformers/models/cohere/modeling_cohere.py", "type": "primary" }, { "label": "CC-BY-NC 4.0 License with Acceptable Use Addendum", "url": "https://cohere.com/cohere-labs-cc-by-nc-license", "type": "primary" }, { "label": "An Overview of Cohere’s Models", "url": "https://docs.cohere.com/docs/models", "type": "primary" }, { "label": "Aya Vision: Advancing the Frontier of Multilingual Multimodality", "url": "https://arxiv.org/html/2505.08751v1", "type": "primary" }, { "label": "Tiny Aya: Bridging Scale and Multilingual Depth", "url": "https://arxiv.org/html/2603.11510v1", "type": "primary" }, { "label": "Command A: An Enterprise-Ready Large Language Model", "url": "https://arxiv.org/pdf/2504.00698v2", "type": "primary" }, { "label": "Command R7B Arabic: A Small, Enterprise Focused, Multilingual, and Culturally Aware Arabic LLM", "url": "https://arxiv.org/html/2503.14603v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/cohere/c4ai-aya-expanse-8b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "8B is Cohere’s nominal total; Hub metadata lists 8,028,033,024 F16 elements, without an independent learned-tensor census. Cohere has not separately specified an active-parameter count. Context is the documented nominal 8K. The license includes an Acceptable Use Addendum. The released template has no dedicated thinking mode or native tool-call format. Cohere retired the hosted c4ai-aya-expanse-8b endpoint on April 4, 2026; the open-weight release remains documented.", "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T11:15:41.989411+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/143.json" }, { "id": "sarvam/sarvam-1", "slug": "sarvam--sarvam-1", "name": "sarvam-1", "developer": "Sarvam AI", "developer_id": "sarvam", "family": "Sarvam", "description": null, "release_date": "2024-10-24", "date_precision": "day", "open_weights": true, "license": "Sarvam AI Research License", "parameters_b": 2.525087744, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.sarvam.ai/blogs/sarvam-1", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://www.sarvam.ai/blogs/sarvam-1", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/sarvamai/sarvam-1", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/sarvamai/sarvam-1", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/sarvamai/sarvam-1/blob/main/LICENSE.md", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "anthropic/claude-3-5-sonnet-20241022", "slug": "anthropic--claude-3-5-sonnet-20241022", "name": "Claude Sonnet 3.5 v2", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude-sonnet", "description": "October 2024 update of Claude 3.5 Sonnet with text and image understanding, improved coding and tool use, and experimental computer-use support.", "release_date": "2024-10-22", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 200000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://www.anthropic.com/news/3-5-models-and-computer-use", "source_type": "primary", "sources": [ { "label": "Anthropic October 2024 release announcement", "url": "https://www.anthropic.com/news/3-5-models-and-computer-use", "type": "primary" }, { "label": "Anthropic October model-card addendum", "url": "https://www-cdn.anthropic.com/c7822cdc35ad788ec87e14b3a9d45010f1f86c38.pdf", "type": "primary" }, { "label": "Anthropic updated Sonnet SWE-bench research", "url": "https://www.anthropic.com/research/swe-bench-sonnet", "type": "primary" }, { "label": "Anthropic API release notes", "url": "https://platform.claude.com/docs/en/release-notes/overview", "type": "primary" }, { "label": "Anthropic model deprecations", "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "type": "primary" }, { "label": "Anthropic interpretability research on Sonnet 3.5 (new)", "url": "https://transformer-circuits.pub/2025/introspection/index.html", "type": "primary" }, { "label": "Anthropic commercial service terms", "url": "https://www.anthropic.com/legal/commercial-terms", "type": "primary" }, { "label": "Anthropic Claude weight-distribution statement", "url": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities", "type": "primary" }, { "label": "Anthropic later pricing table with conflicting context tier", "url": "https://www-cdn.anthropic.com/files/4zrzovbb/website/5678bc2f5978e5bcd4f1fe7c14b2c72284dcf9f8.pdf", "type": "documentation" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/anthropic/claude-3-5-sonnet-20241022.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The October 2024 revision is distinct from claude-3-5-sonnet-20240620. First-party interpretability research supports a generic Transformer architecture; dense/MoE subtype and nominal total/active parameter counts remain undisclosed. No published size for a named base supports a lineage inference. PDF input was added on November 1, 2024. Reasoning=false denotes no extended-thinking mode. Anthropic retired this API snapshot on October 28, 2025; listed capabilities are historical specifications. A May 12, 2026 pricing PDF lists a conflicting 1M API tier after retirement; retained exact-model specifications support 200K.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T10:17:23.618320Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/091.json" }, { "id": "ibm/granite-3.0-1b-a400m-base", "slug": "ibm--granite-3.0-1b-a400m-base", "name": "granite-3.0-1b-a400m-base", "developer": "IBM", "developer_id": "ibm", "family": "Granite 3.0", "description": null, "release_date": "2024-10-21", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 1.384956928, "active_parameters_b": 0.4, "parameters_source": "checkpoint_metadata", "architecture": "moe", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://canada.newsroom.ibm.com/2024-10-21-ibm-introduces-granite-3-0-high-performing-ai-models-built-for-business", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://canada.newsroom.ibm.com/2024-10-21-ibm-introduces-granite-3-0-high-performing-ai-models-built-for-business", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/ibm-granite/granite-3.0-1b-a400m-base", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/ibm-granite/granite-3.0-1b-a400m-base", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "ibm/granite-3.0-1b-a400m-instruct", "slug": "ibm--granite-3.0-1b-a400m-instruct", "name": "granite-3.0-1b-a400m-instruct", "developer": "IBM", "developer_id": "ibm", "family": "Granite 3.0", "description": null, "release_date": "2024-10-21", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 1.334628352, "active_parameters_b": 0.4, "parameters_source": "checkpoint_metadata", "architecture": "moe", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://canada.newsroom.ibm.com/2024-10-21-ibm-introduces-granite-3-0-high-performing-ai-models-built-for-business", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://canada.newsroom.ibm.com/2024-10-21-ibm-introduces-granite-3-0-high-performing-ai-models-built-for-business", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/ibm-granite/granite-3.0-1b-a400m-instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/ibm-granite/granite-3.0-1b-a400m-instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "ibm/granite-3.0-2b-base", "slug": "ibm--granite-3.0-2b-base", "name": "granite-3.0-2b-base", "developer": "IBM", "developer_id": "ibm", "family": "Granite 3.0", "description": null, "release_date": "2024-10-21", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 2.533525504, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://canada.newsroom.ibm.com/2024-10-21-ibm-introduces-granite-3-0-high-performing-ai-models-built-for-business", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://canada.newsroom.ibm.com/2024-10-21-ibm-introduces-granite-3-0-high-performing-ai-models-built-for-business", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/ibm-granite/granite-3.0-2b-base", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/ibm-granite/granite-3.0-2b-base", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "ibm/granite-3.0-2b-instruct", "slug": "ibm--granite-3.0-2b-instruct", "name": "granite-3.0-2b-instruct", "developer": "IBM", "developer_id": "ibm", "family": "Granite 3.0", "description": null, "release_date": "2024-10-21", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 2.634201088, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://canada.newsroom.ibm.com/2024-10-21-ibm-introduces-granite-3-0-high-performing-ai-models-built-for-business", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://canada.newsroom.ibm.com/2024-10-21-ibm-introduces-granite-3-0-high-performing-ai-models-built-for-business", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/ibm-granite/granite-3.0-2b-instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/ibm-granite/granite-3.0-2b-instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "ibm/granite-3.0-3b-a800m-base", "slug": "ibm--granite-3.0-3b-a800m-base", "name": "granite-3.0-3b-a800m-base", "developer": "IBM", "developer_id": "ibm", "family": "Granite 3.0", "description": null, "release_date": "2024-10-21", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 3.374286336, "active_parameters_b": 0.8, "parameters_source": "checkpoint_metadata", "architecture": "moe", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://canada.newsroom.ibm.com/2024-10-21-ibm-introduces-granite-3-0-high-performing-ai-models-built-for-business", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://canada.newsroom.ibm.com/2024-10-21-ibm-introduces-granite-3-0-high-performing-ai-models-built-for-business", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/ibm-granite/granite-3.0-3b-a800m-base", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/ibm-granite/granite-3.0-3b-a800m-base", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "ibm/granite-3.0-3b-a800m-instruct", "slug": "ibm--granite-3.0-3b-a800m-instruct", "name": "granite-3.0-3b-a800m-instruct", "developer": "IBM", "developer_id": "ibm", "family": "Granite 3.0", "description": null, "release_date": "2024-10-21", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 3.374295552, "active_parameters_b": 0.8, "parameters_source": "checkpoint_metadata", "architecture": "moe", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://canada.newsroom.ibm.com/2024-10-21-ibm-introduces-granite-3-0-high-performing-ai-models-built-for-business", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://canada.newsroom.ibm.com/2024-10-21-ibm-introduces-granite-3-0-high-performing-ai-models-built-for-business", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/ibm-granite/granite-3.0-3b-a800m-instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/ibm-granite/granite-3.0-3b-a800m-instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "ibm/granite-3.0-8b-base", "slug": "ibm--granite-3.0-8b-base", "name": "granite-3.0-8b-base", "developer": "IBM", "developer_id": "ibm", "family": "Granite 3.0", "description": null, "release_date": "2024-10-21", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 8.170835968, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://canada.newsroom.ibm.com/2024-10-21-ibm-introduces-granite-3-0-high-performing-ai-models-built-for-business", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://canada.newsroom.ibm.com/2024-10-21-ibm-introduces-granite-3-0-high-performing-ai-models-built-for-business", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/ibm-granite/granite-3.0-8b-base", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/ibm-granite/granite-3.0-8b-base", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "ibm/granite-3.0-8b-instruct", "slug": "ibm--granite-3.0-8b-instruct", "name": "granite-3.0-8b-instruct", "developer": "IBM", "developer_id": "ibm", "family": "Granite 3.0", "description": null, "release_date": "2024-10-21", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 8.170848256, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://canada.newsroom.ibm.com/2024-10-21-ibm-introduces-granite-3-0-high-performing-ai-models-built-for-business", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://canada.newsroom.ibm.com/2024-10-21-ibm-introduces-granite-3-0-high-performing-ai-models-built-for-business", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/ibm-granite/granite-3.0-8b-instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/ibm-granite/granite-3.0-8b-instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "mistral/ministral-3b", "slug": "mistral--ministral-3b", "name": "Ministral 3B", "developer": "Mistral", "developer_id": "mistral", "family": "ministral", "description": "Text model for on-device assistants and function-calling workflows, offered under a commercial license.", "release_date": "2024-10-16", "date_precision": "day", "open_weights": false, "license": "Mistral Commercial License", "parameters_b": 3, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": null, "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://mistral.ai/news/ministraux/", "source_type": "primary", "sources": [ { "label": "Un Ministral, des Ministraux", "url": "https://mistral.ai/news/ministraux/", "type": "primary" }, { "label": "Ministral 3B — exact legacy model reference", "url": "https://docs.mistral.ai/models/ministral-3b-24-1.md", "type": "primary" }, { "label": "API changelog — original Ministral release", "url": "https://docs.mistral.ai/resources/changelogs", "type": "primary" }, { "label": "Ministral 1 - 3B — model lifecycle", "url": "https://legal.mistral.ai/ai-governance/models/ministral-1-3b/", "type": "primary" }, { "label": "Ministral-3B — pinned Qualcomm model card before deprecation", "url": "https://huggingface.co/qualcomm/Ministral-3B/blob/5ba3b0827cc78070e4607d06df805d5398c72661/README.md", "type": "primary" }, { "label": "Ministral-3B implementation metadata v0.40.0", "url": "https://raw.githubusercontent.com/quic/ai-hub-models/v0.40.0/qai_hub_models/models/ministral_3b/info.yaml", "type": "primary" }, { "label": "Qualcomm and Mistral AI partner on original Ministral models", "url": "https://www.qualcomm.com/news/releases/2024/10/qualcomm-and-mistral-ai-partner-to-bring-new-generative-ai-model", "type": "primary" }, { "label": "Phi-4-Mini technical report — original Ministral comparison check", "url": "https://arxiv.org/html/2503.01743v1", "type": "primary" }, { "label": "Phi-4 Microsoft Research January 2025 brief — active-parameter plot check", "url": "https://www.microsoft.com/en-us/research/wp-content/uploads/2025/01/WEF-2025_Leave-Behind_Phi-4.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/ministral-3b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Original October 2024 Ministral 3B (ministral-3b-2410), distinct from the December 2025 multimodal successor. Nominal 3B is supported by Mistral’s model name and Qualcomm’s explicit implementation specification; active parameters and internal architecture remain undisclosed. Weights require a commercial agreement. Public launch is October 16; retrospective developer records give October 9, an unresolved date conflict. Context is nominal 128k; dedicated reasoning mode is unverified. Legacy docs record December 2025 API retirement.", "field_status": { "active_parameters_b": "undisclosed", "architecture": "undisclosed", "reasoning": "unverified" }, "audited_at": "2026-10-27T15:32:17.547228Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/325.json" }, { "id": "mistral/ministral-8b-instruct-2410", "slug": "mistral--ministral-8b-instruct-2410", "name": "Ministral 8B Instruct 2410", "developer": "Mistral", "developer_id": "mistral", "family": "ministral", "description": "Instruction-tuned text model with interleaved sliding-window attention and native function calling.", "release_date": "2024-10-16", "date_precision": "day", "open_weights": true, "license": "Mistral Research License", "parameters_b": 8.019808256, "active_parameters_b": 8, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://mistral.ai/news/ministraux/", "source_type": "primary", "sources": [ { "label": "Un Ministral, des Ministraux", "url": "https://mistral.ai/news/ministraux/", "type": "primary" }, { "label": "Ministral-8B-Instruct-2410 model card at revision 2f494a194c5b980dfb9772cb92d26cbb671fce5a", "url": "https://huggingface.co/mistralai/Ministral-8B-Instruct-2410/blob/2f494a194c5b980dfb9772cb92d26cbb671fce5a/README.md", "type": "weights" }, { "label": "Ministral 8B legacy model reference", "url": "https://docs.mistral.ai/models/ministral-8b-24-1", "type": "primary" }, { "label": "Ministral 8B model reference, Markdown", "url": "https://docs.mistral.ai/models/ministral-8b-24-1.md", "type": "primary" }, { "label": "Ministral-8B-Instruct-2410 Transformers configuration", "url": "https://huggingface.co/mistralai/Ministral-8B-Instruct-2410/blob/2f494a194c5b980dfb9772cb92d26cbb671fce5a/config.json", "type": "weights" }, { "label": "Ministral-8B-Instruct-2410 native inference parameters", "url": "https://huggingface.co/mistralai/Ministral-8B-Instruct-2410/blob/2f494a194c5b980dfb9772cb92d26cbb671fce5a/params.json", "type": "weights" }, { "label": "Mistral AI Research License 0.1", "url": "https://mistral.ai/static/licenses/MRL-0.1.md", "type": "primary" }, { "label": "Mistral API changelog", "url": "https://docs.mistral.ai/resources/changelogs", "type": "primary" }, { "label": "Official checkpoint repository metadata", "url": "https://huggingface.co/api/models/mistralai/Ministral-8B-Instruct-2410", "type": "weights" }, { "label": "Original checkpoint weight shard, bounded public access probe", "url": "https://huggingface.co/mistralai/Ministral-8B-Instruct-2410/resolve/2f494a194c5b980dfb9772cb92d26cbb671fce5a/model-00001-of-00004.safetensors", "type": "weights" }, { "label": "Ragged attention supported in vLLM", "url": "https://huggingface.co/mistralai/Ministral-8B-Instruct-2410/discussions/18", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/ministral-8b-instruct-2410.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Ministral-8B-Instruct-2410", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Original October 2024 instruction checkpoint; distinct from Ministral 3 8B released in December 2025. Contemporaneous Mistral announcement dated October 16, 2024 says available starting today. Retrospective model documentation and API changelog instead give October 9; the public launch date is retained with this conflict recorded. Exact total count: 8,019,808,256 parameters. The official legacy Weights table separately discloses a rounded 8B active count; it is not an exact active-count derivation. Public weights use Mistral Research License 0.1 with research-use restrictions. Commercial use requires a separate license. Native maximum context is 131072 tokens with attention pattern full/32768/32768/32768. The Transformers configuration sets 32768 maximum positions; the card describes an older framework limitation, while a later maintainer discussion reports vLLM ragged-attention support. Dedicated native reasoning/thinking support could not be verified; ordinary reasoning ability does not establish a separate reasoning mode. The legacy API reference lists retirement on December 31, 2025. This does not remove the original public checkpoint or establish that any historical API endpoint currently operates.", "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-26T23:07:54.072532Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/328.json" }, { "id": "google/gemma-2-2b-jpn-it", "slug": "google--gemma-2-2b-jpn-it", "name": "gemma-2-2b-jpn-it", "developer": "Google", "developer_id": "google", "family": "Gemma 2", "description": null, "release_date": "2024-10-03", "date_precision": "day", "open_weights": true, "license": "gemma", "parameters_b": 2.614341888, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.google.dev/gemma/docs/releases", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://ai.google.dev/gemma/docs/releases", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/google/gemma-2-2b-jpn-it", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/google/gemma-2-2b-jpn-it", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "nvidia/llama-3.1-nemotron-70b-instruct", "slug": "nvidia--llama-3.1-nemotron-70b-instruct", "name": "Llama 3.1 Nemotron 70B Instruct", "developer": "NVIDIA", "developer_id": "nvidia", "family": "nemotron", "description": "NVIDIA’s RLHF-tuned Llama 3.1 70B Instruct model, trained with REINFORCE and HelpSteer2 preference data to improve helpfulness in general instruction following.", "release_date": "2024-10", "date_precision": "month", "open_weights": true, "license": "NVIDIA Open Model License + Llama 3.1 Community License", "parameters_b": 70.553706496, "active_parameters_b": 70, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/nvidia/llama-3.1-nemotron-70b-instruct.toml", "source_type": "catalog", "sources": [ { "label": "Original NVIDIA NeMo model card", "url": "https://huggingface.co/nvidia/Llama-3.1-Nemotron-70B-Instruct/blob/a83af1f4968437064635f6726fb745e5b615e863/README.md", "type": "primary" }, { "label": "NVIDIA Transformers conversion model card", "url": "https://huggingface.co/nvidia/Llama-3.1-Nemotron-70B-Instruct-HF/blob/031d4042f36adc1a52cca51b331d25cbe3cf1022/README.md", "type": "primary" }, { "label": "Official HF checkpoint configuration", "url": "https://huggingface.co/nvidia/Llama-3.1-Nemotron-70B-Instruct-HF/resolve/031d4042f36adc1a52cca51b331d25cbe3cf1022/config.json", "type": "weights" }, { "label": "Official checkpoint tensor index", "url": "https://huggingface.co/nvidia/Llama-3.1-Nemotron-70B-Instruct-HF/resolve/031d4042f36adc1a52cca51b331d25cbe3cf1022/model.safetensors.index.json", "type": "weights" }, { "label": "Official safetensors headers (first of 30 indexed shards)", "url": "https://huggingface.co/nvidia/Llama-3.1-Nemotron-70B-Instruct-HF/resolve/031d4042f36adc1a52cca51b331d25cbe3cf1022/model-00001-of-00030.safetensors", "type": "weights" }, { "label": "Original NeMo configuration", "url": "https://huggingface.co/nvidia/Llama-3.1-Nemotron-70B-Instruct/resolve/a83af1f4968437064635f6726fb745e5b615e863/model_config.yaml", "type": "weights" }, { "label": "Original NeMo weight metadata", "url": "https://huggingface.co/nvidia/Llama-3.1-Nemotron-70B-Instruct/resolve/a83af1f4968437064635f6726fb745e5b615e863/model_weights/model.embedding.word_embeddings.weight/.zarray", "type": "weights" }, { "label": "NVIDIA alignment-model announcement", "url": "https://developer.nvidia.com/blog/new-reward-model-helps-improve-llm-alignment-with-human-preferences/", "type": "primary" }, { "label": "HelpSteer2-Preference revised technical report", "url": "https://arxiv.org/html/2410.01257v2", "type": "primary" }, { "label": "Exact-model NVIDIA NIM system card", "url": "https://build.nvidia.com/nvidia/llama-3_1-nemotron-70b-instruct/systemcard", "type": "primary" }, { "label": "Official checkpoint tokenizer and chat template", "url": "https://huggingface.co/nvidia/Llama-3.1-Nemotron-70B-Instruct-HF/resolve/031d4042f36adc1a52cca51b331d25cbe3cf1022/tokenizer_config.json", "type": "weights" }, { "label": "NVIDIA Open Model License Agreement", "url": "https://developer.download.nvidia.com/licenses/nvidia-open-model-license-agreement-june-2024.pdf", "type": "primary" }, { "label": "Llama 3.1 Community License Agreement", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/LICENSE", "type": "primary" }, { "label": "AI21 Jamba-1.5 original ICLR 2025 paper, Table 1", "url": "https://proceedings.iclr.cc/paper_files/paper/2025/file/a9ed43fa31dc8b4a7d7a673d713dcb5f-Paper-Conference.pdf", "type": "primary" }, { "label": "Meta Llama 3.1 model card", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/MODEL_CARD.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/nvidia/llama-3.1-nemotron-70b-instruct.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Total is the exact 70.553706496B complete BF16 NVIDIA Transformers conversion. The original NeMo checkpoint stores 70.566289408B elements, including 12,582,912 extra vocabulary-padding elements; 70B is nominal. Active is a nominal 70B lineage inference from AI21’s Jamba comparison convention for Llama 3.1 70B and NVIDIA’s documented REINFORCE post-training of Llama-3.1-70B-Instruct; NVIDIA does not separately disclose an exact active count. Release is established to October 2024, with the day unresolved. Context is 131072 tokens in both configurations. NVIDIA NIM marks tool calling unsupported, while the checkpoint retains inherited Llama function-call syntax; capability for the fine-tune remains unverified. Dedicated reasoning support is also unverified.", "field_status": { "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T15:49:48.547900Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/374.json" }, { "id": "meta/llama-3.2-11b-vision-instruct", "slug": "meta--llama-3.2-11b-vision-instruct", "name": "Llama-3.2-11B-Vision-Instruct", "developer": "Meta", "developer_id": "meta", "family": "llama", "description": "Meta instruction-tuned multimodal Llama 3.2 model for image understanding, captioning, visual question answering, and text generation.", "release_date": "2024-09-25", "date_precision": "day", "open_weights": true, "license": "Llama 3.2 Community License", "parameters_b": 10.6, "active_parameters_b": 10.6, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://huggingface.co/meta-llama/Llama-3.2-11B-Vision-Instruct", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/meta-llama/Llama-3.2-11B-Vision-Instruct", "type": "weights" }, { "label": "Llama 3.2 Vision model card", "url": "https://raw.githubusercontent.com/meta-llama/llama-models/main/models/llama3_2/MODEL_CARD_VISION.md", "type": "primary" }, { "label": "Llama 3.2: Revolutionizing edge AI and vision with open, customizable models", "url": "https://ai.meta.com/blog/llama-3-2-connect-2024-vision-edge-mobile-devices/", "type": "primary" }, { "label": "Llama 3.2 Community License Agreement", "url": "https://raw.githubusercontent.com/meta-llama/llama-models/main/models/llama3_2/LICENSE", "type": "primary" }, { "label": "Llama 3.2 Acceptable Use Policy", "url": "https://raw.githubusercontent.com/meta-llama/llama-models/main/models/llama3_2/USE_POLICY.md", "type": "primary" }, { "label": "Llama 3.2 vision prompt format", "url": "https://raw.githubusercontent.com/meta-llama/llama-models/main/models/llama3_2/vision_prompt_format.md", "type": "primary" }, { "label": "Official model SKU definitions: Llama3.2-11B-Vision-Instruct", "url": "https://raw.githubusercontent.com/meta-llama/llama-models/main/models/sku_list.py", "type": "primary" }, { "label": "Official model families and maximum sequence lengths", "url": "https://raw.githubusercontent.com/meta-llama/llama-models/main/models/sku_types.py", "type": "primary" }, { "label": "Official cross-attention multimodal Transformer implementation", "url": "https://raw.githubusercontent.com/meta-llama/llama-models/main/models/llama3/multimodal/model.py", "type": "primary" }, { "label": "Kimi-VL Technical Report, Table 3: active parameters of comparison VLMs", "url": "https://arxiv.org/html/2504.07491v3#S3.T3", "type": "paper" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/meta/llama-3.2-11b-vision-instruct.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Meta reports nominal 10.6B total (11B label). Nominal 10.6B active follows the Kimi team's explicit 8B+2.6B LLM+VT convention in Kimi-VL Table 3; it is not a Meta active-count disclosure or an exact per-token measurement. Dense language and vision Transformers use eight cross-attention adapter blocks. Native tool calling is documented only for text-only prompts, not prompts containing images. A dedicated thinking mode is undisclosed. The Llama 3.2 license incorporates multimodal EU restrictions, with an exception for end users of incorporating products or services. The official 128k context is 131072 tokens.", "field_status": { "reasoning": "undisclosed" }, "audited_at": "2026-10-27T14:10:35.082323Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/270.json" }, { "id": "meta/llama-3.2-1b", "slug": "meta--llama-3.2-1b", "name": "Llama 3.2 1B", "developer": "Meta", "developer_id": "meta", "family": "llama-3.2", "description": "Pretrained multilingual Llama 3.2 text model with 1.23B parameters, shared embeddings and 128K context, developed through pruning and distillation.", "release_date": "2024-09-25", "date_precision": "day", "open_weights": true, "license": "Llama 3.2 Community License", "parameters_b": 1.23, "active_parameters_b": 1.23, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/meta-llama/Llama-3.2-1B", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/meta-llama/Llama-3.2-1B", "type": "weights" }, { "label": "Llama 3.2 launch announcement, September 25, 2024", "url": "https://ai.meta.com/blog/llama-3-2-connect-2024-vision-edge-mobile-devices/", "type": "primary" }, { "label": "Meta model registry: exact Llama 3.2 1B base SKU", "url": "https://github.com/meta-llama/llama-models/blob/main/models/sku_list.py", "type": "primary" }, { "label": "Meta model registry: precision defaults and context lengths", "url": "https://github.com/meta-llama/llama-models/blob/main/models/sku_types.py", "type": "primary" }, { "label": "Meta Llama reference Transformer implementation", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3/model.py", "type": "primary" }, { "label": "Llama 3.2 exact prompt formats", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_2/text_prompt_format.md", "type": "primary" }, { "label": "Llama 3.2 Community License Agreement", "url": "https://huggingface.co/meta-llama/Llama-3.2-1B/blob/main/LICENSE.txt", "type": "primary" }, { "label": "Current combined Llama 3.2 model card", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_2/MODEL_CARD.md", "type": "primary" }, { "label": "Meta Llama 3.2 model and quantization documentation", "url": "https://dev.meta.ai/llama/docs/model-cards-and-prompt-formats/llama3_2", "type": "primary" }, { "label": "Mikhaylovskiy: Llama-3.2-1B nominal active count, Table 3", "url": "https://aclanthology.org/2025.findings-emnlp.837.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/meta/llama-3.2-1b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Meta reports a nominal 1.23B total with shared embeddings. The 1.23B active count follows Mikhaylovskiy’s EMNLP 2025 Table 3 convention for the original base model; it is not a separate Meta disclosure or exact per-token tensor count. A dedicated reasoning mode remains undisclosed. Native function calling is documented for the Instruct sibling, not established for this base checkpoint. Original BF16 context is 131072; later INT4 Instruct context is 8192. The September 25, 2024 base release predates the October 24 quantized update.", "field_status": { "reasoning": "undisclosed", "tool_call": "undisclosed" }, "audited_at": "2026-10-27T14:10:19.710Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/271.json" }, { "id": "meta/llama-3.2-1b-instruct", "slug": "meta--llama-3.2-1b-instruct", "name": "Llama 3.2 1B Instruct", "developer": "Meta", "developer_id": "meta", "family": "llama", "description": "Compact instruction-tuned Llama model for multilingual dialogue and on-device assistants", "release_date": "2024-09-25", "date_precision": "day", "open_weights": true, "license": "Llama 3.2 Community License", "parameters_b": 1.23, "active_parameters_b": 1.23, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://huggingface.co/meta-llama/Llama-3.2-1B-Instruct", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/meta-llama/Llama-3.2-1B-Instruct", "type": "weights" }, { "label": "Meta BF16 launch", "url": "https://ai.meta.com/blog/llama-3-2-connect-2024-vision-edge-mobile-devices/", "type": "primary" }, { "label": "Meta model SKU configuration", "url": "https://github.com/meta-llama/llama-models/blob/main/models/sku_list.py", "type": "primary" }, { "label": "Meta context-limit implementation", "url": "https://github.com/meta-llama/llama-models/blob/main/models/sku_types.py", "type": "primary" }, { "label": "Meta dense Transformer implementation", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3/model.py", "type": "primary" }, { "label": "Meta 1B/3B tool-calling documentation", "url": "https://dev.meta.ai/llama/docs/model-cards-and-prompt-formats/llama3_2", "type": "primary" }, { "label": "Exact checkpoint public metadata", "url": "https://huggingface.co/api/models/meta-llama/Llama-3.2-1B-Instruct", "type": "weights" }, { "label": "Llama 3.2 license", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_2/LICENSE", "type": "primary" }, { "label": "Meta reference pythonic function-call format", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_2/text_prompt_format.md", "type": "primary" }, { "label": "Mikhaylovskiy nominal active-parameter table", "url": "https://aclanthology.org/2025.findings-emnlp.837.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/meta/llama-3.2-1b-instruct.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Original BF16 Instruct checkpoint. Meta reports approximately 1.23B total parameters with shared embeddings. Active 1.23B follows Mikhaylovskiy’s published nominal convention for the original base/Instruct models, not an exact Meta per-token count. Dense Transformer with GQA, produced through structured pruning and distillation. Context is 131,072 tokens; later official INT4 variants use 8,192. Native custom function calling is documented for 1B; dedicated reasoning-mode support is undisclosed. Meta reference pythonic calls can contain multiple functions; the Hub JSON template handles one call per message.", "field_status": { "reasoning": "undisclosed" }, "audited_at": "2026-10-27T14:11:05.513740Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/272.json" }, { "id": "meta/llama-3.2-3b", "slug": "meta--llama-3.2-3b", "name": "Llama-3.2-3B", "developer": "Meta", "developer_id": "meta", "family": "llama", "description": "Pretrained Llama 3.2 text model with a reported 3.21B parameters, shared embeddings and a 131,072-token context.", "release_date": "2024-09-25", "date_precision": "day", "open_weights": true, "license": "Llama 3.2 Community License", "parameters_b": 3.21, "active_parameters_b": 3.21, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/meta-llama/Llama-3.2-3B", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/meta-llama/Llama-3.2-3B", "type": "weights" }, { "label": "Meta original BF16 release", "url": "https://ai.meta.com/blog/llama-3-2-connect-2024-vision-edge-mobile-devices/", "type": "primary" }, { "label": "Meta exact model configuration", "url": "https://github.com/meta-llama/llama-models/blob/0e0b8c519242d5833d8c11bffc1232b77ad7f301/models/sku_list.py", "type": "primary" }, { "label": "Meta precision and context limits", "url": "https://github.com/meta-llama/llama-models/blob/0e0b8c519242d5833d8c11bffc1232b77ad7f301/models/sku_types.py", "type": "primary" }, { "label": "Meta dense Transformer implementation", "url": "https://github.com/meta-llama/llama-models/blob/0e0b8c519242d5833d8c11bffc1232b77ad7f301/models/llama3/model.py", "type": "primary" }, { "label": "Meta Llama 3.2 function-call scope", "url": "https://github.com/meta-llama/llama-models/blob/0e0b8c519242d5833d8c11bffc1232b77ad7f301/models/llama3_2/text_prompt_format.md", "type": "primary" }, { "label": "Meta quantized variant documentation", "url": "https://dev.meta.ai/llama/docs/model-cards-and-prompt-formats/llama3_2", "type": "primary" }, { "label": "Llama 3.2 Community License", "url": "https://github.com/meta-llama/llama-models/blob/0e0b8c519242d5833d8c11bffc1232b77ad7f301/models/llama3_2/LICENSE", "type": "primary" }, { "label": "Meta current Llama 3.2 nominal parameter table", "url": "https://raw.githubusercontent.com/meta-llama/llama-models/main/models/llama3_2/MODEL_CARD.md", "type": "primary" }, { "label": "Mikhaylovskiy EMNLP 2025 explicit active-count table", "url": "https://aclanthology.org/2025.findings-emnlp.837.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/meta/llama-3.2-3b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Original BF16 base checkpoint. Meta reports a nominal 3.21B total with shared embeddings and dense GQA blocks. Mikhaylovskiy’s EMNLP 2025 study lists 3.21B active for Llama-3.2-3B; this is the researcher’s nominal convention, not a Meta disclosure or exact per-token count. Developed by pruning and distillation from Llama 3.1. Context is 131072; later quantized Instruct variants use 8192. Dedicated reasoning and native base tool calling are not disclosed.", "field_status": { "reasoning": "undisclosed", "tool_call": "undisclosed" }, "audited_at": "2026-10-27T14:12:59.002274Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/273.json" }, { "id": "meta/llama-3.2-3b-instruct", "slug": "meta--llama-3.2-3b-instruct", "name": "Llama 3.2 3B Instruct", "developer": "Meta", "developer_id": "meta", "family": "llama", "description": "Small instruction-tuned Llama model for multilingual dialogue and agentic retrieval", "release_date": "2024-09-25", "date_precision": "day", "open_weights": true, "license": "Llama 3.2 Community License", "parameters_b": 3.21, "active_parameters_b": 3.21, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct", "type": "weights" }, { "label": "Llama 3.2 launch announcement", "url": "https://ai.meta.com/blog/llama-3-2-connect-2024-vision-edge-mobile-devices/", "type": "primary" }, { "label": "Official Llama model SKU definitions", "url": "https://github.com/meta-llama/llama-models/blob/main/models/sku_list.py", "type": "primary" }, { "label": "Official model types and context limits", "url": "https://github.com/meta-llama/llama-models/blob/main/models/sku_types.py", "type": "primary" }, { "label": "Official Llama Transformer implementation", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3/model.py", "type": "primary" }, { "label": "Llama 3.2 text prompt format", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_2/text_prompt_format.md", "type": "primary" }, { "label": "Introducing quantized Llama models", "url": "https://ai.meta.com/blog/meta-llama-quantized-lightweight-models/", "type": "primary" }, { "label": "Shared Llama 3.2 model card updated for quantization", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_2/MODEL_CARD.md", "type": "primary" }, { "label": "Llama 3.2 Community License Agreement", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_2/LICENSE", "type": "primary" }, { "label": "Mikhaylovskiy: Llama 3.2 nominal active-count convention (Table 3)", "url": "https://aclanthology.org/2025.findings-emnlp.837.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/meta/llama-3.2-3b-instruct.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Original BF16 Instruct model. Meta publishes a nominal 3.21B total with shared embeddings. The 3.21B active count follows Mikhaylovskiy’s research convention (Table 3; original Instruct coverage in sections 4.1/4.4 and Figure 10), not a Meta disclosure or exact per-token tensor census. Dense Transformer with GQA. Original context is 131,072 tokens; later official INT4 variants use 8,192. September 25, 2024 is the original release; October 24 is the quantized update. Custom function calling is documented; dedicated reasoning-mode support is undisclosed.", "field_status": { "reasoning": "undisclosed" }, "audited_at": "2026-10-27T14:13:34.608Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/274.json" }, { "id": "meta/llama-3.2-90b-vision-instruct", "slug": "meta--llama-3.2-90b-vision-instruct", "name": "Llama-3.2-90B-Vision-Instruct", "developer": "Meta", "developer_id": "meta", "family": "llama", "description": "Instruction-tuned dense vision-language model with 88.8B reported parameters, supporting visual question answering, document understanding, captioning, and text chat.", "release_date": "2024-09-25", "date_precision": "day", "open_weights": true, "license": "Llama 3.2 Community License", "parameters_b": 88.8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://huggingface.co/meta-llama/Llama-3.2-90B-Vision-Instruct", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/meta-llama/Llama-3.2-90B-Vision-Instruct", "type": "weights" }, { "label": "Llama 3.2 release announcement", "url": "https://ai.meta.com/blog/llama-3-2-connect-2024-vision-edge-mobile-devices/", "type": "primary" }, { "label": "Official exact model architecture registry", "url": "https://github.com/meta-llama/llama-models/blob/main/models/sku_list.py", "type": "primary" }, { "label": "Official Llama model families and context limits", "url": "https://github.com/meta-llama/llama-models/blob/main/models/sku_types.py", "type": "primary" }, { "label": "Official Llama multimodal Transformer implementation", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3/multimodal/model.py", "type": "primary" }, { "label": "Official Llama 3.2 Vision prompt and tool format", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_2/vision_prompt_format.md", "type": "primary" }, { "label": "Llama 3.2 Community License Agreement", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_2/LICENSE", "type": "primary" }, { "label": "Llama 3.2 Acceptable Use Policy", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_2/USE_POLICY.md", "type": "primary" }, { "label": "Official Llama 3.2 Vision model card, fresh parameter check", "url": "https://raw.githubusercontent.com/meta-llama/llama-models/main/models/llama3_2/MODEL_CARD_VISION.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/meta/llama-3.2-90b-vision-instruct.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Meta reports nominal 88.8B complete multimodal parameters (90B tier); a separate active count remains undisclosed. Architecture combines 80 language Transformer layers with 20 vision cross-attention layers. Context is 131072 tokens. Native tool calling applies only to text-only prompts; image prompts are excluded. The custom license incorporates the AUP’s EU multimodal restriction and end-user exception. A dedicated reasoning mode is not separately disclosed.", "field_status": { "active_parameters_b": "undisclosed", "reasoning": "undisclosed" }, "audited_at": "2026-10-27T14:15:52.116303Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/275.json" }, { "id": "alibaba/qwen2.5-0.5b", "slug": "alibaba--qwen2.5-0.5b", "name": "Qwen2.5-0.5B", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen2.5", "description": null, "release_date": "2024-09-19", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 0.494032768, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen2.5/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen2.5/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen2.5-0.5B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2.5-0.5B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "alibaba/qwen2.5-0.5b-instruct", "slug": "alibaba--qwen2.5-0.5b-instruct", "name": "Qwen2.5-0.5B-Instruct", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen2.5", "description": null, "release_date": "2024-09-19", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 0.494032768, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen2.5/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen2.5/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2.5-0.5B-Instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "alibaba/qwen2.5-1.5b", "slug": "alibaba--qwen2.5-1.5b", "name": "Qwen2.5-1.5B", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen2.5", "description": null, "release_date": "2024-09-19", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 1.543714304, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen2.5/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen2.5/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen2.5-1.5B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2.5-1.5B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "alibaba/qwen2.5-1.5b-instruct", "slug": "alibaba--qwen2.5-1.5b-instruct", "name": "Qwen2.5-1.5B-Instruct", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen2.5", "description": null, "release_date": "2024-09-19", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 1.543714304, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen2.5/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen2.5/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen2.5-1.5B-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2.5-1.5B-Instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "alibaba/qwen2.5-14b", "slug": "alibaba--qwen2.5-14b", "name": "Qwen2.5-14B", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen2.5", "description": null, "release_date": "2024-09-19", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 14.770033664, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen2.5/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen2.5/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen2.5-14B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2.5-14B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "alibaba/qwen2.5-14b-instruct", "slug": "alibaba--qwen2.5-14b-instruct", "name": "Qwen2.5-14B-Instruct", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen2.5", "description": null, "release_date": "2024-09-19", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 14.770033664, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen2.5/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen2.5/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen2.5-14B-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2.5-14B-Instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "alibaba/qwen2.5-32b", "slug": "alibaba--qwen2.5-32b", "name": "Qwen2.5-32B", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen2.5", "description": null, "release_date": "2024-09-19", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 32.763876352, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen2.5/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen2.5/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen2.5-32B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2.5-32B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "alibaba/qwen2.5-32b-instruct", "slug": "alibaba--qwen2.5-32b-instruct", "name": "Qwen2.5-32B-Instruct", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen2.5", "description": null, "release_date": "2024-09-19", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 32.763876352, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen2.5/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen2.5/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen2.5-32B-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2.5-32B-Instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "alibaba/qwen2.5-3b", "slug": "alibaba--qwen2.5-3b", "name": "Qwen2.5-3B", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen2.5", "description": null, "release_date": "2024-09-19", "date_precision": "day", "open_weights": true, "license": "Qwen RESEARCH LICENSE AGREEMENT", "parameters_b": 3.085938688, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen2.5/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen2.5/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen2.5-3B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2.5-3B", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/Qwen/Qwen2.5-3B/blob/main/LICENSE", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "alibaba/qwen2.5-3b-instruct", "slug": "alibaba--qwen2.5-3b-instruct", "name": "Qwen2.5-3B-Instruct", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen2.5", "description": null, "release_date": "2024-09-19", "date_precision": "day", "open_weights": true, "license": "Qwen RESEARCH LICENSE AGREEMENT", "parameters_b": 3.085938688, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen2.5/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen2.5/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen2.5-3B-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2.5-3B-Instruct", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/Qwen/Qwen2.5-3B-Instruct/blob/main/LICENSE", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "alibaba/qwen2.5-72b", "slug": "alibaba--qwen2.5-72b", "name": "Qwen2.5-72B", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen2.5", "description": null, "release_date": "2024-09-19", "date_precision": "day", "open_weights": true, "license": "Qwen LICENSE AGREEMENT", "parameters_b": 72.706203648, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen2.5/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen2.5/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen2.5-72B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2.5-72B", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/Qwen/Qwen2.5-72B/blob/main/LICENSE", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "alibaba/qwen2.5-72b-instruct", "slug": "alibaba--qwen2.5-72b-instruct", "name": "Qwen2.5-72B-Instruct", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen2.5", "description": null, "release_date": "2024-09-19", "date_precision": "day", "open_weights": true, "license": "Qwen LICENSE AGREEMENT", "parameters_b": 72.706203648, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen2.5/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen2.5/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen2.5-72B-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2.5-72B-Instruct", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/Qwen/Qwen2.5-72B-Instruct/blob/main/LICENSE", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "alibaba/qwen2.5-7b", "slug": "alibaba--qwen2.5-7b", "name": "Qwen2.5-7B", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen2.5", "description": null, "release_date": "2024-09-19", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 7.615616512, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen2.5/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen2.5/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen2.5-7B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2.5-7B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "alibaba/qwen2.5-7b-instruct", "slug": "alibaba--qwen2.5-7b-instruct", "name": "Qwen2.5 7B Instruct", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Open-weight 7.61B-parameter dense Qwen2.5 instruction model for text chat, structured output, and tool use", "release_date": "2024-09-19", "date_precision": "day", "open_weights": true, "license": "Apache 2.0", "parameters_b": 7.61, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://qwenlm.github.io/blog/qwen2.5-llm/", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/Qwen/Qwen2.5-7B-Instruct", "type": "weights" }, { "label": "Qwen2.5-7B-Instruct configuration", "url": "https://huggingface.co/Qwen/Qwen2.5-7B-Instruct/blob/main/config.json", "type": "config" }, { "label": "Qwen2.5 release announcement", "url": "https://qwenlm.github.io/blog/qwen2.5/", "type": "announcement" }, { "label": "Qwen2.5-7B-Instruct tool-call template", "url": "https://huggingface.co/Qwen/Qwen2.5-7B-Instruct/blob/main/tokenizer_config.json", "type": "config" }, { "label": "Qwen2.5 detailed release post", "url": "https://qwenlm.github.io/blog/qwen2.5-llm/", "type": "announcement" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen2.5-7b-instruct.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "7.61B is the developer-reported rounded total; 6.53B is non-embedding, not active parameters. Active count is not separately defined for this dense model. Maximum context is 131,072 with YaRN; the shipped configuration defaults to 32,768. Reasoning=false refers to absence of a dedicated thinking mode, not absence of general reasoning ability.", "field_status": { "active_parameters_b": "not_applicable" }, "audited_at": "2026-10-26T14:08:57.313895Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/021.json" }, { "id": "mistral/pixtral-12b", "slug": "mistral--pixtral-12b", "name": "Pixtral 12B", "developer": "Mistral", "developer_id": "mistral", "family": "pixtral", "description": "Mistral vision-language model for image understanding and multimodal chat", "release_date": "2024-09-11", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 12.682739712, "active_parameters_b": 12, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://docs.mistral.ai/models/pixtral-12b-24-09", "source_type": "primary", "sources": [ { "label": "Official Pixtral model documentation", "url": "https://docs.mistral.ai/models/pixtral-12b-24-09", "type": "primary" }, { "label": "Official instruct model card", "url": "https://huggingface.co/mistralai/Pixtral-12B-2409/blob/a5ca2b7a9421a35ea24411c46e970758b8d3c6a7/README.md", "type": "weights" }, { "label": "Official complete checkpoint header", "url": "https://huggingface.co/mistralai/Pixtral-12B-2409/resolve/a5ca2b7a9421a35ea24411c46e970758b8d3c6a7/consolidated.safetensors", "type": "weights" }, { "label": "Official native model configuration", "url": "https://huggingface.co/mistralai/Pixtral-12B-2409/raw/a5ca2b7a9421a35ea24411c46e970758b8d3c6a7/params.json", "type": "weights" }, { "label": "Official schema: exact instruct total and active fields", "url": "https://github.com/mistralai/platform-docs-public/blob/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/schema/models/models/pixtral-12b-24-09.ts", "type": "primary" }, { "label": "Official weights-table billion units", "url": "https://github.com/mistralai/platform-docs-public/blob/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/app/%5Blocale%5D/%28docs%29/model-card/%5Bslug%5D/components/weights-table.tsx", "type": "primary" }, { "label": "Pixtral 12B technical report", "url": "https://arxiv.org/html/2410.07073v2", "type": "primary" }, { "label": "Official original-release lifecycle", "url": "https://legal.mistral.ai/ai-governance/models/pixtral-12b/", "type": "primary" }, { "label": "Official dense Transformer implementation", "url": "https://github.com/mistralai/mistral-inference/blob/9eaeb91c17450e09021b6065a1d5cc69876507c8/src/mistral_inference/transformer_layers.py", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/pixtral-12b.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Pixtral-12B-2409", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Pixtral-12B-2409 instruct checkpoint. Complete parameter count is 12,682,739,712 from its official BF16 safetensors header: decoder 12,247,782,400, vision encoder 403,489,792 and vision-language adapter 31,467,520. Published 12B decoder and 400M vision labels are nominal, not an exact full-model total. Official documentation separately reports nominal Active (B)=12 but leaves its component/execution-phase scope unspecified; it is not a derived equality with the complete parameter count. Native configuration and technical report specify 131072 context tokens, marketed as 128k. Original release was 2024-09-11; the announcement article followed on 2024-09-17. Document-QnA does not add a native PDF modality. Documented function calling does not imply the deprecated API remains operational. Dedicated reasoning classification was not explicitly verified for this exact model; the unsupported baseline false is cleared.", "parameter_components_b": { "multimodal_decoder": 12.2477824, "vision_encoder": 0.403489792, "vision_language_adapter": 0.03146752 }, "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-26T23:41:08.504Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/349.json" }, { "id": "sdaia/allam-2-7b", "slug": "sdaia--allam-2-7b", "name": "ALLaM-2-7B Instruct", "developer": "SDAIA", "developer_id": "sdaia", "family": "allam", "description": "SDAIA's instruction-tuned Arabic-English language model, pretrained from scratch.", "release_date": "2024-09-10", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 7, "active_parameters_b": 7, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/introducing-sdaia-and-their-latest-arabic-llm-on-azure-ai-model-catalog/4240241/", "source_type": "primary", "sources": [ { "label": "September 10, 2024 ALLaM-2-7B availability announcement", "url": "https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/introducing-sdaia-and-their-latest-arabic-llm-on-azure-ai-model-catalog/4240241/", "type": "primary" }, { "label": "SDAIA ALLaM-2-7b-instruct model card on Microsoft Foundry", "url": "https://ai.azure.com/catalog/models/ALLaM-2-7b-instruct?publisher=SDAIA", "type": "primary" }, { "label": "Exact allam-2-7b serving documentation", "url": "https://console.groq.com/docs/model/allam-2-7b", "type": "primary" }, { "label": "Original published checkpoint model card", "url": "https://huggingface.co/humain-ai/ALLaM-7B-Instruct-preview/blob/4cb5169d0b3c95e468aa92185a6f7c57f8499f91/README.md", "type": "primary" }, { "label": "Original checkpoint configuration, internal v1.27.2.25", "url": "https://huggingface.co/humain-ai/ALLaM-7B-Instruct-preview/blob/4cb5169d0b3c95e468aa92185a6f7c57f8499f91/config.json", "type": "primary" }, { "label": "Original three-shard checkpoint index", "url": "https://huggingface.co/humain-ai/ALLaM-7B-Instruct-preview/blob/4cb5169d0b3c95e468aa92185a6f7c57f8499f91/model.safetensors.index.json", "type": "primary" }, { "label": "Later preview card and revision distinction", "url": "https://huggingface.co/humain-ai/ALLaM-7B-Instruct-preview/blob/a28dd1e67420cde72d3629c8633a974cf7d9c366/README.md", "type": "primary" }, { "label": "ALLaM: Large Language Models for Arabic and English", "url": "https://arxiv.org/html/2407.15390v1", "type": "primary" }, { "label": "TII 3LM Figure 11 approximate active-parameter comparison", "url": "https://aclanthology.org/2025.arabicnlp-main.4.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/sdaia/allam-2-7b.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/ALLaM-AI/ALLaM-2.0-7B-Instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "7B is the official nominal total; approximately 7B active follows TII's 3LM Figure 11 comparison convention, not an exact per-token count or SDAIA active-count disclosure. The original preview card identifies ALLaM-2-7b-instruct. Its 7.000559616B checkpoint total is not asserted for undocumented hosted revisions. Azure availability dates to September 10, 2024; published preview weights appeared in February 2025 and changed in July. Apache-2.0 applies to published weights; API terms are separate. Dedicated reasoning/tool support remains unverified.", "field_status": { "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T17:57:54.392499Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/493.json" }, { "id": "deepseek/deepseek-v2.5", "slug": "deepseek--deepseek-v2.5", "name": "DeepSeek-V2.5", "developer": "DeepSeek", "developer_id": "deepseek", "family": "DeepSeek-V2.5", "description": null, "release_date": "2024-09-05", "date_precision": "day", "open_weights": true, "license": "DEEPSEEK LICENSE AGREEMENT", "parameters_b": 235.74143488, "active_parameters_b": 21, "parameters_source": "checkpoint_metadata", "architecture": "moe", "context_window": 163840, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://api-docs.deepseek.com/updates/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://api-docs.deepseek.com/updates/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V2.5", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/deepseek-ai/DeepSeek-V2.5", "type": "weights" }, { "label": "Model license", "url": "https://github.com/deepseek-ai/DeepSeek-V2/blob/main/LICENSE-MODEL", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "openbmb/minicpm3-4b", "slug": "openbmb--minicpm3-4b", "name": "MiniCPM3-4B", "developer": "OpenBMB", "developer_id": "openbmb", "family": "MiniCPM3", "description": null, "release_date": "2024-09-05", "date_precision": "day", "open_weights": true, "license": null, "parameters_b": 4, "active_parameters_b": null, "parameters_source": "model_name", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/OpenBMB/MiniCPM", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/OpenBMB/MiniCPM", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/openbmb/MiniCPM3-4B", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer card metadata says Apache 2.0, but its license section applies a separate model license whose linked file is unavailable. The model license remains unresolved; code licensing is not substituted." }, { "id": "allenai/olmoe-1b-7b-0924", "slug": "allenai--olmoe-1b-7b-0924", "name": "OLMoE-1B-7B-0924", "developer": "Ai2", "developer_id": "allenai", "family": "OLMoE", "description": null, "release_date": "2024-09-03", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 6.919161856, "active_parameters_b": 1, "parameters_source": "checkpoint_metadata", "architecture": "moe", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://docs.allenai.org/release_notes/olmo-release-notes", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://docs.allenai.org/release_notes/olmo-release-notes", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/allenai/OLMoE-1B-7B-0924", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/allenai/OLMoE-1B-7B-0924", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "writer/palmyra-x4", "slug": "writer--palmyra-x4", "name": "Palmyra X4", "developer": "Writer", "developer_id": "writer", "family": "palmyra", "description": "Enterprise language model for workflow automation, coding, data analysis, and tool use", "release_date": "2024-09", "date_precision": "month", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/writer/palmyra-x4.toml", "source_type": "catalog", "sources": [ { "label": "Choose a model: Palmyra X6, X5 and X4 comparison", "url": "https://dev.writer.com/home/models", "type": "primary" }, { "label": "Writer API changelog", "url": "https://dev.writer.com/home/changelog", "type": "primary" }, { "label": "Introducing actions with Palmyra X4", "url": "https://writer.com/engineering/actions-with-palmyra-x4/", "type": "primary" }, { "label": "Bringing AI superpowers to Slack", "url": "https://writer.com/blog/slack-integration/", "type": "primary" }, { "label": "What is new at Writer: September and October 2024 entries", "url": "https://support.writer.com/articles/1313908954-what-s-new-at-writer", "type": "primary" }, { "label": "Writer Palmyra X4 model specification and invocation schema", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-palmyra-x4.html", "type": "provider" }, { "label": "Writer chat-completion API", "url": "https://dev.writer.com/api-reference/completion-api/chat-completion", "type": "primary" }, { "label": "Writer Platform Services Agreement, sections 2 and 5", "url": "https://writer.com/legal/platform-services/", "type": "primary" }, { "label": "Palmyra X 004 dummy repository metadata", "url": "https://huggingface.co/api/models/Writer/palmyra-x-004-dummy-weights", "type": "primary" }, { "label": "Palmyra Creative model card", "url": "https://huggingface.co/Writer/Palmyra-Creative", "type": "primary" }, { "label": "Palmyra X 4.3 model card", "url": "https://huggingface.co/Writer/Palmyra-X-4.3-73B/blob/main/README.md", "type": "primary" }, { "label": "Writer-submitted 2025 transparency disclosure", "url": "https://crfm.stanford.edu/fmti/December-2025/company-reports/Writer_FinalReport_FMTI2025.html", "type": "primary" }, { "label": "Synthetic data: Busting the myths holding back enterprise AI progress", "url": "https://writer.com/engineering/synthetic-data-myths-vs-facts/", "type": "primary" }, { "label": "Introducing self-evolving models", "url": "https://writer.com/engineering/self-evolving-models/", "type": "primary" }, { "label": "Writing in the Margins: Writer GTC presentation", "url": "https://static.rainfocus.com/nvidia/gtcs25/sess/1726246419320001YBXO/FinalPresPDF/S72323_1741999284107001GsWo.pdf", "type": "primary" }, { "label": "Writer staff explanation of Palmyra Creative and X004 lineage", "url": "https://huggingface.co/Writer/Palmyra-Creative/discussions/1", "type": "primary" }, { "label": "Writer-submitted May 2024 Palmyra X transparency report", "url": "https://crfm.stanford.edu/fmti/May-2024/company-reports/Writer_Palymra%20X.html", "type": "primary" }, { "label": "May Habib interview: Palmyra X004 training and model family", "url": "https://www.mattturck.com/podcast/building-the-easy-button-for-generative-ai-may-habib-ceo-writer", "type": "primary" }, { "label": "Writer Palmyra-X-004 32K vendor Marketplace listing", "url": "https://aws.amazon.com/marketplace/pp/prodview-ry66clc5of6gw", "type": "primary" }, { "label": "Palmyra Creative pinned model configuration", "url": "https://huggingface.co/Writer/Palmyra-Creative/raw/11756c451193aa998c6365b1a51a4bbc1918ca06/config.json", "type": "primary" }, { "label": "Palmyra Creative repository metadata and nominal-size corroboration", "url": "https://huggingface.co/api/models/Writer/Palmyra-Creative", "type": "primary" }, { "label": "Writer CTO interview on Creative model merging and layer slicing", "url": "https://digiday.com/media/ai-briefing-writers-cto-on-how-to-make-ai-models-think-more-creatively/", "type": "primary" }, { "label": "Writer Palmyra Creative launch announcement", "url": "https://writer.com/blog/palmyra-creative/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/writer/palmyra-x4.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Palmyra X 004 was renamed Palmyra X4 in April 2025. September 2024 availability is inferred from dated Writer Slack announcements; October 9 is the formal launch, and the exact first day remains unresolved because current pages may contain later edits. The October 27 parameter re-audit verifies Transformer architecture from Writer's explicit Llama-based X004 statement. Creative's nominal 122B and its explicit fine-tune relationship were reassessed as potential size-inheritance evidence; the greater-depth wording does not itself establish a change relative to X004. A separate direct CTO interview does document Creative construction through merging three models and shuffling/slicing layers, so an ordinary size-preserving fine-tune cannot be assumed. It does not establish that the sizes differ either. The X4 total therefore remains unverified and its active count undisclosed; neither dummy metadata nor X4.3's later size resolves this. Generic reasoning claims do not verify a dedicated thinking mode. Writer schedules X4 deprecation for December 14, 2026.", "field_status": { "parameters_b": "unverified", "active_parameters_b": "undisclosed", "reasoning": "unverified" }, "audited_at": "2026-10-27T07:25:15.037Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/523.json" }, { "id": "cohere/command-r-08-2024", "slug": "cohere--command-r-08-2024", "name": "Command R", "developer": "Cohere", "developer_id": "cohere", "family": "command-r", "description": "Cohere's August 2024 text-generation model for long-context chat, multilingual RAG and tool use.", "release_date": "2024-08-30", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-4.0", "parameters_b": 32, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://docs.cohere.com/changelog/command-gets-refreshed", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/c4ai-command-r-08-2024", "type": "weights" }, { "label": "Cohere August 2024 release announcement", "url": "https://docs.cohere.com/changelog/command-gets-refreshed", "type": "primary" }, { "label": "Cohere Command R documentation", "url": "https://docs.cohere.com/docs/command-r", "type": "primary" }, { "label": "Cohere weight license", "url": "https://cohere.com/cohere-labs-cc-by-nc-license", "type": "primary" }, { "label": "Cohere capability definition", "url": "https://raw.githubusercontent.com/cohere-ai/cohere-developer-experience/main/fern/pages/models/the-command-family-of-models/command-r.mdx", "type": "primary" }, { "label": "Cohere capability renderer", "url": "https://raw.githubusercontent.com/cohere-ai/cohere-developer-experience/main/fern/components/model-showcase.tsx", "type": "primary" }, { "label": "Exact checkpoint architecture metadata", "url": "https://huggingface.co/api/models/CohereLabs/c4ai-command-r-08-2024", "type": "primary" }, { "label": "Cohere dense decoder implementation", "url": "https://raw.githubusercontent.com/huggingface/transformers/v4.44.2/src/transformers/models/cohere/modeling_cohere.py", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/cohere/command-r-08-2024.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "32B is Cohere’s nominal total for the August 2024 release; active parameters are not separately disclosed. The checkpoint is a dense autoregressive Transformer with GQA. Research weights use CC-BY-NC-4.0 with acceptable-use terms; hosted API access has separate terms.", "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T12:28:26.119375+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/151.json" }, { "id": "cohere/command-r-plus-08-2024", "slug": "cohere--command-r-plus-08-2024", "name": "Command R+", "developer": "Cohere", "developer_id": "cohere", "family": "command-r", "description": "Cohere's August 2024 Command R+ model for multilingual RAG, long-context text generation, and multi-step tool use", "release_date": "2024-08-30", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-4.0", "parameters_b": 104, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://docs.cohere.com/v1/changelog/command-gets-refreshed", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/CohereLabs/c4ai-command-r-plus-08-2024", "type": "weights" }, { "label": "Cohere August 30 release notes", "url": "https://docs.cohere.com/v1/changelog/command-gets-refreshed", "type": "primary" }, { "label": "Cohere Command R+ documentation", "url": "https://docs.cohere.com/docs/command-r-plus", "type": "primary" }, { "label": "Cohere exact-ID capability data", "url": "https://github.com/cohere-ai/cohere-developer-experience/blob/main/fern/pages/models/the-command-family-of-models/command-r-plus.mdx", "type": "primary" }, { "label": "Cohere capability renderer", "url": "https://github.com/cohere-ai/cohere-developer-experience/blob/main/fern/components/model-showcase.tsx", "type": "primary" }, { "label": "Exact checkpoint architecture metadata", "url": "https://huggingface.co/api/models/CohereLabs/c4ai-command-r-plus-08-2024", "type": "primary" }, { "label": "Cohere dense decoder implementation", "url": "https://raw.githubusercontent.com/huggingface/transformers/v4.44.2/src/transformers/models/cohere/modeling_cohere.py", "type": "primary" }, { "label": "Later Command A technical report", "url": "https://arxiv.org/pdf/2504.00698v2", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/cohere/command-r-plus-08-2024.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "104B is Cohere's nominal total for the August 2024 checkpoint; active parameters are not separately disclosed. Research weights use CC-BY-NC-4.0 with acceptable-use terms; hosted API terms differ. The reasoning flag describes dedicated thinking-mode support; ordinary text reasoning and tool planning remain supported.", "field_status": { "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T12:28:29.363688+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/152.json" }, { "id": "nvidia/nemotron-mini-4b-instruct", "slug": "nvidia--nemotron-mini-4b-instruct", "name": "Nemotron Mini 4B Instruct", "developer": "NVIDIA", "developer_id": "nvidia", "family": "nemotron", "description": "Dense Transformer fine-tuned from Minitron-4B-Base for English roleplay, retrieval-augmented question answering and function calling.", "release_date": "2024-08-20", "date_precision": "day", "open_weights": true, "license": "NVIDIA Open Model License", "parameters_b": 4.190509056, "active_parameters_b": 4, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://huggingface.co/nvidia/Nemotron-Mini-4B-Instruct/blob/791833e92ebddb0bc2c1007f6d2b6764f886a2ae/README.md", "source_type": "primary", "sources": [ { "label": "Pinned Nemotron Mini 4B Instruct model card", "url": "https://huggingface.co/nvidia/Nemotron-Mini-4B-Instruct/blob/791833e92ebddb0bc2c1007f6d2b6764f886a2ae/README.md", "type": "primary" }, { "label": "Pinned Nemotron Mini configuration", "url": "https://huggingface.co/nvidia/Nemotron-Mini-4B-Instruct/blob/791833e92ebddb0bc2c1007f6d2b6764f886a2ae/config.json", "type": "primary" }, { "label": "Complete PyTorch checkpoint metadata", "url": "https://huggingface.co/nvidia/Nemotron-Mini-4B-Instruct/resolve/791833e92ebddb0bc2c1007f6d2b6764f886a2ae/pytorch_model.bin", "type": "primary" }, { "label": "Transformers Nemotron implementation v4.44.2", "url": "https://github.com/huggingface/transformers/blob/v4.44.2/src/transformers/models/nemotron/modeling_nemotron.py", "type": "primary" }, { "label": "August 20, 2024 NVIDIA Mini launch announcement", "url": "https://blogs.nvidia.com/blog/digital-human-technology-mecha-break/", "type": "primary" }, { "label": "NVIDIA February 14, 2025 license metadata update", "url": "https://huggingface.co/nvidia/Nemotron-Mini-4B-Instruct/commit/791833e92ebddb0bc2c1007f6d2b6764f886a2ae", "type": "primary" }, { "label": "NVIDIA Open Model License Agreement, June 14, 2024", "url": "https://developer.download.nvidia.com/licenses/nvidia-open-model-license-agreement-june-2024.pdf", "type": "primary" }, { "label": "Compact Language Models via Pruning and Knowledge Distillation", "url": "https://arxiv.org/html/2407.14679v2", "type": "primary" }, { "label": "LLM Pruning and Distillation in Practice: The Minitron Approach, Tables 1 and 2", "url": "https://arxiv.org/html/2408.11796v1", "type": "primary" }, { "label": "Watt Counts, Table 11 and Section 5.2", "url": "https://arxiv.org/html/2604.09048v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/nvidia/nemotron-mini-4b-instruct.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The BF16 checkpoint has 4.190509056B learned parameters, consistent with NVIDIA’s nominal 4.19B Minitron base specification. Active 4B follows the Watt Counts authors’ nominal benchmark convention, not an NVIDIA disclosure or exact per-token count. Context is 4,096 tokens. August 20, 2024 marks documented NIM availability; Hugging Face distribution followed in September. NVIDIA changed the model license metadata to the Open Model License in February 2025; the card body retains an older Community license link.", "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-27T18:17:52.122626+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/391.json" }, { "id": "nousresearch/hermes-3-llama-3.1-405b", "slug": "nousresearch--hermes-3-llama-3.1-405b", "name": "Hermes 3 405B Instruct", "developer": "Nous Research", "developer_id": "nousresearch", "family": "hermes", "description": "Nous Research’s 405B full-parameter fine-tune of Llama 3.1 for instruction following, multi-turn chat, structured reasoning, code generation, and tool calling.", "release_date": "2024-08-15", "date_precision": "day", "open_weights": true, "license": "Llama 3.1 Community License", "parameters_b": 405, "active_parameters_b": 405, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/NousResearch/Hermes-3-Llama-3.1-405B", "source_type": "primary", "sources": [ { "label": "Nous Research model card", "url": "https://huggingface.co/NousResearch/Hermes-3-Llama-3.1-405B", "type": "primary" }, { "label": "Hermes 3 technical report", "url": "https://arxiv.org/html/2408.11857v1", "type": "primary" }, { "label": "Exact 405B configuration", "url": "https://huggingface.co/NousResearch/Hermes-3-Llama-3.1-405B/blob/b9fe8ede1f5075f171b2c0a235e20ae4aba33767/config.json", "type": "primary" }, { "label": "Exact 405B tokenizer and templates", "url": "https://huggingface.co/NousResearch/Hermes-3-Llama-3.1-405B/blob/b9fe8ede1f5075f171b2c0a235e20ae4aba33767/tokenizer_config.json", "type": "primary" }, { "label": "Explicitly mapped Llama architecture", "url": "https://arxiv.org/html/2407.21783v3#S3.SS2", "type": "primary" }, { "label": "Llama 3.1 governing base license", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/LICENSE", "type": "primary" }, { "label": "August 15 launch announcement", "url": "https://lambda.ai/blog/unveiling-hermes-3-the-first-fine-tuned-llama-3.1-405b-model-is-on-lambdas-cloud", "type": "primary" }, { "label": "Nous original Hermes 3 announcement", "url": "https://nousresearch.com/freedom-at-the-frontier-hermes-3", "type": "primary" }, { "label": "Nous retrospective release index (date conflict)", "url": "https://nousresearch.com/releases", "type": "primary" }, { "label": "AI21 Jamba Table 1: base-model active-count convention", "url": "https://proceedings.iclr.cc/paper_files/paper/2025/file/a9ed43fa31dc8b4a7d7a673d713dcb5f-Paper-Conference.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/nousresearch/hermes-3-llama-3.1-405b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "405B is Nous’s nominal total. Nominal 405B active is inferred from the explicit Llama 3.1 405B full-parameter fine-tune lineage using AI21’s Jamba comparison table; it is not a direct Nous/Meta active-count disclosure or an exact per-token census. This record describes the original BF16 405B fine-tune, not the separate FP8 hosting checkpoint. Reasoning denotes trained structured reasoning/scratchpad tokens, without an enforced thinking-mode or effort control. The explicit Llama 3.1 base agreement covers derivatives; Nous’s card retains a legacy llama3 metadata tag and no separate license file. Earliest public launch is August 15, 2024; the current Nous retrospective index says August 24, conflicting with the contemporaneous launch.", "field_status": {}, "audited_at": "2026-10-27T15:45:54Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/371.json" }, { "id": "nousresearch/hermes-3-llama-3.1-70b", "slug": "nousresearch--hermes-3-llama-3.1-70b", "name": "Hermes 3 70B Instruct", "developer": "Nous Research", "developer_id": "nousresearch", "family": "hermes", "description": "Nous Research instruction-tuned Llama 3.1 70B model for text chat, coding, structured reasoning, and function calling.", "release_date": "2024-08-15", "date_precision": "day", "open_weights": true, "license": "Llama 3.1 Community License", "parameters_b": 70.553706496, "active_parameters_b": 70, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://huggingface.co/NousResearch/Hermes-3-Llama-3.1-70B", "source_type": "primary", "sources": [ { "label": "Official 70B model card", "url": "https://huggingface.co/NousResearch/Hermes-3-Llama-3.1-70B/blob/1fc86da0ce9cdb14cd775ad270bc7d1b4bf70ede/README.md", "type": "primary" }, { "label": "Exact 70B checkpoint configuration", "url": "https://huggingface.co/NousResearch/Hermes-3-Llama-3.1-70B/blob/1fc86da0ce9cdb14cd775ad270bc7d1b4bf70ede/config.json", "type": "weights" }, { "label": "Complete native checkpoint tensor index", "url": "https://huggingface.co/NousResearch/Hermes-3-Llama-3.1-70B/blob/1fc86da0ce9cdb14cd775ad270bc7d1b4bf70ede/model.safetensors.index.json", "type": "weights" }, { "label": "Exact 70B tokenizer and templates", "url": "https://huggingface.co/NousResearch/Hermes-3-Llama-3.1-70B/blob/1fc86da0ce9cdb14cd775ad270bc7d1b4bf70ede/tokenizer_config.json", "type": "weights" }, { "label": "Hermes 3 Technical Report", "url": "https://arxiv.org/html/2408.11857v1", "type": "primary" }, { "label": "Nous public launch including 70B (X oEmbed)", "url": "https://publish.twitter.com/oembed?url=https://twitter.com/NousResearch/status/1824131520375951454", "type": "primary" }, { "label": "Exact 70B repository inventory", "url": "https://huggingface.co/api/models/NousResearch/Hermes-3-Llama-3.1-70B/tree/1fc86da0ce9cdb14cd775ad270bc7d1b4bf70ede?recursive=true&expand=false", "type": "weights" }, { "label": "Governing Llama 3.1 Community License", "url": "https://raw.githubusercontent.com/meta-llama/llama-models/main/models/llama3_1/LICENSE", "type": "primary" }, { "label": "Exact Meta Llama 3.1 70B base license", "url": "https://huggingface.co/api/models/meta-llama/Llama-3.1-70B", "type": "primary" }, { "label": "AI21 Jamba Table 1: nominal base-model active-count convention", "url": "https://proceedings.iclr.cc/paper_files/paper/2025/file/a9ed43fa31dc8b4a7d7a673d713dcb5f-Paper-Conference.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/nousresearch/hermes-3-llama-3.1-70b.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete native BF16 checkpoint has exactly 70.553706496B learned parameters in 723 tensors; 70B is nominal. Nominal 70B active is inferred from its documented Llama 3.1 70B fine-tuning lineage under AI21's Jamba Table 1 comparison convention; it is not a direct Nous/Meta active-count disclosure or an exact per-token census. Dense decoder-only Llama architecture and 131072-token configuration are verified. Public launch including 70B was August 15, 2024; earlier private/public repository visibility is unknown. Reasoning denotes explicitly trained structured reasoning tokens, while tools denote trained function calls; provider controls may differ. The exact card declares English and has a legacy llama3 tag; the explicitly mapped Llama 3.1 70B base is governed by the Llama 3.1 Community License, and no separate fine-tune license appears in the 70B repository.", "field_status": {}, "audited_at": "2026-10-27T15:45:56.035027Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/372.json" }, { "id": "lg-ai/exaone-3.0-7.8b-instruct", "slug": "lg-ai--exaone-3.0-7.8b-instruct", "name": "EXAONE-3.0-7.8B-Instruct", "developer": "LG AI Research", "developer_id": "lg-ai", "family": "EXAONE 3.0", "description": null, "release_date": "2024-08-07", "date_precision": "day", "open_weights": true, "license": "EXAONE AI Model License Agreement 1.1 - NC", "parameters_b": 7.818448896, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/LG-AI-EXAONE/EXAONE-3.0", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/LG-AI-EXAONE/EXAONE-3.0", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/LGAI-EXAONE/EXAONE-3.0-7.8B-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/LGAI-EXAONE/EXAONE-3.0-7.8B-Instruct", "type": "weights" }, { "label": "Model license", "url": "https://github.com/LG-AI-EXAONE/EXAONE-3.0/blob/main/LICENSE", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "openai/gpt-4o-2024-08-06", "slug": "openai--gpt-4o-2024-08-06", "name": "GPT-4o (2024-08-06)", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-4o", "description": "August 2024 GPT-4o API snapshot with text and image input, text output, function calling and Structured Outputs.", "release_date": "2024-08-06", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 128000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://openai.com/index/introducing-structured-outputs-in-the-api/", "source_type": "primary", "sources": [ { "label": "Introducing Structured Outputs in the API", "url": "https://openai.com/index/introducing-structured-outputs-in-the-api/", "type": "primary" }, { "label": "GPT-4o model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-4o.md", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201", "type": "primary" }, { "label": "GPT-5.1 Prompting Guide", "url": "https://developers.openai.com/cookbook/examples/gpt-5/gpt-5-1_prompting_guide", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "OpenAI API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "Introducing gpt-oss", "url": "https://openai.com/index/introducing-gpt-oss/", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "Toward understanding and preventing misalignment generalization", "url": "https://cdn.openai.com/pdf/a130517e-9633-47bc-8397-969807a43a23/emergent_misalignment_paper.pdf", "type": "primary" }, { "label": "GPT-4o System Card", "url": "https://openai.com/index/gpt-4o-system-card/", "type": "primary" }, { "label": "GPT-5 System Card", "url": "https://cdn.openai.com/gpt-5-system-card.pdf", "type": "primary" }, { "label": "FILBENCH: Can LLMs Understand and Generate Filipino?", "url": "https://aclanthology.org/2025.emnlp-main.127.pdf", "type": "primary" }, { "label": "RAD-Bench: Evaluating Large Language Models’ Capabilities in Retrieval Augmented Dialogues", "url": "https://aclanthology.org/2025.naacl-industry.66.pdf", "type": "primary" }, { "label": "Incompressible Knowledge Probes: Estimating Black-Box LLM Parameter Counts via Factual Capacity", "url": "https://arxiv.org/html/2604.24827v2", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-4o-2024-08-06.toml", "type": "catalog" }, { "label": "Alias documentation", "url": "https://developers.openai.com/api/docs/models/gpt-4o", "type": "docs" } ], "aliases": [ "openai/gpt-4o" ], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The gpt-4o alias defaults to this August 6, 2024 Structured Outputs snapshot. It supports 128,000-token context, text/image inputs and text output; PDF files are converted to text and page images. Broad Transformer structure is documented, but nominal total and active parameter counts remain undisclosed. SAE feature counts and black-box capacity estimates are not model parameters. OpenAI marks the snapshot deprecated with API removal scheduled for January 14, 2027.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T16:28:16.379378Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/409.json", "alias_sources": { "openai/gpt-4o": { "url": "https://developers.openai.com/api/docs/models/gpt-4o", "note": "Alias or alternate catalog name for the same released model; only the retained ID counts as a release." } } }, { "id": "openbmb/minicpm-v-2_6", "slug": "openbmb--minicpm-v-2_6", "name": "MiniCPM-V 2.6", "developer": "OpenBMB", "developer_id": "openbmb", "family": "minicpm", "description": null, "release_date": "2024-08-06", "date_precision": "day", "open_weights": true, "license": "MiniCPM Model License", "parameters_b": 8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/openbmb/MiniCPM-V-2_6", "source_type": "primary", "sources": [ { "label": "Publisher dated release log", "url": "https://github.com/OpenBMB/MiniCPM-V/blob/main/docs/changelog_en.md", "type": "announcement" }, { "label": "Publisher legacy model documentation", "url": "https://github.com/OpenBMB/MiniCPM-V/blob/main/docs/minicpm_v2dot6.md", "type": "model_card" }, { "label": "Official checkpoint metadata", "url": "https://huggingface.co/api/models/openbmb/MiniCPM-V-2_6", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T11:22:26.610Z", "last_seen": "2026-10-29T11:22:26.610Z", "upstream_present": false, "availability": "public", "notes": "OpenBMB trained multimodal adaptation of Qwen2-7B with SigLIP-400M. The dated release log verifies August 6 availability. The publisher gives a nominal 8B total and identifies the MiniCPM Model License for weights, separately from its Apache-2.0 code license. Supported context remains unverified." }, { "id": "abacusai/dracarys-llama-3.1-70b-instruct", "slug": "abacusai--dracarys-llama-3.1-70b-instruct", "name": "Dracarys Llama 3.1 70B Instruct", "developer": "Abacus.AI", "developer_id": "abacusai", "family": "llama", "description": "Abacus.AI coding fine-tune of Llama 3.1 70B Instruct", "release_date": "2024-08", "date_precision": "month", "open_weights": true, "license": null, "parameters_b": 70, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/api/models/abacusai/Dracarys-Llama-3.1-70B-Instruct/commits/main", "source_type": "primary", "sources": [ { "label": "Abacus model card", "url": "https://huggingface.co/abacusai/Dracarys-Llama-3.1-70B-Instruct/blob/7e3f4e0d296fd08a944bfad5dcb1ed675c4b38d2/README.md", "type": "primary" }, { "label": "Abacus checkpoint configuration", "url": "https://huggingface.co/abacusai/Dracarys-Llama-3.1-70B-Instruct/raw/main/config.json", "type": "primary" }, { "label": "Abacus checkpoint tensor index", "url": "https://huggingface.co/abacusai/Dracarys-Llama-3.1-70B-Instruct/raw/main/model.safetensors.index.json", "type": "weights" }, { "label": "Hugging Face checkpoint metadata", "url": "https://huggingface.co/api/models/abacusai/Dracarys-Llama-3.1-70B-Instruct", "type": "weights" }, { "label": "Abacus checkpoint upload history", "url": "https://huggingface.co/abacusai/Dracarys-Llama-3.1-70B-Instruct/commits/main", "type": "primary" }, { "label": "Meta base model card", "url": "https://huggingface.co/meta-llama/Llama-3.1-70B-Instruct", "type": "primary" }, { "label": "Abacus tokenizer configuration", "url": "https://huggingface.co/abacusai/Dracarys-Llama-3.1-70B-Instruct/raw/main/tokenizer_config.json", "type": "primary" }, { "label": "Exact Dracarys checkpoint publication history", "url": "https://huggingface.co/api/models/abacusai/Dracarys-Llama-3.1-70B-Instruct/commits/main", "type": "primary" }, { "label": "Contemporaneous public discussion on exact Dracarys model", "url": "https://huggingface.co/api/models/abacusai/Dracarys-Llama-3.1-70B-Instruct/discussions/2", "type": "primary" }, { "label": "Abacus founder Dracarys announcement", "url": "https://cdn.syndication.twimg.com/tweet-result?id=1826757521635455115&lang=en&token=x", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/abacusai/dracarys-llama-3.1-70b-instruct.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/abacusai/Dracarys-Llama-3.1-70B-Instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "70B is nominal; the published BF16 checkpoint contains 70,553,706,496 parameters. Dense model; active parameters are not separately reported. Released in August 2024. The checkpoint was uploaded August 14 and had an external public discussion August 16; the founder announcement followed August 22. Exact first-public day remains unresolved. License conflict: Abacus card declares Llama 3, while its named Llama 3.1 base declares Llama 3.1. Model config allows 131072 positions, but tokenizer model_max_length is 2048. Fine-tune-specific reasoning-mode and tool-call support remain unverified.", "field_status": { "license": "unverified", "active_parameters_b": "not_applicable", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T19:29:54.122501Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/003.json" }, { "id": "internlm/internlm2_5-1_8b", "slug": "internlm--internlm2_5-1_8b", "name": "internlm2_5-1_8b", "developer": "Shanghai AI Laboratory", "developer_id": "internlm", "family": "InternLM2.5", "description": null, "release_date": "2024-08", "date_precision": "month", "open_weights": true, "license": "apache-2.0", "parameters_b": 1.8, "active_parameters_b": null, "parameters_source": "model_name", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/InternLM/InternLM", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/InternLM/InternLM", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/internlm/internlm2_5-1_8b", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer repository news dates this release August 1; its model table dates these checkpoints August 5. Month precision records the unresolved earliest public availability." }, { "id": "internlm/internlm2_5-1_8b-chat", "slug": "internlm--internlm2_5-1_8b-chat", "name": "internlm2_5-1_8b-chat", "developer": "Shanghai AI Laboratory", "developer_id": "internlm", "family": "InternLM2.5", "description": null, "release_date": "2024-08", "date_precision": "month", "open_weights": true, "license": null, "parameters_b": 1.889110016, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/InternLM/InternLM", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/InternLM/InternLM", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/internlm/internlm2_5-1_8b-chat", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/internlm/internlm2_5-1_8b-chat", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer repository news dates this release August 1; its model table dates these checkpoints August 5. Month precision records the unresolved earliest public availability. The developer card describes separate model-use terms and code licensing but gives no named model license; no license label is inferred." }, { "id": "internlm/internlm2_5-20b", "slug": "internlm--internlm2_5-20b", "name": "internlm2_5-20b", "developer": "Shanghai AI Laboratory", "developer_id": "internlm", "family": "InternLM2.5", "description": null, "release_date": "2024-08", "date_precision": "month", "open_weights": true, "license": "apache-2.0", "parameters_b": 19.861149696, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/InternLM/InternLM", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/InternLM/InternLM", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/internlm/internlm2_5-20b", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/internlm/internlm2_5-20b", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer repository news dates this release August 1; its model table dates these checkpoints August 5. Month precision records the unresolved earliest public availability." }, { "id": "internlm/internlm2_5-20b-chat", "slug": "internlm--internlm2_5-20b-chat", "name": "internlm2_5-20b-chat", "developer": "Shanghai AI Laboratory", "developer_id": "internlm", "family": "InternLM2.5", "description": null, "release_date": "2024-08", "date_precision": "month", "open_weights": true, "license": null, "parameters_b": 19.861149696, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/InternLM/InternLM", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/InternLM/InternLM", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/internlm/internlm2_5-20b-chat", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/internlm/internlm2_5-20b-chat", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer repository news dates this release August 1; its model table dates these checkpoints August 5. Month precision records the unresolved earliest public availability. The developer card describes separate model-use terms and code licensing but gives no named model license; no license label is inferred." }, { "id": "microsoft/phi-3.5-mini-instruct", "slug": "microsoft--phi-3.5-mini-instruct", "name": "Phi-3.5-mini-instruct", "developer": "Microsoft", "developer_id": "microsoft", "family": "Phi-3.5", "description": null, "release_date": "2024-08", "date_precision": "month", "open_weights": true, "license": "mit", "parameters_b": 3.821079552, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://techcommunity.microsoft.com/blog/educatordeveloperblog/a-better-phi-family-is-coming---multi-language-support-better-vision-intelligenc/4224181", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://techcommunity.microsoft.com/blog/educatordeveloperblog/a-better-phi-family-is-coming---multi-language-support-better-vision-intelligenc/4224181", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/microsoft/Phi-3.5-mini-instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/microsoft/Phi-3.5-mini-instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "Release month verified from the developer source; an exact public availability day is not established here." }, { "id": "microsoft/phi-3.5-moe-instruct", "slug": "microsoft--phi-3.5-moe-instruct", "name": "Phi-3.5-MoE-instruct", "developer": "Microsoft", "developer_id": "microsoft", "family": "Phi-3.5", "description": null, "release_date": "2024-08", "date_precision": "month", "open_weights": true, "license": "mit", "parameters_b": 41.873153344, "active_parameters_b": 6.6, "parameters_source": "checkpoint_metadata", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://techcommunity.microsoft.com/blog/educatordeveloperblog/a-better-phi-family-is-coming---multi-language-support-better-vision-intelligenc/4224181", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://techcommunity.microsoft.com/blog/educatordeveloperblog/a-better-phi-family-is-coming---multi-language-support-better-vision-intelligenc/4224181", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/microsoft/Phi-3.5-MoE-instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/microsoft/Phi-3.5-MoE-instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "Release month verified from the developer source; an exact public availability day is not established here." }, { "id": "microsoft/phi-3.5-vision-instruct", "slug": "microsoft--phi-3.5-vision-instruct", "name": "Phi-3.5-vision-instruct", "developer": "Microsoft", "developer_id": "microsoft", "family": "Phi-3.5", "description": null, "release_date": "2024-08", "date_precision": "month", "open_weights": true, "license": "mit", "parameters_b": 4.14662144, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://techcommunity.microsoft.com/blog/educatordeveloperblog/a-better-phi-family-is-coming---multi-language-support-better-vision-intelligenc/4224181", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://techcommunity.microsoft.com/blog/educatordeveloperblog/a-better-phi-family-is-coming---multi-language-support-better-vision-intelligenc/4224181", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/microsoft/Phi-3.5-vision-instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/microsoft/Phi-3.5-vision-instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "Release month verified from the developer source; an exact public availability day is not established here." }, { "id": "sarvam/sarvam-1-v0.5", "slug": "sarvam--sarvam-1-v0.5", "name": "sarvam-1-v0.5", "developer": "Sarvam AI", "developer_id": "sarvam", "family": "Sarvam", "description": null, "release_date": "2024-08", "date_precision": "month", "open_weights": true, "license": "Sarvam AI Research License", "parameters_b": 2.508834816, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.sarvam.ai/blogs/sarvam-launch", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://www.sarvam.ai/blogs/sarvam-launch", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/sarvamai/sarvam-1-v0.5", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/sarvamai/sarvam-1-v0.5", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/sarvamai/sarvam-1-v0.5/blob/main/LICENSE.md", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer launch page is dated August 14 but its announcement body says August 13. The released Sarvam 2B model is the sarvam-1-v0.5 checkpoint, distinct from the later October Sarvam-1." }, { "id": "sarvam/shuka-1", "slug": "sarvam--shuka-1", "name": "Shuka 1", "developer": "Sarvam AI", "developer_id": "sarvam", "family": "shuka", "description": null, "release_date": "2024-08", "date_precision": "month", "open_weights": true, "license": "Llama 3 Community License", "parameters_b": 8.72578048, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text", "audio" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/sarvamai/shuka-1", "source_type": "primary", "sources": [ { "label": "Publisher launch announcement", "url": "https://www.sarvam.ai/blogs/sarvam-launch", "type": "announcement" }, { "label": "Publisher model card", "url": "https://huggingface.co/sarvamai/shuka-1/blob/cdc20cf7b92857bb2b59c4e9b6b09702586e9835/README.md", "type": "model_card" }, { "label": "Publisher checkpoint metadata including total tensor parameters", "url": "https://huggingface.co/api/models/sarvamai/shuka-1", "type": "documentation" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/sarvamai/shuka-1/blob/cdc20cf7b92857bb2b59c4e9b6b09702586e9835/config.json", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T11:22:26.610Z", "last_seen": "2026-10-29T11:22:26.610Z", "upstream_present": false, "availability": "public", "notes": "Sarvam’s trained audio-language model connects Whisper-large-v2 and Llama 3 8B Instruct using a learned projector. The launch page header says August 14 while its body dates the launch August 13, so month precision is retained. Parameter count is the complete safetensors total reported for the official checkpoint; context is the text decoder limit." }, { "id": "allenai/olmo-1b-0724", "slug": "allenai--olmo-1b-0724", "name": "OLMo-1B-0724", "developer": "Ai2", "developer_id": "allenai", "family": "OLMo", "description": null, "release_date": "2024-07-31", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 1.279787008, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://docs.allenai.org/release_notes/olmo-release-notes", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://docs.allenai.org/release_notes/olmo-release-notes", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/allenai/OLMo-1B-0724-hf", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/allenai/OLMo-1B-0724-hf", "type": "weights" } ], "aliases": [ "allenai/olmo-1b-0724-hf" ], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "allenai/olmo-7b-0724", "slug": "allenai--olmo-7b-0724", "name": "OLMo-7B-0724", "developer": "Ai2", "developer_id": "allenai", "family": "OLMo", "description": null, "release_date": "2024-07-31", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 6.888095744, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://docs.allenai.org/release_notes/olmo-release-notes", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://docs.allenai.org/release_notes/olmo-release-notes", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/allenai/OLMo-7B-0724-hf", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/allenai/OLMo-7B-0724-hf", "type": "weights" } ], "aliases": [ "allenai/olmo-7b-0724-hf" ], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "allenai/olmo-7b-0724-instruct", "slug": "allenai--olmo-7b-0724-instruct", "name": "OLMo-7B-0724-Instruct", "developer": "Ai2", "developer_id": "allenai", "family": "OLMo", "description": null, "release_date": "2024-07-31", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 6.888095744, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://docs.allenai.org/release_notes/olmo-release-notes", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://docs.allenai.org/release_notes/olmo-release-notes", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/allenai/OLMo-7B-0724-Instruct-hf", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/allenai/OLMo-7B-0724-Instruct-hf", "type": "weights" } ], "aliases": [ "allenai/olmo-7b-0724-instruct-hf" ], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "allenai/olmo-7b-0724-sft", "slug": "allenai--olmo-7b-0724-sft", "name": "OLMo-7B-0724-SFT", "developer": "Ai2", "developer_id": "allenai", "family": "OLMo", "description": null, "release_date": "2024-07-31", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 6.888095744, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://docs.allenai.org/release_notes/olmo-release-notes", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://docs.allenai.org/release_notes/olmo-release-notes", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/allenai/OLMo-7B-0724-SFT-hf", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/allenai/OLMo-7B-0724-SFT-hf", "type": "weights" } ], "aliases": [ "allenai/olmo-7b-0724-sft-hf" ], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "google/gemma-2-2b", "slug": "google--gemma-2-2b", "name": "gemma-2-2b", "developer": "Google", "developer_id": "google", "family": "Gemma 2", "description": null, "release_date": "2024-07-31", "date_precision": "day", "open_weights": true, "license": "gemma", "parameters_b": 2.614341888, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.google.dev/gemma/docs/releases", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://ai.google.dev/gemma/docs/releases", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/google/gemma-2-2b", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/google/gemma-2-2b", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "google/gemma-2-2b-it", "slug": "google--gemma-2-2b-it", "name": "Gemma 2 2B IT", "developer": "Google", "developer_id": "google", "family": "gemma", "description": "Instruction-tuned Gemma 2 text-generation model from Google, with open weights and an 8,192-token context window.", "release_date": "2024-07-31", "date_precision": "day", "open_weights": true, "license": "Gemma Terms of Use", "parameters_b": 2.614341888, "active_parameters_b": 2.61, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://developers.googleblog.com/en/smaller-safer-more-transparent-advancing-responsible-ai-with-gemma/", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/google/gemma-2-2b-it", "type": "weights" }, { "label": "Smaller, Safer, More Transparent: Advancing Responsible AI with Gemma", "url": "https://developers.googleblog.com/en/smaller-safer-more-transparent-advancing-responsible-ai-with-gemma/", "type": "primary" }, { "label": "Official Gemma 2 2B Release collection", "url": "https://huggingface.co/collections/google/gemma-2-2b-release-66a20f3796a2ff2a7c76f98f", "type": "primary" }, { "label": "Pinned official instruction-tuned checkpoint metadata", "url": "https://huggingface.co/api/models/google/gemma-2-2b-it/revision/299a8560bedf22ed1c72a8a11e7dce4a7f9f51f8", "type": "weights" }, { "label": "Official Gemma 2 PyTorch configuration, pinned revision", "url": "https://github.com/google/gemma_pytorch/blob/014acb7ac4563a5f77c76d7ff98f31b568c16508/gemma/config.py", "type": "primary" }, { "label": "Official Gemma PyTorch model implementation, pinned revision", "url": "https://github.com/google/gemma_pytorch/blob/014acb7ac4563a5f77c76d7ff98f31b568c16508/gemma/model.py", "type": "primary" }, { "label": "Gemma 2: Improving Open Language Models at a Practical Size", "url": "https://storage.googleapis.com/deepmind-media/gemma/gemma-2-report.pdf", "type": "primary" }, { "label": "Building a chatbot with Gemma", "url": "https://ai.google.dev/gemma/docs/gemma_chat", "type": "primary" }, { "label": "Gemma Terms of Use", "url": "https://ai.google.dev/gemma/terms", "type": "primary" }, { "label": "Agentic AI with Gemma 2, official pinned cookbook notebook", "url": "https://github.com/google-gemini/gemma-cookbook/blob/461e9f6fd44015171ca0efdbbfe7979427da8ea3/Gemma/%5BGemma_2%5DAgentic_AI.ipynb", "type": "primary" }, { "label": "Beyond the Chatbot: Agentic AI with Gemma", "url": "https://developers.googleblog.com/en/beyond-the-chatbot-agentic-ai-with-gemma/", "type": "primary" }, { "label": "Gemma 2 model card", "url": "https://ai.google.dev/gemma/docs/core/model_card_2", "type": "primary" }, { "label": "Watt Counts: Energy-Aware Benchmark for Sustainable LLM Inference on Heterogeneous GPU Architectures", "url": "https://arxiv.org/html/2604.09048v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemma-2-2b-it.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The 2B name excludes embeddings; the checkpoint has 2.614341888B unique learned parameters. Active size is nominally 2.61B under the Watt Counts authors’ dense-model comparison convention, not a Google per-token disclosure. Google demonstrates prompted ReAct calls but documents no built-in function calling; broader tool capability remains unverified.", "field_status": { "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T18:18:27.379370+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/216.json" }, { "id": "mistral/mistral-large-2407", "slug": "mistral--mistral-large-2407", "name": "Mistral Large 2407", "developer": "Mistral", "developer_id": "mistral", "family": "mistral-large", "description": "July 2024 Mistral Large 2 instruction model with multilingual text, code generation and native function calling", "release_date": "2024-07-24", "date_precision": "day", "open_weights": true, "license": "Mistral Research License", "parameters_b": 123, "active_parameters_b": 123, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://mistral.ai/news/mistral-large-2407/", "source_type": "primary", "sources": [ { "label": "Mistral Large 2407 official release", "url": "https://mistral.ai/news/mistral-large-2407/", "type": "primary" }, { "label": "Mistral Large 2.0 nominal total and active counts", "url": "https://docs.mistral.ai/models/mistral-large-2-0-24-07", "type": "primary" }, { "label": "Mistral Large 2.0 native modalities", "url": "https://docs.mistral.ai/models/mistral-large-2-0-24-07.md", "type": "primary" }, { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Mistral-Large-Instruct-2407", "type": "weights" }, { "label": "Mistral Research License", "url": "https://mistral.ai/licenses/MRL-0.1.md", "type": "primary" }, { "label": "Merged native 131072-token context correction", "url": "https://huggingface.co/mistralai/Mistral-Large-Instruct-2407/discussions/11", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/mistral-large-2407.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Mistral separately discloses nominal 123B total and 123B active parameters; these are not exact counts. Native checkpoint context is 131072, confirmed by a merged Mistral configuration correction; official API documentation advertises 128k without an exact serving boundary. Native modalities are text only. Dedicated reasoning-mode support remains unverified. Released weights use MRL-0.1 with research/non-commercial restrictions. Official docs mark the API retired March 30, 2025.", "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-26T23:12:05Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/333.json" }, { "id": "meta/llama-3.1-70b-instruct", "slug": "meta--llama-3.1-70b-instruct", "name": "Llama-3.1-70B-Instruct", "developer": "Meta", "developer_id": "meta", "family": "llama", "description": "Open-weight 70B dense Transformer instruction model for multilingual text dialogue and tool calling.", "release_date": "2024-07-23", "date_precision": "day", "open_weights": true, "license": "Llama 3.1 Community License", "parameters_b": 70, "active_parameters_b": 70, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://huggingface.co/meta-llama/Llama-3.1-70B-Instruct", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/meta-llama/Llama-3.1-70B-Instruct", "type": "weights" }, { "label": "Meta Llama 3 Herd report, v3", "url": "https://arxiv.org/html/2407.21783v3", "type": "primary" }, { "label": "Llama 3.1 Community License", "url": "https://github.com/meta-llama/llama-models/blob/0e0b8c519242d5833d8c11bffc1232b77ad7f301/models/llama3_1/LICENSE", "type": "primary" }, { "label": "Meta sequence-length definitions", "url": "https://github.com/meta-llama/llama-models/blob/0e0b8c519242d5833d8c11bffc1232b77ad7f301/models/sku_types.py", "type": "primary" }, { "label": "Meta 3.1 context evaluation details", "url": "https://github.com/meta-llama/llama-models/blob/0e0b8c519242d5833d8c11bffc1232b77ad7f301/models/llama3_1/eval_details.md", "type": "primary" }, { "label": "Meta 3.1 native prompt and tool protocol", "url": "https://github.com/meta-llama/llama-models/blob/0e0b8c519242d5833d8c11bffc1232b77ad7f301/models/llama3_1/prompt_format.md", "type": "primary" }, { "label": "Meta exact Instruct architecture configuration", "url": "https://github.com/meta-llama/llama-models/blob/0e0b8c519242d5833d8c11bffc1232b77ad7f301/models/sku_list.py", "type": "primary" }, { "label": "Mikhaylovskiy active-parameter convention, Table 3", "url": "https://aclanthology.org/2025.findings-emnlp.837.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/meta/llama-3.1-70b-instruct.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "70B total is Meta’s nominal specification. Nominal 70B active is inferred from Mikhaylovskiy’s base-model convention through Meta’s documented Instruct lineage and matching architecture, not a separate Meta active-count disclosure. Native context is 131072 tokens including generation. Dedicated reasoning mode is undisclosed; this does not deny ordinary reasoning ability.", "field_status": { "reasoning": "undisclosed" }, "audited_at": "2026-10-27T14:07:44.826Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/268.json" }, { "id": "meta/llama-3.1-8b-instruct", "slug": "meta--llama-3.1-8b-instruct", "name": "Llama-3.1-8B-Instruct", "developer": "Meta", "developer_id": "meta", "family": "llama", "description": "Meta open-weight 8B Llama 3.1 instruction-tuned text model for multilingual dialogue and tool use.", "release_date": "2024-07-23", "date_precision": "day", "open_weights": true, "license": "Llama 3.1 Community License", "parameters_b": 8, "active_parameters_b": 8, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct", "source_type": "primary", "sources": [ { "label": "Hugging Face", "url": "https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct", "type": "weights" }, { "label": "Original Llama 3.1 license", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/LICENSE", "type": "primary" }, { "label": "Llama 3 technical report v3", "url": "https://arxiv.org/html/2407.21783v3", "type": "primary" }, { "label": "Llama 3.1 prompt protocol", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/prompt_format.md", "type": "primary" }, { "label": "Meta model SKU limits", "url": "https://github.com/meta-llama/llama-models/blob/main/models/sku_types.py", "type": "primary" }, { "label": "Meta exact model SKU mapping", "url": "https://github.com/meta-llama/llama-models/blob/main/models/sku_list.py", "type": "primary" }, { "label": "Llama 3.1 exact evaluation limits", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/eval_details.md", "type": "primary" }, { "label": "Original Llama 3.1 release", "url": "https://ai.meta.com/blog/meta-llama-3-1/", "type": "primary" }, { "label": "Qwen3 exact-model activated-parameter comparison (Table 18)", "url": "https://arxiv.org/html/2505.09388v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/meta/llama-3.1-8b-instruct.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Meta reports a nominal 8B total. Qwen3 Table 18 reports 8B activated parameters for this exact Instruct model under Qwen Team's benchmark convention; this is not a separate Meta disclosure or exact per-token census. Dense GQA model with 131072-token context and native tool calls; a dedicated reasoning mode remains undisclosed.", "field_status": { "reasoning": "undisclosed" }, "audited_at": "2026-10-27T14:08:00.487Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/269.json" }, { "id": "meta/llama-guard-3-8b", "slug": "meta--llama-guard-3-8b", "name": "Llama-Guard-3-8B", "developer": "Meta", "developer_id": "meta", "family": "llama", "description": "Llama 3.1 8B-based text safety classifier for prompts and responses in eight languages, covering 14 hazard categories including code interpreter abuse.", "release_date": "2024-07-23", "date_precision": "day", "open_weights": true, "license": "Llama 3.1 Community License", "parameters_b": 8.030261248, "active_parameters_b": 8, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/meta-llama/Llama-Guard-3-8B", "source_type": "primary", "sources": [ { "label": "Meta Llama Guard 3 8B model card", "url": "https://raw.githubusercontent.com/meta-llama/PurpleLlama/main/Llama-Guard3/8B/MODEL_CARD.md", "type": "primary" }, { "label": "Meta exact model architecture registry", "url": "https://raw.githubusercontent.com/meta-llama/llama-models/main/models/sku_list.py", "type": "primary" }, { "label": "Meta learned-weight implementation", "url": "https://raw.githubusercontent.com/meta-llama/llama-models/main/models/llama3/model.py", "type": "primary" }, { "label": "Meta exact context registry", "url": "https://raw.githubusercontent.com/meta-llama/llama-models/main/models/sku_types.py", "type": "primary" }, { "label": "Meta July 23 release announcement", "url": "https://ai.meta.com/blog/meta-llama-3-1-ai-responsibility/", "type": "primary" }, { "label": "Meta checkpoint metadata and license", "url": "https://huggingface.co/api/models/meta-llama/Llama-Guard-3-8B", "type": "weights" }, { "label": "Meta current GitHub license copy (conflicting)", "url": "https://raw.githubusercontent.com/meta-llama/PurpleLlama/main/Llama-Guard3/8B/LICENSE", "type": "primary" }, { "label": "Mikhaylovskiy Table 3: Llama 3.1 8B nominal active convention", "url": "https://aclanthology.org/2025.findings-emnlp.837.pdf", "type": "primary" }, { "label": "Qwen3 Table 18: Llama 3.1 8B Instruct active-count corroboration", "url": "https://arxiv.org/html/2505.09388v1", "type": "primary" }, { "label": "Meta Llama Guard 3 model suite and nominal sizes", "url": "https://raw.githubusercontent.com/meta-llama/PurpleLlama/main/Llama-Guard3/README.md", "type": "primary" }, { "label": "Meta separate Llama Guard 3 8B INT8 variant", "url": "https://huggingface.co/meta-llama/Llama-Guard-3-8B-INT8/blob/main/README.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/meta/llama-guard-3-8b.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/meta-llama/Llama-Guard-3-8B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "8.030261248B total is derived from Meta’s exact registry and reference weights, corroborated by BF16 repository metadata; nominal size is 8B. Nominal active 8B is inferred from the documented Llama 3.1 8B fine-tune lineage using Mikhaylovskiy’s Table 3 convention, corroborated for the Instruct sibling by Qwen3 Table 18. It is not a separate Meta active-count disclosure or exact per-token count. The standard checkpoint retains its full vocabulary; INT8 and pruned 1B variants are separate. Checkpoint headers were not accessed. The exact Hugging Face weights carry Llama 3.1 terms, while the current PurpleLlama 8B GitHub license copy says Llama 3.2. Tool-use support here means safety classification of tool conversations, not tool-call generation. Dedicated reasoning and native tool-call generation are not separately disclosed.", "field_status": { "reasoning": "undisclosed", "tool_call": "undisclosed" }, "audited_at": "2026-10-27T14:17:11.468583Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/279.json" }, { "id": "mistral/mistral-nemo", "slug": "mistral--mistral-nemo", "name": "Mistral Nemo", "developer": "Mistral AI and NVIDIA", "developer_id": "mistral", "family": "mistral-nemo", "description": "Multilingual 12B instruction-tuned dense Transformer jointly developed by Mistral AI and NVIDIA, with Apache 2.0 open weights and function calling.", "release_date": "2024-07-18", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 12, "active_parameters_b": 12, "parameters_source": "primary_source", "architecture": "dense", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://mistral.ai/news/mistral-nemo/", "source_type": "primary", "sources": [ { "label": "Mistral NeMo release announcement", "url": "https://mistral.ai/news/mistral-nemo/", "type": "primary" }, { "label": "Mistral NeMo official model metadata", "url": "https://raw.githubusercontent.com/mistralai/platform-docs-public/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/schema/models/models/mistral-nemo-12b-24-07.ts", "type": "primary" }, { "label": "NVIDIA technical architecture article", "url": "https://developer.nvidia.com/blog/power-text-generation-applications-with-mistral-nemo-12b-running-on-a-single-gpu/", "type": "primary" }, { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407", "type": "weights" }, { "label": "Mistral-NeMo Instruct pinned configuration", "url": "https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407/blob/04d8a90549d23fc6bd7f642064003592df51e9b3/config.json", "type": "weights" }, { "label": "Mistral documentation parameter units", "url": "https://raw.githubusercontent.com/mistralai/platform-docs-public/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/app/%5Blocale%5D/(docs)/models/model-cards/%5Bslug%5D/components/weights-table.tsx", "type": "primary" }, { "label": "Mistral documentation capability schema", "url": "https://raw.githubusercontent.com/mistralai/platform-docs-public/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/schema/models/schema.ts", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/mistral-nemo.toml", "type": "catalog" }, { "label": "Mistral-NeMo Instruct native parameters", "url": "https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407/blob/04d8a90549d23fc6bd7f642064003592df51e9b3/params.json", "type": "weights" }, { "label": "NVIDIA NIM serving limitations", "url": "https://docs.nvidia.com/nim/large-language-models/1.8.0/release-notes.html", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Alias resolves to Mistral-Nemo-Instruct-2407. Total and active counts are separately published nominal 12B figures. Native params.json and Transformers config.json at revision 04d8a90549d23fc6bd7f642064003592df51e9b3 both specify 131072 positions, consistent with the nominal 128k context; deployment limits may be lower. Dedicated reasoning remains unverified. Official API metadata records retirement on 2026-07-31; this audit does not assert a currently operating endpoint.", "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-26T23:27:03.492548Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/340.json" }, { "id": "openai/gpt-4o-mini", "slug": "openai--gpt-4o-mini", "name": "GPT-4o mini", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-4o", "description": "GPT-4o variant with text and image input, text output, a 128,000-token context window and function calling.", "release_date": "2024-07-18", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 128000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://openai.com/index/gpt-4o-mini-advancing-cost-efficient-intelligence/", "source_type": "primary", "sources": [ { "label": "GPT-4o mini: advancing cost-efficient intelligence", "url": "https://openai.com/index/gpt-4o-mini-advancing-cost-efficient-intelligence/", "type": "primary" }, { "label": "GPT-4o Mini model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-4o-mini.md", "type": "primary" }, { "label": "File inputs", "url": "https://developers.openai.com/api/docs/guides/file-inputs.md", "type": "primary" }, { "label": "Reasoning best practices", "url": "https://developers.openai.com/api/docs/guides/reasoning-best-practices.md", "type": "primary" }, { "label": "Prompt Caching in the API", "url": "https://openai.com/index/api-prompt-caching/", "type": "primary" }, { "label": "Prompt Caching 201", "url": "https://developers.openai.com/cookbook/examples/prompt_caching_201", "type": "primary" }, { "label": "API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "Introducing gpt-oss", "url": "https://openai.com/index/introducing-gpt-oss/", "type": "primary" }, { "label": "OpenAI o1 System Card", "url": "https://cdn.openai.com/o1-system-card-20241205.pdf", "type": "primary" }, { "label": "GPT-5 System Card", "url": "https://cdn.openai.com/gpt-5-system-card.pdf", "type": "primary" }, { "label": "The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions", "url": "https://openai.com/index/the-instruction-hierarchy/", "type": "primary" }, { "label": "Qwen3 Technical Report, Table 14", "url": "https://arxiv.org/html/2505.09388v1", "type": "primary" }, { "label": "Kimi-VL Technical Report, Table 3", "url": "https://arxiv.org/html/2504.07491v1", "type": "primary" }, { "label": "Scam2Prompt: A Scalable Framework for Auditing Malicious Scam Endpoints in Production LLMs, Table 3", "url": "https://arxiv.org/html/2509.02372v3", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-4o-mini.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The gpt-4o-mini alias defaults to gpt-4o-mini-2024-07-18, released July 18, 2024. PDF input is processed into text and page images. Broad Transformer attention is inferred from documented prompt caching; full structure and routing remain undisclosed. Total and active parameter counts remain undisclosed: original comparison tables omit them, while published 40B/8B figures are expressly speculative. This text/vision model is separate from audio and realtime variants. API shutdown is scheduled for January 14, 2027 following October 14, 2026 deprecation; no live endpoint request was made.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T16:29:26.316771Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/411.json" }, { "id": "huggingface/smollm-1.7b", "slug": "huggingface--smollm-1.7b", "name": "SmolLM-1.7B", "developer": "Hugging Face", "developer_id": "huggingface", "family": "SmolLM", "description": null, "release_date": "2024-07-16", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 1.711376384, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/blog/smollm", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://huggingface.co/blog/smollm", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/HuggingFaceTB/SmolLM-1.7B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/HuggingFaceTB/SmolLM-1.7B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "huggingface/smollm-1.7b-instruct", "slug": "huggingface--smollm-1.7b-instruct", "name": "SmolLM-1.7B-Instruct", "developer": "Hugging Face", "developer_id": "huggingface", "family": "SmolLM", "description": null, "release_date": "2024-07-16", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 1.711376384, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/blog/smollm", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://huggingface.co/blog/smollm", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/HuggingFaceTB/SmolLM-1.7B-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/HuggingFaceTB/SmolLM-1.7B-Instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "huggingface/smollm-135m", "slug": "huggingface--smollm-135m", "name": "SmolLM-135M", "developer": "Hugging Face", "developer_id": "huggingface", "family": "SmolLM", "description": null, "release_date": "2024-07-16", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 0.134515008, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/blog/smollm", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://huggingface.co/blog/smollm", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/HuggingFaceTB/SmolLM-135M", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/HuggingFaceTB/SmolLM-135M", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "huggingface/smollm-135m-instruct", "slug": "huggingface--smollm-135m-instruct", "name": "SmolLM-135M-Instruct", "developer": "Hugging Face", "developer_id": "huggingface", "family": "SmolLM", "description": null, "release_date": "2024-07-16", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 0.134515008, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/blog/smollm", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://huggingface.co/blog/smollm", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/HuggingFaceTB/SmolLM-135M-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/HuggingFaceTB/SmolLM-135M-Instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "huggingface/smollm-360m", "slug": "huggingface--smollm-360m", "name": "SmolLM-360M", "developer": "Hugging Face", "developer_id": "huggingface", "family": "SmolLM", "description": null, "release_date": "2024-07-16", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 0.36182112, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/blog/smollm", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://huggingface.co/blog/smollm", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/HuggingFaceTB/SmolLM-360M", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/HuggingFaceTB/SmolLM-360M", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "huggingface/smollm-360m-instruct", "slug": "huggingface--smollm-360m-instruct", "name": "SmolLM-360M-Instruct", "developer": "Hugging Face", "developer_id": "huggingface", "family": "SmolLM", "description": null, "release_date": "2024-07-16", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 0.36182112, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/blog/smollm", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://huggingface.co/blog/smollm", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/HuggingFaceTB/SmolLM-360M-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/HuggingFaceTB/SmolLM-360M-Instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "openbmb/minicpm-s-1b-sft", "slug": "openbmb--minicpm-s-1b-sft", "name": "MiniCPM-S-1B-sft", "developer": "OpenBMB", "developer_id": "openbmb", "family": "MiniCPM-S", "description": null, "release_date": "2024-07-05", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 1.360246272, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/OpenBMB/MiniCPM", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/OpenBMB/MiniCPM", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/openbmb/MiniCPM-S-1B-sft", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/openbmb/MiniCPM-S-1B-sft", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "internlm/internlm2_5-7b", "slug": "internlm--internlm2_5-7b", "name": "internlm2_5-7b", "developer": "Shanghai AI Laboratory", "developer_id": "internlm", "family": "InternLM2.5", "description": null, "release_date": "2024-07-03", "date_precision": "day", "open_weights": true, "license": null, "parameters_b": 7.737708544, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/InternLM/InternLM", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/InternLM/InternLM", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/internlm/internlm2_5-7b", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/internlm/internlm2_5-7b", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "internlm/internlm2_5-7b-chat", "slug": "internlm--internlm2_5-7b-chat", "name": "internlm2_5-7b-chat", "developer": "Shanghai AI Laboratory", "developer_id": "internlm", "family": "InternLM2.5", "description": null, "release_date": "2024-07-03", "date_precision": "day", "open_weights": true, "license": null, "parameters_b": 7.737708544, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/InternLM/InternLM", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/InternLM/InternLM", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/internlm/internlm2_5-7b-chat", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/internlm/internlm2_5-7b-chat", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer card describes separate model-use terms and code licensing but gives no named model license; no license label is inferred." }, { "id": "internlm/internlm2_5-7b-chat-1m", "slug": "internlm--internlm2_5-7b-chat-1m", "name": "internlm2_5-7b-chat-1m", "developer": "Shanghai AI Laboratory", "developer_id": "internlm", "family": "InternLM2.5", "description": null, "release_date": "2024-07-03", "date_precision": "day", "open_weights": true, "license": null, "parameters_b": 7.737708544, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/InternLM/InternLM", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/InternLM/InternLM", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/internlm/internlm2_5-7b-chat-1m", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/internlm/internlm2_5-7b-chat-1m", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer card describes separate model-use terms and code licensing but gives no named model license; no license label is inferred." }, { "id": "h2oai/h2o-danube3-4b-base", "slug": "h2oai--h2o-danube3-4b-base", "name": "h2o-danube3-4b-base", "developer": "H2O.ai", "developer_id": "h2oai", "family": "H2O Danube3", "description": null, "release_date": "2024-07", "date_precision": "month", "open_weights": true, "license": "apache-2.0", "parameters_b": 3.96183936, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://h2o.ai/company/press-media/2024/h2o-announces-the-launch-of-danube3-series/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://h2o.ai/company/press-media/2024/h2o-announces-the-launch-of-danube3-series/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/h2oai/h2o-danube3-4b-base", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/h2oai/h2o-danube3-4b-base", "type": "weights" }, { "label": "Developer media archive", "url": "https://h2o.ai/content/h2o/language-masters/ja/company/press-releases/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer launch press release is dated July 17, while its press index records July 15 coverage of the already-released model family. The release month is retained; exact first external availability requires earlier dated developer evidence." }, { "id": "h2oai/h2o-danube3-4b-chat", "slug": "h2oai--h2o-danube3-4b-chat", "name": "h2o-danube3-4b-chat", "developer": "H2O.ai", "developer_id": "h2oai", "family": "H2O Danube3", "description": null, "release_date": "2024-07", "date_precision": "month", "open_weights": true, "license": "apache-2.0", "parameters_b": 3.96183936, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://h2o.ai/company/press-media/2024/h2o-announces-the-launch-of-danube3-series/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://h2o.ai/company/press-media/2024/h2o-announces-the-launch-of-danube3-series/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/h2oai/h2o-danube3-4b-chat", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/h2oai/h2o-danube3-4b-chat", "type": "weights" }, { "label": "Developer media archive", "url": "https://h2o.ai/content/h2o/language-masters/ja/company/press-releases/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer launch press release is dated July 17, while its press index records July 15 coverage of the already-released model family. The release month is retained; exact first external availability requires earlier dated developer evidence." }, { "id": "h2oai/h2o-danube3-500m-base", "slug": "h2oai--h2o-danube3-500m-base", "name": "h2o-danube3-500m-base", "developer": "H2O.ai", "developer_id": "h2oai", "family": "H2O Danube3", "description": null, "release_date": "2024-07", "date_precision": "month", "open_weights": true, "license": "apache-2.0", "parameters_b": 0.513590784, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://h2o.ai/company/press-media/2024/h2o-announces-the-launch-of-danube3-series/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://h2o.ai/company/press-media/2024/h2o-announces-the-launch-of-danube3-series/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/h2oai/h2o-danube3-500m-base", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/h2oai/h2o-danube3-500m-base", "type": "weights" }, { "label": "Developer media archive", "url": "https://h2o.ai/content/h2o/language-masters/ja/company/press-releases/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer launch press release is dated July 17, while its press index records July 15 coverage of the already-released model family. The release month is retained; exact first external availability requires earlier dated developer evidence." }, { "id": "h2oai/h2o-danube3-500m-chat", "slug": "h2oai--h2o-danube3-500m-chat", "name": "h2o-danube3-500m-chat", "developer": "H2O.ai", "developer_id": "h2oai", "family": "H2O Danube3", "description": null, "release_date": "2024-07", "date_precision": "month", "open_weights": true, "license": "apache-2.0", "parameters_b": 0.513590784, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://h2o.ai/company/press-media/2024/h2o-announces-the-launch-of-danube3-series/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://h2o.ai/company/press-media/2024/h2o-announces-the-launch-of-danube3-series/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/h2oai/h2o-danube3-500m-chat", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/h2oai/h2o-danube3-500m-chat", "type": "weights" }, { "label": "Developer media archive", "url": "https://h2o.ai/content/h2o/language-masters/ja/company/press-releases/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer launch press release is dated July 17, while its press index records July 15 coverage of the already-released model family. The release month is retained; exact first external availability requires earlier dated developer evidence." }, { "id": "google/gemma-2-27b", "slug": "google--gemma-2-27b", "name": "gemma-2-27b", "developer": "Google", "developer_id": "google", "family": "Gemma 2", "description": null, "release_date": "2024-06-27", "date_precision": "day", "open_weights": true, "license": "gemma", "parameters_b": 27.22712832, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.google.dev/gemma/docs/releases", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://ai.google.dev/gemma/docs/releases", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/google/gemma-2-27b", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/google/gemma-2-27b", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "google/gemma-2-27b-it", "slug": "google--gemma-2-27b-it", "name": "Gemma 2 27B IT", "developer": "Google", "developer_id": "google", "family": "gemma", "description": "Gemma 2 27B instruction-tuned text model with a dense decoder-only Transformer and an 8,192-token context window.", "release_date": "2024-06-27", "date_precision": "day", "open_weights": true, "license": "gemma", "parameters_b": 27.22712832, "active_parameters_b": 27, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://blog.google/innovation-and-ai/technology/developers-tools/google-gemma-2/", "source_type": "primary", "sources": [ { "label": "Gemma 2 IT release (June 27, 2024)", "url": "https://developers.googleblog.com/en/fine-tuning-gemma-2-with-keras-hugging-face-update/", "type": "primary" }, { "label": "Official Gemma 2 27B IT model card", "url": "https://huggingface.co/google/gemma-2-27b-it", "type": "primary" }, { "label": "Pinned BF16 checkpoint metadata", "url": "https://huggingface.co/api/models/google/gemma-2-27b-it/revision/aaf20e6b9f4c0fcf043f6fb2a2068419086d77b0", "type": "weights" }, { "label": "Google Gemma 27B configuration", "url": "https://github.com/google/gemma_pytorch/blob/014acb7ac4563a5f77c76d7ff98f31b568c16508/gemma/config.py", "type": "primary" }, { "label": "Google Gemma model implementation", "url": "https://github.com/google/gemma_pytorch/blob/014acb7ac4563a5f77c76d7ff98f31b568c16508/gemma/model.py", "type": "primary" }, { "label": "Gemma 2 technical report", "url": "https://arxiv.org/html/2408.00118v3", "type": "primary" }, { "label": "Gemma Terms of Use", "url": "https://ai.google.dev/gemma/terms", "type": "primary" }, { "label": "Google Gemma 2 prompt-based tool use", "url": "https://developers.googleblog.com/en/beyond-the-chatbot-agentic-ai-with-gemma/", "type": "primary" }, { "label": "Microsoft Research nominal active-size comparison", "url": "https://www.microsoft.com/en-us/research/wp-content/uploads/2025/01/WEF-2025_Leave-Behind_Phi-4.pdf", "type": "primary" }, { "label": "Gemma 2 release", "url": "https://blog.google/innovation-and-ai/technology/developers-tools/google-gemma-2/", "type": "announcement" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/google/gemma-2-27b-it.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Nominal size: 27B. Total is 27,227,128,320 unique learned parameters from pinned BF16 metadata and Google’s implementation; the paper uses 128 additional vocabulary rows. Active size is approximately 27B under Microsoft Research’s benchmark convention, not an exact count or a separate Google specification. A dedicated reasoning mode and native structured tool-call support are not established; prompt-based ReAct tool use is documented.", "field_status": { "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T13:11:20.756300Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/215.json" }, { "id": "google/gemma-2-9b", "slug": "google--gemma-2-9b", "name": "gemma-2-9b", "developer": "Google", "developer_id": "google", "family": "Gemma 2", "description": null, "release_date": "2024-06-27", "date_precision": "day", "open_weights": true, "license": "gemma", "parameters_b": 9.241705984, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.google.dev/gemma/docs/releases", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://ai.google.dev/gemma/docs/releases", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/google/gemma-2-9b", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/google/gemma-2-9b", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "google/gemma-2-9b-it", "slug": "google--gemma-2-9b-it", "name": "gemma-2-9b-it", "developer": "Google", "developer_id": "google", "family": "Gemma 2", "description": null, "release_date": "2024-06-27", "date_precision": "day", "open_weights": true, "license": "gemma", "parameters_b": 9.241705984, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.google.dev/gemma/docs/releases", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://ai.google.dev/gemma/docs/releases", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/google/gemma-2-9b-it", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/google/gemma-2-9b-it", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "anthropic/claude-3-5-sonnet-20240620", "slug": "anthropic--claude-3-5-sonnet-20240620", "name": "Claude Sonnet 3.5", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude", "description": "Original June 2024 Claude 3.5 Sonnet release, with a 200,000-token context window, text and image inputs, text output, and tool calling.", "release_date": "2024-06-20", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 200000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://platform.claude.com/docs/en/release-notes/overview", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://platform.claude.com/docs/en/release-notes/overview", "type": "announcement" }, { "label": "Original Claude 3.5 Sonnet announcement", "url": "https://www.anthropic.com/news/claude-3-5-sonnet", "type": "announcement" }, { "label": "Original model card addendum", "url": "https://www-cdn.anthropic.com/fed9cc193a14b84131812372d8d5857f8f304c52/Model_Card_Claude_3_Addendum.pdf", "type": "model_card" }, { "label": "Later original-versus-upgraded model comparisons", "url": "https://www-cdn.anthropic.com/c7822cdc35ad788ec87e14b3a9d45010f1f86c38.pdf", "type": "model_card" }, { "label": "Extended thinking introduced with Claude 3.7", "url": "https://www-cdn.anthropic.com/9ff93dfa8f445c932415d335c88852ef47f1201e.pdf", "type": "model_card" }, { "label": "Exact June revision and multimodal example", "url": "https://aws.amazon.com/blogs/aws/anthropics-claude-3-5-sonnet-model-now-available-in-amazon-bedrock-the-most-intelligent-claude-model-yet/", "type": "announcement" }, { "label": "Claude API retirement record", "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "type": "documentation" }, { "label": "Responsible Scaling Policy, Version 1.0", "url": "https://www-cdn.anthropic.com/1adf000c8f675958c2ee23805d91aaade1cd4613/responsible-scaling-policy.pdf", "type": "primary" }, { "label": "Anthropic later pricing table with conflicting context tier", "url": "https://www-cdn.anthropic.com/files/4zrzovbb/website/5678bc2f5978e5bcd4f1fe7c14b2c72284dcf9f8.pdf", "type": "documentation" }, { "label": "Combined Claude 3 card and original/upgraded Sonnet addenda", "url": "https://www-cdn.anthropic.com/c6a80a657af445f40e31afac050f3bf76d3b1404.pdf", "type": "model_card" }, { "label": "Later Claude 4 system-card parameter recheck", "url": "https://www.anthropic.com/claude-4-system-card", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "Total and active parameter counts remain undisclosed; no published nominal count or size-disclosed base lineage was found. Detailed architecture is also undisclosed. This is the June snapshot, distinct from claude-3-5-sonnet-20241022. Anthropic's release notes establish June 20, 2024 despite the current announcement page displaying June 21. Anthropic documents Claude API retirement on October 28, 2025. PDF support was added November 13, 2024 for all Sonnet 3.5 models. A May 12, 2026 pricing PDF lists a conflicting 1M API tier after retirement; retained exact-model specifications support 200K.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T10:17:00Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/090.json" }, { "id": "google/recurrentgemma-9b", "slug": "google--recurrentgemma-9b", "name": "recurrentgemma-9b", "developer": "Google", "developer_id": "google", "family": "RecurrentGemma", "description": null, "release_date": "2024-06-11", "date_precision": "day", "open_weights": true, "license": "gemma", "parameters_b": 9.628553216, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.google.dev/gemma/docs/releases", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://ai.google.dev/gemma/docs/releases", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/google/recurrentgemma-9b", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/google/recurrentgemma-9b", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "google/recurrentgemma-9b-it", "slug": "google--recurrentgemma-9b-it", "name": "recurrentgemma-9b-it", "developer": "Google", "developer_id": "google", "family": "RecurrentGemma", "description": null, "release_date": "2024-06-11", "date_precision": "day", "open_weights": true, "license": "gemma", "parameters_b": 9.628553216, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.google.dev/gemma/docs/releases", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://ai.google.dev/gemma/docs/releases", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/google/recurrentgemma-9b-it", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/google/recurrentgemma-9b-it", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "zhipuai/glm-4-9b", "slug": "zhipuai--glm-4-9b", "name": "glm-4-9b", "developer": "Z.ai", "developer_id": "zhipuai", "family": "GLM-4", "description": null, "release_date": "2024-06-05", "date_precision": "day", "open_weights": true, "license": "The glm-4-9b License", "parameters_b": 9.399951392, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/zai-org/GLM-4/blob/main/README_20240605.md", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/zai-org/GLM-4/blob/main/README_20240605.md", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/zai-org/glm-4-9b", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/zai-org/glm-4-9b", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/THUDM/glm-4-9b/raw/main/LICENSE", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "zhipuai/glm-4-9b-chat", "slug": "zhipuai--glm-4-9b-chat", "name": "glm-4-9b-chat", "developer": "Z.ai", "developer_id": "zhipuai", "family": "GLM-4", "description": null, "release_date": "2024-06-05", "date_precision": "day", "open_weights": true, "license": "The glm-4-9b License", "parameters_b": 9.399951392, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/zai-org/GLM-4/blob/main/README_20240605.md", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/zai-org/GLM-4/blob/main/README_20240605.md", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/zai-org/glm-4-9b-chat", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/zai-org/glm-4-9b-chat", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/THUDM/glm-4-9b-chat/blob/main/LICENSE", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "zhipuai/glm-4-9b-chat-1m", "slug": "zhipuai--glm-4-9b-chat-1m", "name": "glm-4-9b-chat-1m", "developer": "Z.ai", "developer_id": "zhipuai", "family": "GLM-4", "description": null, "release_date": "2024-06-05", "date_precision": "day", "open_weights": true, "license": "The glm-4-9b License", "parameters_b": 9.483857952, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 1048576, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/zai-org/GLM-4/blob/main/README_20240605.md", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/zai-org/GLM-4/blob/main/README_20240605.md", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/zai-org/glm-4-9b-chat-1m", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/zai-org/glm-4-9b-chat-1m", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/THUDM/glm-4-9b-chat-1m/blob/main/LICENSE", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "zhipuai/glm-4v-9b", "slug": "zhipuai--glm-4v-9b", "name": "glm-4v-9b", "developer": "Z.ai", "developer_id": "zhipuai", "family": "GLM-4V", "description": null, "release_date": "2024-06-05", "date_precision": "day", "open_weights": true, "license": "The glm-4-9b License", "parameters_b": 13.906320928, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 8192, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/zai-org/GLM-4/blob/main/README_20240605.md", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/zai-org/GLM-4/blob/main/README_20240605.md", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/zai-org/glm-4v-9b", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/zai-org/glm-4v-9b", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/THUDM/glm-4v-9b/blob/main/LICENSE", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "alibaba/qwen2-0.5b", "slug": "alibaba--qwen2-0.5b", "name": "Qwen2-0.5B", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen2", "description": null, "release_date": "2024-06", "date_precision": "month", "open_weights": true, "license": "apache-2.0", "parameters_b": 0.494032768, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen2/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen2/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen2-0.5B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2-0.5B", "type": "weights" }, { "label": "Developer release changelog (June 6)", "url": "https://github.com/QwenLM/Qwen3", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The Qwen2 release blog is dated June 7, while the developer repository changelog dates release June 6. The release month is recorded because the earliest external availability day is unresolved." }, { "id": "alibaba/qwen2-0.5b-instruct", "slug": "alibaba--qwen2-0.5b-instruct", "name": "Qwen2-0.5B-Instruct", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen2", "description": null, "release_date": "2024-06", "date_precision": "month", "open_weights": true, "license": "apache-2.0", "parameters_b": 0.494032768, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen2/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen2/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen2-0.5B-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2-0.5B-Instruct", "type": "weights" }, { "label": "Developer release changelog (June 6)", "url": "https://github.com/QwenLM/Qwen3", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The Qwen2 release blog is dated June 7, while the developer repository changelog dates release June 6. The release month is recorded because the earliest external availability day is unresolved." }, { "id": "alibaba/qwen2-1.5b", "slug": "alibaba--qwen2-1.5b", "name": "Qwen2-1.5B", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen2", "description": null, "release_date": "2024-06", "date_precision": "month", "open_weights": true, "license": "apache-2.0", "parameters_b": 1.543714304, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen2/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen2/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen2-1.5B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2-1.5B", "type": "weights" }, { "label": "Developer release changelog (June 6)", "url": "https://github.com/QwenLM/Qwen3", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The Qwen2 release blog is dated June 7, while the developer repository changelog dates release June 6. The release month is recorded because the earliest external availability day is unresolved." }, { "id": "alibaba/qwen2-1.5b-instruct", "slug": "alibaba--qwen2-1.5b-instruct", "name": "Qwen2-1.5B-Instruct", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen2", "description": null, "release_date": "2024-06", "date_precision": "month", "open_weights": true, "license": "apache-2.0", "parameters_b": 1.543714304, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen2/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen2/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen2-1.5B-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2-1.5B-Instruct", "type": "weights" }, { "label": "Developer release changelog (June 6)", "url": "https://github.com/QwenLM/Qwen3", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The Qwen2 release blog is dated June 7, while the developer repository changelog dates release June 6. The release month is recorded because the earliest external availability day is unresolved." }, { "id": "alibaba/qwen2-57b-a14b", "slug": "alibaba--qwen2-57b-a14b", "name": "Qwen2-57B-A14B", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen2", "description": null, "release_date": "2024-06", "date_precision": "month", "open_weights": true, "license": "apache-2.0", "parameters_b": 57.408658944, "active_parameters_b": 14, "parameters_source": "checkpoint_metadata", "architecture": "moe", "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen2/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen2/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen2-57B-A14B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2-57B-A14B", "type": "weights" }, { "label": "Developer release changelog (June 6)", "url": "https://github.com/QwenLM/Qwen3", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The Qwen2 release blog is dated June 7, while the developer repository changelog dates release June 6. The release month is recorded because the earliest external availability day is unresolved." }, { "id": "alibaba/qwen2-57b-a14b-instruct", "slug": "alibaba--qwen2-57b-a14b-instruct", "name": "Qwen2-57B-A14B-Instruct", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen2", "description": null, "release_date": "2024-06", "date_precision": "month", "open_weights": true, "license": "apache-2.0", "parameters_b": 57.408658944, "active_parameters_b": 14, "parameters_source": "checkpoint_metadata", "architecture": "moe", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen2/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen2/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen2-57B-A14B-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2-57B-A14B-Instruct", "type": "weights" }, { "label": "Developer release changelog (June 6)", "url": "https://github.com/QwenLM/Qwen3", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The Qwen2 release blog is dated June 7, while the developer repository changelog dates release June 6. The release month is recorded because the earliest external availability day is unresolved." }, { "id": "alibaba/qwen2-72b", "slug": "alibaba--qwen2-72b", "name": "Qwen2-72B", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen2", "description": null, "release_date": "2024-06", "date_precision": "month", "open_weights": true, "license": "Tongyi Qianwen LICENSE AGREEMENT", "parameters_b": 72.706203648, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen2/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen2/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen2-72B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2-72B", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/Qwen/Qwen2-72B/blob/main/LICENSE", "type": "license" }, { "label": "Developer release changelog (June 6)", "url": "https://github.com/QwenLM/Qwen3", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The Qwen2 release blog is dated June 7, while the developer repository changelog dates release June 6. The release month is recorded because the earliest external availability day is unresolved." }, { "id": "alibaba/qwen2-72b-instruct", "slug": "alibaba--qwen2-72b-instruct", "name": "Qwen2-72B-Instruct", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen2", "description": null, "release_date": "2024-06", "date_precision": "month", "open_weights": true, "license": "Tongyi Qianwen LICENSE AGREEMENT", "parameters_b": 72.706203648, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen2/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen2/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen2-72B-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2-72B-Instruct", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/Qwen/Qwen2-72B-Instruct/blob/main/LICENSE", "type": "license" }, { "label": "Developer release changelog (June 6)", "url": "https://github.com/QwenLM/Qwen3", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The Qwen2 release blog is dated June 7, while the developer repository changelog dates release June 6. The release month is recorded because the earliest external availability day is unresolved." }, { "id": "alibaba/qwen2-7b", "slug": "alibaba--qwen2-7b", "name": "Qwen2-7B", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen2", "description": null, "release_date": "2024-06", "date_precision": "month", "open_weights": true, "license": "apache-2.0", "parameters_b": 7.615616512, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen2/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen2/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen2-7B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2-7B", "type": "weights" }, { "label": "Developer release changelog (June 6)", "url": "https://github.com/QwenLM/Qwen3", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The Qwen2 release blog is dated June 7, while the developer repository changelog dates release June 6. The release month is recorded because the earliest external availability day is unresolved." }, { "id": "alibaba/qwen2-7b-instruct", "slug": "alibaba--qwen2-7b-instruct", "name": "Qwen2-7B-Instruct", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen2", "description": null, "release_date": "2024-06", "date_precision": "month", "open_weights": true, "license": "apache-2.0", "parameters_b": 7.615616512, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen2/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen2/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen2-7B-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen2-7B-Instruct", "type": "weights" }, { "label": "Developer release changelog (June 6)", "url": "https://github.com/QwenLM/Qwen3", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The Qwen2 release blog is dated June 7, while the developer repository changelog dates release June 6. The release month is recorded because the earliest external availability day is unresolved." }, { "id": "deepseek/deepseek-coder-v2-base", "slug": "deepseek--deepseek-coder-v2-base", "name": "DeepSeek-Coder-V2-Base", "developer": "DeepSeek", "developer_id": "deepseek", "family": "DeepSeek-Coder-V2", "description": null, "release_date": "2024-06", "date_precision": "month", "open_weights": true, "license": "DEEPSEEK LICENSE AGREEMENT", "parameters_b": 235.74143488, "active_parameters_b": 21, "parameters_source": "checkpoint_metadata", "architecture": "moe", "context_window": 163840, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/deepseek-ai/DeepSeek-Coder-V2", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/deepseek-ai/DeepSeek-Coder-V2", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/deepseek-ai/DeepSeek-Coder-V2-Base", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/deepseek-ai/DeepSeek-Coder-V2-Base", "type": "weights" }, { "label": "Model license", "url": "https://github.com/deepseek-ai/DeepSeek-Coder-V2/blob/main/LICENSE-MODEL", "type": "license" }, { "label": "Dated developer technical report", "url": "https://arxiv.org/abs/2406.11931", "type": "paper" }, { "label": "Developer API release changelog", "url": "https://api-docs.deepseek.com/updates/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The API changelog records Coder-V2-0614 on June 14; the downloadable-weight family was announced later in June. Exact earliest availability of this specific checkpoint remains unresolved; June is recorded." }, { "id": "deepseek/deepseek-coder-v2-instruct", "slug": "deepseek--deepseek-coder-v2-instruct", "name": "DeepSeek-Coder-V2-Instruct", "developer": "DeepSeek", "developer_id": "deepseek", "family": "DeepSeek-Coder-V2", "description": null, "release_date": "2024-06", "date_precision": "month", "open_weights": true, "license": "DEEPSEEK LICENSE AGREEMENT", "parameters_b": 235.74143488, "active_parameters_b": 21, "parameters_source": "checkpoint_metadata", "architecture": "moe", "context_window": 163840, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/deepseek-ai/DeepSeek-Coder-V2", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/deepseek-ai/DeepSeek-Coder-V2", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/deepseek-ai/DeepSeek-Coder-V2-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/deepseek-ai/DeepSeek-Coder-V2-Instruct", "type": "weights" }, { "label": "Model license", "url": "https://github.com/deepseek-ai/DeepSeek-Coder-V2/blob/main/LICENSE-MODEL", "type": "license" }, { "label": "Dated developer technical report", "url": "https://arxiv.org/abs/2406.11931", "type": "paper" }, { "label": "Developer API release changelog", "url": "https://api-docs.deepseek.com/updates/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The API changelog records Coder-V2-0614 on June 14; the downloadable-weight family was announced later in June. Exact earliest availability of this specific checkpoint remains unresolved; June is recorded." }, { "id": "deepseek/deepseek-coder-v2-lite-base", "slug": "deepseek--deepseek-coder-v2-lite-base", "name": "DeepSeek-Coder-V2-Lite-Base", "developer": "DeepSeek", "developer_id": "deepseek", "family": "DeepSeek-Coder-V2", "description": null, "release_date": "2024-06", "date_precision": "month", "open_weights": true, "license": "DEEPSEEK LICENSE AGREEMENT", "parameters_b": 15.706484224, "active_parameters_b": 2.4, "parameters_source": "checkpoint_metadata", "architecture": "moe", "context_window": 163840, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/deepseek-ai/DeepSeek-Coder-V2", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/deepseek-ai/DeepSeek-Coder-V2", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/deepseek-ai/DeepSeek-Coder-V2-Lite-Base", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/deepseek-ai/DeepSeek-Coder-V2-Lite-Base", "type": "weights" }, { "label": "Model license", "url": "https://github.com/deepseek-ai/DeepSeek-Coder-V2/blob/main/LICENSE-MODEL", "type": "license" }, { "label": "Dated developer technical report", "url": "https://arxiv.org/abs/2406.11931", "type": "paper" }, { "label": "Developer API release changelog", "url": "https://api-docs.deepseek.com/updates/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "Release month verified from developer evidence; exact public availability day is unresolved." }, { "id": "deepseek/deepseek-coder-v2-lite-instruct", "slug": "deepseek--deepseek-coder-v2-lite-instruct", "name": "DeepSeek-Coder-V2-Lite-Instruct", "developer": "DeepSeek", "developer_id": "deepseek", "family": "DeepSeek-Coder-V2", "description": null, "release_date": "2024-06", "date_precision": "month", "open_weights": true, "license": "DEEPSEEK LICENSE AGREEMENT", "parameters_b": 15.706484224, "active_parameters_b": 2.4, "parameters_source": "checkpoint_metadata", "architecture": "moe", "context_window": 163840, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/deepseek-ai/DeepSeek-Coder-V2", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/deepseek-ai/DeepSeek-Coder-V2", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct", "type": "weights" }, { "label": "Model license", "url": "https://github.com/deepseek-ai/DeepSeek-Coder-V2/blob/main/LICENSE-MODEL", "type": "license" }, { "label": "Dated developer technical report", "url": "https://arxiv.org/abs/2406.11931", "type": "paper" }, { "label": "Developer API release changelog", "url": "https://api-docs.deepseek.com/updates/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "Release month verified from developer evidence; exact public availability day is unresolved." }, { "id": "mistral/codestral-22b-v0.1", "slug": "mistral--codestral-22b-v0.1", "name": "Codestral-22B-v0.1", "developer": "Mistral", "developer_id": "mistral", "family": "codestral", "description": "Dense code model trained on over 80 programming languages, supporting instruction following, fill-in-the-middle completion and native tool calls.", "release_date": "2024-05-29", "date_precision": "day", "open_weights": true, "license": "Mistral AI Non-Production License", "parameters_b": 22, "active_parameters_b": 22, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://mistral.ai/news/codestral/", "source_type": "primary", "sources": [ { "label": "Codestral launch announcement", "url": "https://mistral.ai/news/codestral/", "type": "primary" }, { "label": "Codestral 24.05 model documentation", "url": "https://docs.mistral.ai/models/codestral-24-05", "type": "primary" }, { "label": "Codestral-22B-v0.1 model card", "url": "https://huggingface.co/mistralai/Codestral-22B-v0.1/blob/28b1c1a51dabe9d86ca8c41420ada1984632498f/README.md", "type": "weights" }, { "label": "Codestral-22B-v0.1 Transformers configuration", "url": "https://huggingface.co/mistralai/Codestral-22B-v0.1/blob/28b1c1a51dabe9d86ca8c41420ada1984632498f/config.json", "type": "weights" }, { "label": "Codestral-22B-v0.1 native parameters", "url": "https://huggingface.co/mistralai/Codestral-22B-v0.1/blob/28b1c1a51dabe9d86ca8c41420ada1984632498f/params.json", "type": "weights" }, { "label": "Mistral AI Non-Production License 0.1", "url": "https://mistral.ai/licences/MNPL-0.1.md", "type": "primary" }, { "label": "Mistral reference inference library README", "url": "https://github.com/mistralai/mistral-inference/blob/main/README.md", "type": "primary" }, { "label": "Mistral reference Transformer layers", "url": "https://github.com/mistralai/mistral-inference/blob/main/src/mistral_inference/transformer_layers.py", "type": "primary" }, { "label": "Codestral-22B-v0.1 tokenizer configuration", "url": "https://huggingface.co/mistralai/Codestral-22B-v0.1/blob/28b1c1a51dabe9d86ca8c41420ada1984632498f/tokenizer_config.json", "type": "weights" }, { "label": "Codestral official repository metadata and bounded access probe", "url": "https://huggingface.co/api/models/mistralai/Codestral-22B-v0.1", "type": "weights" }, { "label": "Codestral-22B-v0.1 checkpoint index", "url": "https://huggingface.co/mistralai/Codestral-22B-v0.1/blob/28b1c1a51dabe9d86ca8c41420ada1984632498f/model.safetensors.index.json", "type": "weights" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/codestral-22b-v0.1.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Codestral-22B-v0.1", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "22B total and 22B active are developer nominal figures; the original model documentation explicitly gives both, without defining an exact tensor-count convention. The checkpoint is a dense GQA Transformer with 32768 positions. Weights use MNPL 0.1, with commercial licensing separate. Native tool calling is documented through mistral_common; the bundled Hugging Face Jinja template does not itself serialize tools. Dedicated thinking-mode status is unverified. Mistral marks codestral-2405 retired June 16, 2025; published weights remain accessible.", "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-26T22:45:55.188975Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/312.json" }, { "id": "mistral/mistral-7b-instruct-v0.3", "slug": "mistral--mistral-7b-instruct-v0.3", "name": "Mistral 7B Instruct v0.3", "developer": "Mistral", "developer_id": "mistral", "family": "mistral", "description": "Open 7B Mistral instruction model with function calling", "release_date": "2024-05-22", "date_precision": "day", "open_weights": true, "license": "Apache 2.0", "parameters_b": 7.248023552, "active_parameters_b": 7, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://docs.mistral.ai/models/mistral-7b-0-3", "source_type": "primary", "sources": [ { "label": "Mistral 7B v0.3 official release and active parameters", "url": "https://docs.mistral.ai/models/mistral-7b-0-3", "type": "primary" }, { "label": "Mistral 7B Instruct v0.3 official model card", "url": "https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.3/blob/c170c708c41dac9275d15a8fff4eca08d52bab71/README.md", "type": "primary" }, { "label": "Mistral 7B Instruct v0.3 configuration", "url": "https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.3/blob/c170c708c41dac9275d15a8fff4eca08d52bab71/config.json", "type": "primary" }, { "label": "Mistral 7B Instruct v0.3 checkpoint index", "url": "https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.3/blob/c170c708c41dac9275d15a8fff4eca08d52bab71/model.safetensors.index.json", "type": "weights" }, { "label": "Mistral 7B Instruct v0.3 native weights", "url": "https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.3/resolve/c170c708c41dac9275d15a8fff4eca08d52bab71/consolidated.safetensors", "type": "weights" }, { "label": "Mistral official dense Transformer implementation", "url": "https://github.com/mistralai/mistral-inference/blob/9eaeb91c17450e09021b6065a1d5cc69876507c8/src/mistral_inference/transformer_layers.py", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/mistral-7b-instruct-v0.3.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.3", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Total parameters count all 291 learned BF16 tensors: 7,248,023,552. Mistral separately reports a nominal 7B active; this is not an exact active count. Native reasoning support remains unverified.", "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-26T23:09:39.780292Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/330.json" }, { "id": "microsoft/phi-3-medium-128k-instruct", "slug": "microsoft--phi-3-medium-128k-instruct", "name": "Phi-3-medium-128k-instruct", "developer": "Microsoft", "developer_id": "microsoft", "family": "Phi-3", "description": null, "release_date": "2024-05-21", "date_precision": "day", "open_weights": true, "license": "mit", "parameters_b": 13.96023808, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://azure.microsoft.com/en-us/blog/new-models-added-to-the-phi-3-family-available-on-microsoft-azure/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://azure.microsoft.com/en-us/blog/new-models-added-to-the-phi-3-family-available-on-microsoft-azure/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/microsoft/Phi-3-medium-128k-instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/microsoft/Phi-3-medium-128k-instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "microsoft/phi-3-medium-4k-instruct", "slug": "microsoft--phi-3-medium-4k-instruct", "name": "Phi-3-medium-4k-instruct", "developer": "Microsoft", "developer_id": "microsoft", "family": "Phi-3", "description": null, "release_date": "2024-05-21", "date_precision": "day", "open_weights": true, "license": "mit", "parameters_b": 13.96023808, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://azure.microsoft.com/en-us/blog/new-models-added-to-the-phi-3-family-available-on-microsoft-azure/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://azure.microsoft.com/en-us/blog/new-models-added-to-the-phi-3-family-available-on-microsoft-azure/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/microsoft/Phi-3-medium-4k-instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/microsoft/Phi-3-medium-4k-instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "microsoft/phi-3-small-128k-instruct", "slug": "microsoft--phi-3-small-128k-instruct", "name": "Phi-3-small-128k-instruct", "developer": "Microsoft", "developer_id": "microsoft", "family": "Phi-3", "description": null, "release_date": "2024-05-21", "date_precision": "day", "open_weights": true, "license": "mit", "parameters_b": 7.392272384, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://azure.microsoft.com/en-us/blog/new-models-added-to-the-phi-3-family-available-on-microsoft-azure/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://azure.microsoft.com/en-us/blog/new-models-added-to-the-phi-3-family-available-on-microsoft-azure/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/microsoft/Phi-3-small-128k-instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/microsoft/Phi-3-small-128k-instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "microsoft/phi-3-small-8k-instruct", "slug": "microsoft--phi-3-small-8k-instruct", "name": "Phi-3-small-8k-instruct", "developer": "Microsoft", "developer_id": "microsoft", "family": "Phi-3", "description": null, "release_date": "2024-05-21", "date_precision": "day", "open_weights": true, "license": "mit", "parameters_b": 7.392274432, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://azure.microsoft.com/en-us/blog/new-models-added-to-the-phi-3-family-available-on-microsoft-azure/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://azure.microsoft.com/en-us/blog/new-models-added-to-the-phi-3-family-available-on-microsoft-azure/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/microsoft/Phi-3-small-8k-instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/microsoft/Phi-3-small-8k-instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "microsoft/phi-3-vision-128k-instruct", "slug": "microsoft--phi-3-vision-128k-instruct", "name": "Phi-3-vision-128k-instruct", "developer": "Microsoft", "developer_id": "microsoft", "family": "Phi-3", "description": null, "release_date": "2024-05-21", "date_precision": "day", "open_weights": true, "license": "mit", "parameters_b": 4.14662144, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 131072, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://azure.microsoft.com/en-us/blog/new-models-added-to-the-phi-3-family-available-on-microsoft-azure/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://azure.microsoft.com/en-us/blog/new-models-added-to-the-phi-3-family-available-on-microsoft-azure/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/microsoft/Phi-3-vision-128k-instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/microsoft/Phi-3-vision-128k-instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "openbmb/minicpm-llama3-v-2_5", "slug": "openbmb--minicpm-llama3-v-2_5", "name": "MiniCPM-Llama3-V 2.5", "developer": "OpenBMB", "developer_id": "openbmb", "family": "minicpm", "description": null, "release_date": "2024-05-20", "date_precision": "day", "open_weights": true, "license": "MiniCPM Model License; Llama 3 Community License", "parameters_b": 8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/openbmb/MiniCPM-Llama3-V-2_5", "source_type": "primary", "sources": [ { "label": "Publisher dated release log", "url": "https://github.com/OpenBMB/MiniCPM-V/blob/main/docs/changelog_en.md", "type": "announcement" }, { "label": "Publisher model card", "url": "https://huggingface.co/openbmb/MiniCPM-Llama3-V-2_5/blob/fd7f352fac0e06d0d818b23f98e3ec8c64267a57/README.md", "type": "model_card" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/openbmb/MiniCPM-Llama3-V-2_5/blob/fd7f352fac0e06d0d818b23f98e3ec8c64267a57/config.json", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T11:22:26.610Z", "last_seen": "2026-10-29T11:22:26.610Z", "upstream_present": false, "availability": "public", "notes": "OpenBMB trained multimodal adaptation of Llama 3 8B Instruct plus SigLIP. Parameter count is the publisher’s nominal total. The exact checkpoint configuration specifies an 8192-token context." }, { "id": "deepseek/deepseek-v2-lite", "slug": "deepseek--deepseek-v2-lite", "name": "DeepSeek-V2-Lite", "developer": "DeepSeek", "developer_id": "deepseek", "family": "DeepSeek-V2", "description": null, "release_date": "2024-05-16", "date_precision": "day", "open_weights": true, "license": "DEEPSEEK LICENSE AGREEMENT", "parameters_b": 15.706484224, "active_parameters_b": 2.4, "parameters_source": "checkpoint_metadata", "architecture": "moe", "context_window": 163840, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/deepseek-ai/DeepSeek-V2", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/deepseek-ai/DeepSeek-V2", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V2-Lite", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/deepseek-ai/DeepSeek-V2-Lite", "type": "weights" }, { "label": "Model license", "url": "https://github.com/deepseek-ai/DeepSeek-V2/blob/main/LICENSE-MODEL", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "deepseek/deepseek-v2-lite-chat", "slug": "deepseek--deepseek-v2-lite-chat", "name": "DeepSeek-V2-Lite-Chat", "developer": "DeepSeek", "developer_id": "deepseek", "family": "DeepSeek-V2", "description": null, "release_date": "2024-05-16", "date_precision": "day", "open_weights": true, "license": "DEEPSEEK LICENSE AGREEMENT", "parameters_b": 15.706484224, "active_parameters_b": 2.4, "parameters_source": "checkpoint_metadata", "architecture": "moe", "context_window": 163840, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/deepseek-ai/DeepSeek-V2", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/deepseek-ai/DeepSeek-V2", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V2-Lite-Chat", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/deepseek-ai/DeepSeek-V2-Lite-Chat", "type": "weights" }, { "label": "Model license", "url": "https://github.com/deepseek-ai/DeepSeek-V2/blob/main/LICENSE-MODEL", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "01-ai/yi-1.5-34b", "slug": "01-ai--yi-1.5-34b", "name": "Yi-1.5-34B", "developer": "01.AI", "developer_id": "01-ai", "family": "Yi-1.5", "description": null, "release_date": "2024-05-13", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 34.388917248, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/01-ai/Yi-1.5", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/01-ai/Yi-1.5", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/01-ai/Yi-1.5-34B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/01-ai/Yi-1.5-34B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "01-ai/yi-1.5-34b-chat", "slug": "01-ai--yi-1.5-34b-chat", "name": "Yi-1.5-34B-Chat", "developer": "01.AI", "developer_id": "01-ai", "family": "Yi-1.5", "description": null, "release_date": "2024-05-13", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 34.388917248, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/01-ai/Yi-1.5", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/01-ai/Yi-1.5", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/01-ai/Yi-1.5-34B-Chat", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/01-ai/Yi-1.5-34B-Chat", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "01-ai/yi-1.5-6b", "slug": "01-ai--yi-1.5-6b", "name": "Yi-1.5-6B", "developer": "01.AI", "developer_id": "01-ai", "family": "Yi-1.5", "description": null, "release_date": "2024-05-13", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 6.06103552, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/01-ai/Yi-1.5", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/01-ai/Yi-1.5", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/01-ai/Yi-1.5-6B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/01-ai/Yi-1.5-6B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "01-ai/yi-1.5-6b-chat", "slug": "01-ai--yi-1.5-6b-chat", "name": "Yi-1.5-6B-Chat", "developer": "01.AI", "developer_id": "01-ai", "family": "Yi-1.5", "description": null, "release_date": "2024-05-13", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 6.06103552, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/01-ai/Yi-1.5", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/01-ai/Yi-1.5", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/01-ai/Yi-1.5-6B-Chat", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/01-ai/Yi-1.5-6B-Chat", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "01-ai/yi-1.5-9b", "slug": "01-ai--yi-1.5-9b", "name": "Yi-1.5-9B", "developer": "01.AI", "developer_id": "01-ai", "family": "Yi-1.5", "description": null, "release_date": "2024-05-13", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 8.829407232, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/01-ai/Yi-1.5", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/01-ai/Yi-1.5", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/01-ai/Yi-1.5-9B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/01-ai/Yi-1.5-9B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "01-ai/yi-1.5-9b-chat", "slug": "01-ai--yi-1.5-9b-chat", "name": "Yi-1.5-9B-Chat", "developer": "01.AI", "developer_id": "01-ai", "family": "Yi-1.5", "description": null, "release_date": "2024-05-13", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 8.829407232, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/01-ai/Yi-1.5", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/01-ai/Yi-1.5", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/01-ai/Yi-1.5-9B-Chat", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/01-ai/Yi-1.5-9B-Chat", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "openai/gpt-4o-2024-05-13", "slug": "openai--gpt-4o-2024-05-13", "name": "GPT-4o (2024-05-13)", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-4o", "description": "Initial GPT-4o API snapshot, with text and image input, text output, a 128,000-token context window and function calling.", "release_date": "2024-05-13", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 128000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://openai.com/index/hello-gpt-4o/", "source_type": "primary", "sources": [ { "label": "Hello GPT-4o", "url": "https://openai.com/index/hello-gpt-4o/", "type": "primary" }, { "label": "GPT-4o model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-4o.md", "type": "primary" }, { "label": "Foundry Models sold by Azure: GPT-4o capabilities", "url": "https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure", "type": "primary" }, { "label": "Introducing Structured Outputs in the API", "url": "https://openai.com/index/introducing-structured-outputs-in-the-api/", "type": "primary" }, { "label": "OpenAI API deprecations", "url": "https://developers.openai.com/api/docs/deprecations.md", "type": "primary" }, { "label": "Introducing gpt-oss", "url": "https://openai.com/index/introducing-gpt-oss/", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "GPT-4o System Card", "url": "https://cdn.openai.com/gpt-4o-system-card.pdf", "type": "primary" }, { "label": "Toward understanding and preventing misalignment generalization", "url": "https://cdn.openai.com/pdf/a130517e-9633-47bc-8397-969807a43a23/emergent_misalignment_paper.pdf", "type": "primary" }, { "label": "GPT-5 System Card", "url": "https://cdn.openai.com/gpt-5-system-card.pdf", "type": "primary" }, { "label": "GPT-5.1 Prompting Guide", "url": "https://developers.openai.com/cookbook/examples/gpt-5/gpt-5-1_prompting_guide", "type": "primary" }, { "label": "Helpful assistant features suppress emergent misalignment", "url": "https://alignment.openai.com/helpful-assistant-features/", "type": "primary" }, { "label": "DeepSeek-Coder-V2: Breaking the Barrier of Closed-Source Models in Code Intelligence", "url": "https://arxiv.org/html/2406.11931v1", "type": "primary" }, { "label": "HELMET: How to Evaluate Long-Context Language Models Effectively and Thoroughly", "url": "https://proceedings.iclr.cc/paper_files/paper/2025/file/f5332c8273d02729730a9c24dec2135e-Paper-Conference.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-4o-2024-05-13.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Specifications describe gpt-4o-2024-05-13, the initial text/vision API snapshot, not later GPT-4o, ChatGPT, audio or realtime variants. Its documented context is 128,000 tokens with a separate 4,096-token output limit. OpenAI lists this snapshot as shut down October 23, 2026; no live endpoint call was made. Total and active parameter counts remain undisclosed; original DeepSeek-Coder-V2 and HELMET comparisons leave the exact May snapshot counts unknown. Exact architecture and dedicated reasoning-mode classification remain unverified; a launch topic tag, later unpinned family descriptions and August-snapshot interpretability research are not substituted for exact May evidence.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "unverified", "reasoning": "unverified" }, "audited_at": "2026-10-27T16:26:30.316961Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/408.json" }, { "id": "deepseek/deepseek-v2", "slug": "deepseek--deepseek-v2", "name": "DeepSeek-V2", "developer": "DeepSeek", "developer_id": "deepseek", "family": "DeepSeek-V2", "description": null, "release_date": "2024-05-06", "date_precision": "day", "open_weights": true, "license": "DEEPSEEK LICENSE AGREEMENT", "parameters_b": 235.74143488, "active_parameters_b": 21, "parameters_source": "checkpoint_metadata", "architecture": "moe", "context_window": 163840, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/deepseek-ai/DeepSeek-V2", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/deepseek-ai/DeepSeek-V2", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V2", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/deepseek-ai/DeepSeek-V2", "type": "weights" }, { "label": "Model license", "url": "https://github.com/deepseek-ai/DeepSeek-V2/blob/main/LICENSE-MODEL", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "deepseek/deepseek-v2-chat", "slug": "deepseek--deepseek-v2-chat", "name": "DeepSeek-V2-Chat", "developer": "DeepSeek", "developer_id": "deepseek", "family": "DeepSeek-V2", "description": null, "release_date": "2024-05-06", "date_precision": "day", "open_weights": true, "license": "DEEPSEEK LICENSE AGREEMENT", "parameters_b": 235.74143488, "active_parameters_b": 21, "parameters_source": "checkpoint_metadata", "architecture": "moe", "context_window": 163840, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/deepseek-ai/DeepSeek-V2", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/deepseek-ai/DeepSeek-V2", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V2-Chat", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/deepseek-ai/DeepSeek-V2-Chat", "type": "weights" }, { "label": "Model license", "url": "https://github.com/deepseek-ai/DeepSeek-V2/blob/main/LICENSE-MODEL", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "nousresearch/hermes-2-pro-llama-3-8b", "slug": "nousresearch--hermes-2-pro-llama-3-8b", "name": "Hermes 2 Pro Llama 3 8B", "developer": "Nous Research", "developer_id": "nousresearch", "family": "hermes", "description": "Llama 3 fine-tune optimized for function calling, structured JSON responses, and dialogue", "release_date": "2024-05-01", "date_precision": "day", "open_weights": true, "license": "Llama 3 Community License", "parameters_b": 8.030523392, "active_parameters_b": 8, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://huggingface.co/NousResearch/Hermes-2-Pro-Llama-3-8B", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/NousResearch/Hermes-2-Pro-Llama-3-8B/blob/f798274b30e7d2d4797c369edcc0cc7473b6e6f2/README.md", "type": "primary" }, { "label": "Official checkpoint configuration", "url": "https://huggingface.co/NousResearch/Hermes-2-Pro-Llama-3-8B/blob/f798274b30e7d2d4797c369edcc0cc7473b6e6f2/config.json", "type": "weights" }, { "label": "Complete checkpoint tensor index", "url": "https://huggingface.co/NousResearch/Hermes-2-Pro-Llama-3-8B/blob/f798274b30e7d2d4797c369edcc0cc7473b6e6f2/model.safetensors.index.json", "type": "weights" }, { "label": "Official safetensors metadata", "url": "https://huggingface.co/api/models/NousResearch/Hermes-2-Pro-Llama-3-8B", "type": "weights" }, { "label": "Nous launch announcement (X oEmbed)", "url": "https://publish.twitter.com/oembed?url=https://twitter.com/NousResearch/status/1785779313826308096", "type": "primary" }, { "label": "Official launch timestamp", "url": "https://cdn.syndication.twimg.com/tweet-result?id=1785779313826308096&lang=en&token=1", "type": "primary" }, { "label": "Checkpoint history", "url": "https://huggingface.co/api/models/NousResearch/Hermes-2-Pro-Llama-3-8B/commits/main", "type": "weights" }, { "label": "Meta Llama 3 model card", "url": "https://raw.githubusercontent.com/meta-llama/llama3/main/MODEL_CARD.md", "type": "primary" }, { "label": "Meta Llama 3 license", "url": "https://raw.githubusercontent.com/meta-llama/llama3/main/LICENSE", "type": "primary" }, { "label": "Hermes function-calling code", "url": "https://raw.githubusercontent.com/NousResearch/Hermes-Function-Calling/main/README.md", "type": "primary" }, { "label": "Named Llama 3 8B base configuration", "url": "https://huggingface.co/NousResearch/Meta-Llama-3-8B/raw/main/config.json", "type": "weights" }, { "label": "Qwen3 Table 6: Llama 3 8B nominal active convention", "url": "https://arxiv.org/html/2505.09388v1#S3.T6", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/nousresearch/hermes-2-pro-llama-3-8b.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/NousResearch/Hermes-2-Pro-Llama-3-8B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Complete main FP16 checkpoint contains exactly 8.030523392B learned parameters (291 tensors); 8B is nominal. Active 8B is a nominal lineage inference from Qwen3 Table 6's explicit Llama-3-8B Base convention, not a direct Nous disclosure or exact per-token count. Hermes retains the base layer dimensions and adds 32 vocabulary tokens, below that convention's precision. Current card specifies llama3 despite initial Apache-2.0 metadata. Public launch announcement is May 1, 2024; April 30 is the earlier repository artifact-upload date, with historical public visibility unverified. Reasoning-mode support remains unverified; Hermes 3 scratch_pad documentation is not transferred.", "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-27T15:46:04.940281Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/370.json" }, { "id": "microsoft/phi-3-mini-128k-instruct", "slug": "microsoft--phi-3-mini-128k-instruct", "name": "Phi-3-mini-128k-instruct", "developer": "Microsoft", "developer_id": "microsoft", "family": "Phi-3", "description": null, "release_date": "2024-04-23", "date_precision": "day", "open_weights": true, "license": "mit", "parameters_b": 3.821079552, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://azure.microsoft.com/en-us/blog/introducing-phi-3-redefining-whats-possible-with-slms/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://azure.microsoft.com/en-us/blog/introducing-phi-3-redefining-whats-possible-with-slms/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/microsoft/Phi-3-mini-128k-instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/microsoft/Phi-3-mini-128k-instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "microsoft/phi-3-mini-4k-instruct", "slug": "microsoft--phi-3-mini-4k-instruct", "name": "Phi-3-mini-4k-instruct", "developer": "Microsoft", "developer_id": "microsoft", "family": "Phi-3", "description": null, "release_date": "2024-04-23", "date_precision": "day", "open_weights": true, "license": "mit", "parameters_b": 3.821079552, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://azure.microsoft.com/en-us/blog/introducing-phi-3-redefining-whats-possible-with-slms/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://azure.microsoft.com/en-us/blog/introducing-phi-3-redefining-whats-possible-with-slms/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/microsoft/Phi-3-mini-4k-instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/microsoft/Phi-3-mini-4k-instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "meta/llama-3-70b", "slug": "meta--llama-3-70b", "name": "Llama 3 70B", "developer": "Meta", "developer_id": "meta", "family": "llama-3", "description": "Original Llama 3 70B pretrained text-generation model with a dense decoder-only Transformer, grouped-query attention, and an 8,192-token context.", "release_date": "2024-04-18", "date_precision": "day", "open_weights": true, "license": "Llama 3 Community License", "parameters_b": 70, "active_parameters_b": 70, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": false, "source_url": "https://ai.meta.com/blog/meta-llama-3", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://ai.meta.com/blog/meta-llama-3/", "type": "primary" }, { "label": "Official model card", "url": "https://github.com/meta-llama/llama3/blob/main/MODEL_CARD.md", "type": "weights" }, { "label": "Official base checkpoint", "url": "https://huggingface.co/meta-llama/Meta-Llama-3-70B", "type": "weights" }, { "label": "Reference architecture implementation", "url": "https://github.com/meta-llama/llama3/blob/main/llama/model.py", "type": "primary" }, { "label": "Original model usage and context support", "url": "https://github.com/meta-llama/llama3/blob/main/README.md", "type": "primary" }, { "label": "Llama 3 Community License", "url": "https://github.com/meta-llama/llama3/blob/main/LICENSE", "type": "primary" }, { "label": "Meta comparison of original Llama 3 and 3.1", "url": "https://arxiv.org/html/2407.21783v3#S1.T1", "type": "primary" }, { "label": "DeepSeek-AI nominal active-parameter comparison, Table 2", "url": "https://arxiv.org/html/2405.04434v1", "type": "primary" }, { "label": "DeepSeek-AI original base-model comparator identity", "url": "https://github.com/deepseek-ai/DeepSeek-V2", "type": "primary" }, { "label": "Current Meta copy of original Llama 3 model card", "url": "https://github.com/meta-llama/llama-models/blob/main/models/llama3/MODEL_CARD.md", "type": "weights" }, { "label": "Original release source", "url": "https://ai.meta.com/blog/meta-llama-3", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "70B total is Meta's nominal size; 70B active follows DeepSeek-V2 Table 2's published comparison convention. Neither is an exact tensor or per-token census. Original April 2024 base checkpoint with 8192-token context, distinct from Llama 3.1. A dedicated reasoning mode is not documented; Meta marks native tool use absent.", "field_status": { "reasoning": "undisclosed" }, "audited_at": "2026-10-27T14:02:55.124696Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/264.json" }, { "id": "meta/llama-3-70b-instruct", "slug": "meta--llama-3-70b-instruct", "name": "Llama 3 70B Instruct", "developer": "Meta", "developer_id": "meta", "family": "llama-3", "description": "Llama 3 instruction-tuned checkpoint.", "release_date": "2024-04-18", "date_precision": "day", "open_weights": true, "license": "Llama 3 Community License", "parameters_b": 70, "active_parameters_b": 70, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": false, "source_url": "https://ai.meta.com/blog/meta-llama-3", "source_type": "primary", "sources": [ { "label": "Original Llama 3 release announcement", "url": "https://ai.meta.com/blog/meta-llama-3/", "type": "primary" }, { "label": "Exact official Instruct model card", "url": "https://huggingface.co/meta-llama/Meta-Llama-3-70B-Instruct", "type": "weights" }, { "label": "Official checkpoint parameter metadata", "url": "https://huggingface.co/api/models/meta-llama/Meta-Llama-3-70B-Instruct", "type": "weights" }, { "label": "Official model card", "url": "https://github.com/meta-llama/llama3/blob/main/MODEL_CARD.md", "type": "weights" }, { "label": "Meta model comparison: original Llama 3 versus 3.1", "url": "https://arxiv.org/html/2407.21783v3#S1.T1", "type": "primary" }, { "label": "DeepSeek-V2 Table 3: Llama 3 70B Instruct activated parameters", "url": "https://arxiv.org/html/2405.04434v1#S4.T3", "type": "primary" }, { "label": "Original release source", "url": "https://ai.meta.com/blog/meta-llama-3", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "Original April 2024 Llama 3 70B Instruct: Meta specifies nominal 70B total parameters. DeepSeek-V2 Table 3 explicitly lists this Instruct model at nominal 70B activated parameters under DeepSeek's comparison convention; no exact per-token breakdown is established. Native tool use is absent in Meta's comparison table; Llama 3.1 tool support and 128K context do not apply.", "field_status": { "reasoning": "undisclosed" }, "audited_at": "2026-10-27T14:02:54.596149Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/265.json" }, { "id": "meta/llama-3-8b", "slug": "meta--llama-3-8b", "name": "Llama 3 8B", "developer": "Meta", "developer_id": "meta", "family": "llama-3", "description": "Meta’s original Llama 3 8B pretrained text-generation model, using a dense decoder-only Transformer with grouped-query attention and an 8,192-token context window.", "release_date": "2024-04-18", "date_precision": "day", "open_weights": true, "license": "Llama 3 Community License", "parameters_b": 8, "active_parameters_b": 8, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": false, "source_url": "https://ai.meta.com/blog/meta-llama-3", "source_type": "primary", "sources": [ { "label": "Original release announcement", "url": "https://ai.meta.com/blog/meta-llama-3/", "type": "primary" }, { "label": "Exact 8B base model card", "url": "https://huggingface.co/meta-llama/Meta-Llama-3-8B", "type": "weights" }, { "label": "Official model card", "url": "https://github.com/meta-llama/llama3/blob/main/MODEL_CARD.md", "type": "weights" }, { "label": "Original model usage and context", "url": "https://github.com/meta-llama/llama3/blob/main/README.md", "type": "primary" }, { "label": "Original Transformer implementation", "url": "https://github.com/meta-llama/llama3/blob/main/llama/model.py", "type": "primary" }, { "label": "Original Llama 3 license", "url": "https://github.com/meta-llama/llama3/blob/main/LICENSE", "type": "primary" }, { "label": "Meta comparison of original Llama 3 and 3.1", "url": "https://arxiv.org/html/2407.21783v3#S1.T1", "type": "primary" }, { "label": "Public exact-repository tensor metadata", "url": "https://huggingface.co/api/models/meta-llama/Meta-Llama-3-8B", "type": "catalog" }, { "label": "Qwen3 Table 6: original Llama 3 8B Base activated parameters", "url": "https://arxiv.org/html/2505.09388v1#S3.T6", "type": "paper" }, { "label": "DeepSeek comparison: approximate activated parameters", "url": "https://arxiv.org/html/2405.04434v1#S0.F1", "type": "paper" }, { "label": "Original release source", "url": "https://ai.meta.com/blog/meta-llama-3", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "Total parameters are Meta’s nominal 8B. The nominal 8B active count follows the Qwen3 authors’ explicit comparison table for Llama-3-8B Base; it is not a separate Meta disclosure or an exact per-token tensor count. A dedicated reasoning mode remains unspecified. Meta marks native tool use absent for this original pretrained checkpoint.", "field_status": { "reasoning": "undisclosed" }, "audited_at": "2026-10-27T14:05:10.409Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/266.json" }, { "id": "meta/llama-3-8b-instruct", "slug": "meta--llama-3-8b-instruct", "name": "Llama 3 8B Instruct", "developer": "Meta", "developer_id": "meta", "family": "llama-3", "description": "Meta’s original Llama 3 8B instruction-tuned text model, aligned for dialogue using supervised fine-tuning and RLHF, with a dense decoder-only Transformer, grouped-query attention and an 8,192-token context.", "release_date": "2024-04-18", "date_precision": "day", "open_weights": true, "license": "Llama 3 Community License", "parameters_b": 8, "active_parameters_b": 8, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": false, "source_url": "https://ai.meta.com/blog/meta-llama-3", "source_type": "primary", "sources": [ { "label": "Original release announcement", "url": "https://ai.meta.com/blog/meta-llama-3/", "type": "primary" }, { "label": "Exact official checkpoint card", "url": "https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct", "type": "weights" }, { "label": "Original reference architecture", "url": "https://github.com/meta-llama/llama3/blob/main/llama/model.py", "type": "primary" }, { "label": "Original inference and context documentation", "url": "https://github.com/meta-llama/llama3/blob/main/README.md", "type": "primary" }, { "label": "Original community license", "url": "https://github.com/meta-llama/llama3/blob/main/LICENSE", "type": "primary" }, { "label": "Meta retrospective model capabilities, Table 1", "url": "https://arxiv.org/html/2407.21783v3#S1.T1", "type": "primary" }, { "label": "Qwen original Llama 3 base activated-count convention, Table 6", "url": "https://arxiv.org/html/2505.09388v1#S3.T6", "type": "primary" }, { "label": "Original release source", "url": "https://ai.meta.com/blog/meta-llama-3", "type": "announcement" }, { "label": "Official model card", "url": "https://github.com/meta-llama/llama3/blob/main/MODEL_CARD.md", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "8B total is Meta’s nominal specification. Nominal 8B active is inferred through Meta’s documented Instruct post-training lineage from Qwen’s explicit original Llama-3-8B Base activated-count table; it is not a Meta per-token tensor census. Original April 2024 context is 8192 tokens, with no native tool use in Meta’s retrospective table.", "field_status": { "reasoning": "undisclosed" }, "audited_at": "2026-10-27T14:05:44.967Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/267.json" }, { "id": "allenai/olmo-7b-0424", "slug": "allenai--olmo-7b-0424", "name": "OLMo-7B-0424", "developer": "Ai2", "developer_id": "allenai", "family": "OLMo", "description": null, "release_date": "2024-04-17", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "model_name", "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://docs.allenai.org/release_notes/olmo-release-notes", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://docs.allenai.org/release_notes/olmo-release-notes", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/allenai/OLMo-7B-0424", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "mistral/mixtral-8x22b-instruct", "slug": "mistral--mixtral-8x22b-instruct", "name": "Mixtral 8x22B Instruct", "developer": "Mistral", "developer_id": "mistral", "family": "mixtral", "description": "Instruction-tuned multilingual text MoE model with native function calling, 141B total parameters, 39B active parameters, and a 64K-token context window.", "release_date": "2024-04-17", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 141, "active_parameters_b": 39, "parameters_source": "primary_source", "architecture": "moe", "context_window": 65536, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://docs.mistral.ai/models/mixtral-8x22b-0-1-0-3", "source_type": "primary", "sources": [ { "label": "Cheaper, Better, Faster, Stronger — Mixtral 8x22B release", "url": "https://mistral.ai/news/mixtral-8x22b/", "type": "primary" }, { "label": "Mixtral-8x22B-Instruct-v0.1 official model card", "url": "https://huggingface.co/mistralai/Mixtral-8x22B-Instruct-v0.1/blob/cc88a6cc19fbd17d9f1c0ee0b0d70a748dce698d/README.md", "type": "weights" }, { "label": "Mixtral-8x22B-Instruct-v0.1 checkpoint configuration", "url": "https://huggingface.co/mistralai/Mixtral-8x22B-Instruct-v0.1/blob/cc88a6cc19fbd17d9f1c0ee0b0d70a748dce698d/config.json", "type": "weights" }, { "label": "Official model parameters", "url": "https://docs.mistral.ai/models/mixtral-8x22b-0-1-0-3", "type": "weights" }, { "label": "Pinned official Mixtral 8x22B model specification", "url": "https://github.com/mistralai/platform-docs-public/blob/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/schema/models/models/mixtral-8x22b-0-1-0-3.ts", "type": "primary" }, { "label": "Mistral inference model download and version notes", "url": "https://github.com/mistralai/mistral-inference/blob/main/README.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/mixtral-8x22b-instruct.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Audited exact Mixtral-8x22B-Instruct-v0.1; Mistral states the v0.3 Instruct archive is the same model in safetensors format. Official Instruct weights table separately publishes nominal 141B total and 39B active parameters. Supported 64K context is corroborated by the 65536-position checkpoint configuration. Native input/output are text. Dedicated reasoning status remains unverified and is cleared to null; reasoning benchmark performance is not evidence of a dedicated mode. The historical API is documented as retired.", "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-26T23:36:53.435398Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/347.json" }, { "id": "microsoft/wizardlm-2-8x22b", "slug": "microsoft--wizardlm-2-8x22b", "name": "WizardLM-2 8x22B", "developer": "Microsoft", "developer_id": "microsoft", "family": "wizardlm", "description": "Microsoft WizardLM team's multilingual text model, post-trained for complex instructions and multi-turn conversation using a Mixtral-8x22B-based mixture of experts.", "release_date": "2024-04-15", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 141, "active_parameters_b": 39, "parameters_source": "primary_source", "architecture": "moe", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://wizardlm.github.io/WizardLM2/", "source_type": "primary", "sources": [ { "label": "WizardLM-2 official release", "url": "https://wizardlm.github.io/WizardLM2/", "type": "primary" }, { "label": "Original Microsoft 8x22B card (April 15 archive)", "url": "https://web.archive.org/web/20240415175129id_/https://huggingface.co/microsoft/WizardLM-2-8x22B", "type": "primary" }, { "label": "WizardLM release-process correction", "url": "https://x.com/WizardLM_AI/status/1780101465950105775", "type": "primary" }, { "label": "Arena Learning first-party report", "url": "https://arxiv.org/html/2407.10627v1", "type": "primary" }, { "label": "Community archive provenance", "url": "https://huggingface.co/alpindale/WizardLM-2-8x22B/blob/661fa9dcdb48da87a59da37a97df289ee6f5dd50/huggingface-metadata.txt", "type": "weights" }, { "label": "Community mirror configuration (unverified original provenance)", "url": "https://huggingface.co/alpindale/WizardLM-2-8x22B/blob/661fa9dcdb48da87a59da37a97df289ee6f5dd50/config.json", "type": "weights" }, { "label": "Cheaper, Better, Faster, Stronger: original Mixtral 8x22B announcement", "url": "https://mistral.ai/news/mixtral-8x22b/", "type": "primary" }, { "label": "Mistral current Mixtral 8x22B base and instruct weights table", "url": "https://docs.mistral.ai/models/mixtral-8x22b-0-1-0-3", "type": "primary" }, { "label": "Official Mixtral 8x22B v0.1 base configuration", "url": "https://huggingface.co/mistralai/Mixtral-8x22B-v0.1/raw/main/config.json", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/microsoft/wizardlm-2-8x22b.toml", "type": "catalog" }, { "label": "Hugging Face archive", "url": "https://huggingface.co/alpindale/WizardLM-2-8x22B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Microsoft reports nominal 141B total. Nominal 39B active is inferred from the explicitly identified Mixtral-8x22B-v0.1 base and preserved MoE geometry. Official release was April 15, 2024; the team acknowledged missed toxicity testing the next day. Original hosting is inaccessible; community weights remain available without independent byte-authenticity verification. Original context remains unverified (mirror config 65536). Dedicated reasoning and native tool-call features are not separately documented.", "field_status": { "context_window": "unverified", "reasoning": "undisclosed", "tool_call": "undisclosed" }, "audited_at": "2026-10-27T10:53:19.177215+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/297.json" }, { "id": "openbmb/minicpm-v-2", "slug": "openbmb--minicpm-v-2", "name": "MiniCPM-V-2", "developer": "OpenBMB", "developer_id": "openbmb", "family": "MiniCPM-V", "description": null, "release_date": "2024-04-12", "date_precision": "day", "open_weights": true, "license": null, "parameters_b": 3.434965792, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/OpenBMB/MiniCPM-V/blob/main/docs/changelog_en.md", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/OpenBMB/MiniCPM-V/blob/main/docs/changelog_en.md", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/openbmb/MiniCPM-V-2", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/openbmb/MiniCPM-V-2", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "openbmb/minicpm-1b-sft", "slug": "openbmb--minicpm-1b-sft", "name": "MiniCPM-1B-sft", "developer": "OpenBMB", "developer_id": "openbmb", "family": "MiniCPM", "description": null, "release_date": "2024-04-11", "date_precision": "day", "open_weights": true, "license": "General Model License Agreement - Source Notes - Publicity Restrictions - Commercial License", "parameters_b": 1.2, "active_parameters_b": null, "parameters_source": "primary", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/OpenBMB/MiniCPM", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/OpenBMB/MiniCPM", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/openbmb/MiniCPM-1B-sft-bf16", "type": "model_card" }, { "label": "Model license", "url": "https://github.com/OpenBMB/General-Model-License/blob/main/%E9%80%9A%E7%94%A8%E6%A8%A1%E5%9E%8B%E8%AE%B8%E5%8F%AF%E5%8D%8F%E8%AE%AE-%E6%9D%A5%E6%BA%90%E8%AF%B4%E6%98%8E-%E5%AE%A3%E4%BC%A0%E9%99%90%E5%88%B6-%E5%95%86%E4%B8%9A%E6%8E%88%E6%9D%83.md", "type": "license" } ], "aliases": [ "MiniCPM-1B-sft-bf16" ], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "parameters_scope": "non_embedding", "parameter_components_b": { "non_embedding": 1.2 }, "notes": "The developer parameter count excludes embeddings; it is stored as a scoped component rather than a total." }, { "id": "openbmb/minicpm-2b-128k", "slug": "openbmb--minicpm-2b-128k", "name": "MiniCPM-2B-128k", "developer": "OpenBMB", "developer_id": "openbmb", "family": "MiniCPM", "description": null, "release_date": "2024-04-11", "date_precision": "day", "open_weights": true, "license": "General Model License Agreement - Source Notes - Publicity Restrictions - Commercial License", "parameters_b": 2.4, "active_parameters_b": null, "parameters_source": "primary", "architecture": null, "context_window": 65536, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/OpenBMB/MiniCPM", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/OpenBMB/MiniCPM", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/openbmb/MiniCPM-2B-128k", "type": "model_card" }, { "label": "Model license", "url": "https://github.com/OpenBMB/General-Model-License/blob/main/%E9%80%9A%E7%94%A8%E6%A8%A1%E5%9E%8B%E8%AE%B8%E5%8F%AF%E5%8D%8F%E8%AE%AE-%E6%9D%A5%E6%BA%90%E8%AF%B4%E6%98%8E-%E5%AE%A3%E4%BC%A0%E9%99%90%E5%88%B6-%E5%95%86%E4%B8%9A%E6%8E%88%E6%9D%83.md", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "parameters_scope": "non_embedding", "parameter_components_b": { "non_embedding": 2.4 }, "notes": "The developer parameter count excludes embeddings; it is stored as a scoped component rather than a total." }, { "id": "openbmb/minicpm-moe-8x2b", "slug": "openbmb--minicpm-moe-8x2b", "name": "MiniCPM-MoE-8x2B", "developer": "OpenBMB", "developer_id": "openbmb", "family": "MiniCPM", "description": null, "release_date": "2024-04-11", "date_precision": "day", "open_weights": true, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/OpenBMB/MiniCPM", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/OpenBMB/MiniCPM", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/openbmb/MiniCPM-MoE-8x2B", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "Developer release changelog verifies this checkpoint. Total and active parameter counts remain unresolved; the 8x2B name is not multiplied to fabricate a total." }, { "id": "google/codegemma-2b", "slug": "google--codegemma-2b", "name": "codegemma-2b", "developer": "Google", "developer_id": "google", "family": "CodeGemma", "description": null, "release_date": "2024-04-09", "date_precision": "day", "open_weights": true, "license": "gemma", "parameters_b": 2.506172416, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.google.dev/gemma/docs/releases", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://ai.google.dev/gemma/docs/releases", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/google/codegemma-2b", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/google/codegemma-2b", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "google/codegemma-7b", "slug": "google--codegemma-7b", "name": "codegemma-7b", "developer": "Google", "developer_id": "google", "family": "CodeGemma", "description": null, "release_date": "2024-04-09", "date_precision": "day", "open_weights": true, "license": "gemma", "parameters_b": 8.537680896, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.google.dev/gemma/docs/releases", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://ai.google.dev/gemma/docs/releases", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/google/codegemma-7b", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/google/codegemma-7b", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "google/codegemma-7b-it", "slug": "google--codegemma-7b-it", "name": "codegemma-7b-it", "developer": "Google", "developer_id": "google", "family": "CodeGemma", "description": null, "release_date": "2024-04-09", "date_precision": "day", "open_weights": true, "license": "gemma", "parameters_b": 8.537680896, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.google.dev/gemma/docs/releases", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://ai.google.dev/gemma/docs/releases", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/google/codegemma-7b-it", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/google/codegemma-7b-it", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "google/recurrentgemma-2b", "slug": "google--recurrentgemma-2b", "name": "recurrentgemma-2b", "developer": "Google", "developer_id": "google", "family": "RecurrentGemma", "description": null, "release_date": "2024-04-09", "date_precision": "day", "open_weights": true, "license": "gemma", "parameters_b": 2.68286208, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.google.dev/gemma/docs/releases", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://ai.google.dev/gemma/docs/releases", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/google/recurrentgemma-2b", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/google/recurrentgemma-2b", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "google/recurrentgemma-2b-it", "slug": "google--recurrentgemma-2b-it", "name": "recurrentgemma-2b-it", "developer": "Google", "developer_id": "google", "family": "RecurrentGemma", "description": null, "release_date": "2024-04-09", "date_precision": "day", "open_weights": true, "license": "gemma", "parameters_b": 2.68286208, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.google.dev/gemma/docs/releases", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://ai.google.dev/gemma/docs/releases", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/google/recurrentgemma-2b-it", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/google/recurrentgemma-2b-it", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "stabilityai/stablelm-2-12b", "slug": "stabilityai--stablelm-2-12b", "name": "stablelm-2-12b", "developer": "Stability AI", "developer_id": "stabilityai", "family": "Stable LM 2", "description": null, "release_date": "2024-04-08", "date_precision": "day", "open_weights": true, "license": "STABILITY AI COMMUNITY LICENSE AGREEMENT", "parameters_b": 12.14318592, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://stability-ai.squarespace.com/news-updates/introducing-stable-lm-2-12b", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://stability-ai.squarespace.com/news-updates/introducing-stable-lm-2-12b", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/stabilityai/stablelm-2-12b", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/stabilityai/stablelm-2-12b", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/stabilityai/stablelm-2-12b/blob/main/LICENSE.md", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "stabilityai/stablelm-2-12b-chat", "slug": "stabilityai--stablelm-2-12b-chat", "name": "stablelm-2-12b-chat", "developer": "Stability AI", "developer_id": "stabilityai", "family": "Stable LM 2", "description": null, "release_date": "2024-04-08", "date_precision": "day", "open_weights": true, "license": "STABILITY AI NON-COMMERCIAL RESEARCH COMMUNITY LICENSE AGREEMENT", "parameters_b": 12.14318592, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://stability-ai.squarespace.com/news-updates/introducing-stable-lm-2-12b", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://stability-ai.squarespace.com/news-updates/introducing-stable-lm-2-12b", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/stabilityai/stablelm-2-12b-chat", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/stabilityai/stablelm-2-12b-chat", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/stabilityai/stablelm-2-12b-chat/blob/main/LICENSE", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "google/gemma-1.1-2b-it", "slug": "google--gemma-1.1-2b-it", "name": "gemma-1.1-2b-it", "developer": "Google", "developer_id": "google", "family": "Gemma 1.1", "description": null, "release_date": "2024-04-05", "date_precision": "day", "open_weights": true, "license": "gemma", "parameters_b": 2.506172416, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.google.dev/gemma/docs/releases", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://ai.google.dev/gemma/docs/releases", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/google/gemma-1.1-2b-it", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/google/gemma-1.1-2b-it", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "google/gemma-1.1-7b-it", "slug": "google--gemma-1.1-7b-it", "name": "gemma-1.1-7b-it", "developer": "Google", "developer_id": "google", "family": "Gemma 1.1", "description": null, "release_date": "2024-04-05", "date_precision": "day", "open_weights": true, "license": "gemma", "parameters_b": 8.537680896, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.google.dev/gemma/docs/releases", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://ai.google.dev/gemma/docs/releases", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/google/gemma-1.1-7b-it", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/google/gemma-1.1-7b-it", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "alibaba/qwen1.5-110b", "slug": "alibaba--qwen1.5-110b", "name": "Qwen1.5-110B", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen1.5", "description": null, "release_date": "2024-04", "date_precision": "month", "open_weights": true, "license": "Tongyi Qianwen LICENSE AGREEMENT", "parameters_b": 111.209914368, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen1.5-110b/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen1.5-110b/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen1.5-110B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen1.5-110B", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/Qwen/Qwen1.5-110B/blob/main/LICENSE", "type": "license" }, { "label": "Developer release announcement (April 26)", "url": "https://huggingface.co/posts/JustinLin610/788974065057418", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer release blog is dated April 25; a developer Hugging Face announcement describes release on April 26. Month precision is retained because earliest external availability is unresolved." }, { "id": "alibaba/qwen1.5-110b-chat", "slug": "alibaba--qwen1.5-110b-chat", "name": "Qwen1.5-110B-Chat", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen1.5", "description": null, "release_date": "2024-04", "date_precision": "month", "open_weights": true, "license": "Tongyi Qianwen LICENSE AGREEMENT", "parameters_b": 111.209914368, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen1.5-110b/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen1.5-110b/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen1.5-110B-Chat", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen1.5-110B-Chat", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/Qwen/Qwen1.5-110B-Chat/blob/main/LICENSE", "type": "license" }, { "label": "Developer release announcement (April 26)", "url": "https://huggingface.co/posts/JustinLin610/788974065057418", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer release blog is dated April 25; a developer Hugging Face announcement describes release on April 26. Month precision is retained because earliest external availability is unresolved." }, { "id": "alibaba/qwen1.5-32b", "slug": "alibaba--qwen1.5-32b", "name": "Qwen1.5-32B", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen1.5", "description": null, "release_date": "2024-04", "date_precision": "month", "open_weights": true, "license": "Tongyi Qianwen LICENSE AGREEMENT", "parameters_b": 32.512218112, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen1.5-32b/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen1.5-32b/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen1.5-32B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen1.5-32B", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/Qwen/Qwen1.5-32B/blob/main/LICENSE", "type": "license" }, { "label": "Developer checkpoint revision history", "url": "https://huggingface.co/Qwen/Qwen1.5-32B/commits/main", "type": "repository" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The dedicated developer release blog is dated April 2; checkpoint upload history begins April 3. The release month is recorded because the exact earliest external availability day is unresolved. This size was added after the original February family release." }, { "id": "alibaba/qwen1.5-32b-chat", "slug": "alibaba--qwen1.5-32b-chat", "name": "Qwen1.5-32B-Chat", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen1.5", "description": null, "release_date": "2024-04", "date_precision": "month", "open_weights": true, "license": "Tongyi Qianwen LICENSE AGREEMENT", "parameters_b": 32.512218112, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen1.5-32b/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen1.5-32b/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen1.5-32B-Chat", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen1.5-32B-Chat", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/Qwen/Qwen1.5-32B-Chat/blob/main/LICENSE", "type": "license" }, { "label": "Developer checkpoint revision history", "url": "https://huggingface.co/Qwen/Qwen1.5-32B-Chat/commits/main", "type": "repository" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The dedicated developer release blog is dated April 2; checkpoint upload history begins April 3. The release month is recorded because the exact earliest external availability day is unresolved. This size was added after the original February family release." }, { "id": "apple/openelm-1_1b", "slug": "apple--openelm-1_1b", "name": "OpenELM-1_1B", "developer": "Apple", "developer_id": "apple", "family": "OpenELM", "description": null, "release_date": "2024-04", "date_precision": "month", "open_weights": true, "license": "apple-amlr", "parameters_b": 1.079891456, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://machinelearning.apple.com/research/openelm", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://machinelearning.apple.com/research/openelm", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/apple/OpenELM-1_1B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/apple/OpenELM-1_1B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "Release month verified from the developer source; an exact public availability day is not established here." }, { "id": "apple/openelm-1_1b-instruct", "slug": "apple--openelm-1_1b-instruct", "name": "OpenELM-1_1B-Instruct", "developer": "Apple", "developer_id": "apple", "family": "OpenELM", "description": null, "release_date": "2024-04", "date_precision": "month", "open_weights": true, "license": "apple-amlr", "parameters_b": 1.079891456, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://machinelearning.apple.com/research/openelm", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://machinelearning.apple.com/research/openelm", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/apple/OpenELM-1_1B-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/apple/OpenELM-1_1B-Instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "Release month verified from the developer source; an exact public availability day is not established here." }, { "id": "apple/openelm-270m", "slug": "apple--openelm-270m", "name": "OpenELM-270M", "developer": "Apple", "developer_id": "apple", "family": "OpenELM", "description": null, "release_date": "2024-04", "date_precision": "month", "open_weights": true, "license": "apple-amlr", "parameters_b": 0.271527168, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://machinelearning.apple.com/research/openelm", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://machinelearning.apple.com/research/openelm", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/apple/OpenELM-270M", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/apple/OpenELM-270M", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "Release month verified from the developer source; an exact public availability day is not established here." }, { "id": "apple/openelm-270m-instruct", "slug": "apple--openelm-270m-instruct", "name": "OpenELM-270M-Instruct", "developer": "Apple", "developer_id": "apple", "family": "OpenELM", "description": null, "release_date": "2024-04", "date_precision": "month", "open_weights": true, "license": "apple-amlr", "parameters_b": 0.271527168, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://machinelearning.apple.com/research/openelm", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://machinelearning.apple.com/research/openelm", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/apple/OpenELM-270M-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/apple/OpenELM-270M-Instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "Release month verified from the developer source; an exact public availability day is not established here." }, { "id": "apple/openelm-3b", "slug": "apple--openelm-3b", "name": "OpenELM-3B", "developer": "Apple", "developer_id": "apple", "family": "OpenELM", "description": null, "release_date": "2024-04", "date_precision": "month", "open_weights": true, "license": "apple-amlr", "parameters_b": 3.036647424, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://machinelearning.apple.com/research/openelm", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://machinelearning.apple.com/research/openelm", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/apple/OpenELM-3B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/apple/OpenELM-3B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "Release month verified from the developer source; an exact public availability day is not established here." }, { "id": "apple/openelm-3b-instruct", "slug": "apple--openelm-3b-instruct", "name": "OpenELM-3B-Instruct", "developer": "Apple", "developer_id": "apple", "family": "OpenELM", "description": null, "release_date": "2024-04", "date_precision": "month", "open_weights": true, "license": "apple-amlr", "parameters_b": 3.036647424, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://machinelearning.apple.com/research/openelm", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://machinelearning.apple.com/research/openelm", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/apple/OpenELM-3B-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/apple/OpenELM-3B-Instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "Release month verified from the developer source; an exact public availability day is not established here." }, { "id": "apple/openelm-450m", "slug": "apple--openelm-450m", "name": "OpenELM-450M", "developer": "Apple", "developer_id": "apple", "family": "OpenELM", "description": null, "release_date": "2024-04", "date_precision": "month", "open_weights": true, "license": "apple-amlr", "parameters_b": 0.457179136, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://machinelearning.apple.com/research/openelm", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://machinelearning.apple.com/research/openelm", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/apple/OpenELM-450M", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/apple/OpenELM-450M", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "Release month verified from the developer source; an exact public availability day is not established here." }, { "id": "apple/openelm-450m-instruct", "slug": "apple--openelm-450m-instruct", "name": "OpenELM-450M-Instruct", "developer": "Apple", "developer_id": "apple", "family": "OpenELM", "description": null, "release_date": "2024-04", "date_precision": "month", "open_weights": true, "license": "apple-amlr", "parameters_b": 0.457179136, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://machinelearning.apple.com/research/openelm", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://machinelearning.apple.com/research/openelm", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/apple/OpenELM-450M-Instruct", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/apple/OpenELM-450M-Instruct", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "Release month verified from the developer source; an exact public availability day is not established here." }, { "id": "h2oai/h2o-danube2-1.8b-base", "slug": "h2oai--h2o-danube2-1.8b-base", "name": "h2o-danube2-1.8b-base", "developer": "H2O.ai", "developer_id": "h2oai", "family": "H2O Danube2", "description": null, "release_date": "2024-04", "date_precision": "month", "open_weights": true, "license": "apache-2.0", "parameters_b": 1.83120128, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://arxiv.org/abs/2401.16818v2", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://arxiv.org/abs/2401.16818v2", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/h2oai/h2o-danube2-1.8b-base", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/h2oai/h2o-danube2-1.8b-base", "type": "weights" }, { "label": "Developer April 10 announcement", "url": "https://h2o.ai/company/press-media/2024/H2O-Danube2-1-8B-Achieves-Top-Ranking-for-2-Billion-Parameters-Range/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The April 15 revised developer paper describes the released Danube2 base and chat models. An April 10 developer press release already describes the base model as available; month precision preserves the unresolved first-access day for each variant." }, { "id": "h2oai/h2o-danube2-1.8b-chat", "slug": "h2oai--h2o-danube2-1.8b-chat", "name": "h2o-danube2-1.8b-chat", "developer": "H2O.ai", "developer_id": "h2oai", "family": "H2O Danube2", "description": null, "release_date": "2024-04", "date_precision": "month", "open_weights": true, "license": "apache-2.0", "parameters_b": 1.83120128, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://arxiv.org/abs/2401.16818v2", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://arxiv.org/abs/2401.16818v2", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/h2oai/h2o-danube2-1.8b-chat", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/h2oai/h2o-danube2-1.8b-chat", "type": "weights" }, { "label": "Developer April 10 announcement", "url": "https://h2o.ai/company/press-media/2024/H2O-Danube2-1-8B-Achieves-Top-Ranking-for-2-Billion-Parameters-Range/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The April 15 revised developer paper describes the released Danube2 base and chat models. An April 10 developer press release already describes the base model as available; month precision preserves the unresolved first-access day for each variant." }, { "id": "h2oai/h2o-danube2-1.8b-sft", "slug": "h2oai--h2o-danube2-1.8b-sft", "name": "h2o-danube2-1.8b-sft", "developer": "H2O.ai", "developer_id": "h2oai", "family": "H2O Danube2", "description": null, "release_date": "2024-04", "date_precision": "month", "open_weights": true, "license": "apache-2.0", "parameters_b": 1.83120128, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://arxiv.org/abs/2401.16818v2", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://arxiv.org/abs/2401.16818v2", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/h2oai/h2o-danube2-1.8b-sft", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/h2oai/h2o-danube2-1.8b-sft", "type": "weights" }, { "label": "Developer April 10 announcement", "url": "https://h2o.ai/company/press-media/2024/H2O-Danube2-1-8B-Achieves-Top-Ranking-for-2-Billion-Parameters-Range/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The April 15 revised developer paper describes the released Danube2 base and chat models. An April 10 developer press release already describes the base model as available; month precision preserves the unresolved first-access day for each variant." }, { "id": "alibaba/qwen1.5-moe-a2.7b", "slug": "alibaba--qwen1.5-moe-a2.7b", "name": "Qwen1.5-MoE-A2.7B", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen1.5", "description": null, "release_date": "2024-03-28", "date_precision": "day", "open_weights": true, "license": "Tongyi Qianwen LICENSE AGREEMENT", "parameters_b": 14.315784192, "active_parameters_b": 2.7, "parameters_source": "checkpoint_metadata", "architecture": "moe", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen-moe/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen-moe/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen1.5-MoE-A2.7B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen1.5-MoE-A2.7B", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/Qwen/Qwen1.5-MoE-A2.7B/blob/main/LICENSE", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "alibaba/qwen1.5-moe-a2.7b-chat", "slug": "alibaba--qwen1.5-moe-a2.7b-chat", "name": "Qwen1.5-MoE-A2.7B-Chat", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen1.5", "description": null, "release_date": "2024-03-28", "date_precision": "day", "open_weights": true, "license": "Tongyi Qianwen LICENSE AGREEMENT", "parameters_b": 14.315784192, "active_parameters_b": 2.7, "parameters_source": "checkpoint_metadata", "architecture": "moe", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen-moe/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen-moe/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen1.5-MoE-A2.7B-Chat", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen1.5-MoE-A2.7B-Chat", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/Qwen/Qwen1.5-MoE-A2.7B-Chat/blob/main/LICENSE", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "01-ai/yi-9b-200k", "slug": "01-ai--yi-9b-200k", "name": "Yi-9B-200K", "developer": "01.AI", "developer_id": "01-ai", "family": "Yi", "description": null, "release_date": "2024-03-16", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 8.829407232, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 262144, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/01-ai/Yi", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/01-ai/Yi", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/01-ai/Yi-9B-200K", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/01-ai/Yi-9B-200K", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "anthropic/claude-3-haiku-20240307", "slug": "anthropic--claude-3-haiku-20240307", "name": "Claude Haiku 3", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude-haiku", "description": "Anthropic's compact Claude 3 model for fast text and image tasks, with a 200K-token context window and tool use. Retired from the Claude API on April 20, 2026.", "release_date": "2024-03-13", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 200000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://www.anthropic.com/news/claude-3-haiku", "source_type": "primary", "sources": [ { "label": "Anthropic Haiku release announcement", "url": "https://www.anthropic.com/news/claude-3-haiku", "type": "primary" }, { "label": "Claude 3 model card and later comparisons", "url": "https://assets.claude.com/c6a80a657af445f40e31afac050f3bf76d3b1404.pdf", "type": "primary" }, { "label": "Anthropic model deprecations", "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "type": "primary" }, { "label": "Unsupervised Elicitation proprietary-model classification", "url": "https://alignment.anthropic.com/2025/unsupervised-elicitation/paper.pdf", "type": "primary" }, { "label": "Anthropic production-model interpretability research", "url": "https://transformer-circuits.pub/2025/introspection/index.html", "type": "primary" }, { "label": "Google Cloud retained Haiku 3 capabilities", "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/deprecations/partner-models", "type": "primary" }, { "label": "AWS PDF launch support scope", "url": "https://aws.amazon.com/about-aws/whats-new/2025/06/citations-api-pdf-claude-models-amazon-bedrock/", "type": "primary" }, { "label": "Claude 3.5 Haiku on AWS Trainium2 and model distillation in Amazon Bedrock", "url": "https://claude.com/resources/articles/trainium2-and-distillation", "type": "primary" }, { "label": "Fine-tune Claude 3 Haiku in Amazon Bedrock", "url": "https://claude.com/resources/articles/fine-tune-claude-3-haiku", "type": "primary" }, { "label": "Claude Haiku 4.5 System Card", "url": "https://assets.anthropic.com/m/99128ddd009bdcb/original/Claude-Haiku-4-5-System-Card.pdf", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/anthropic/claude-3-haiku-20240307.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Publicly released March 13, 2024 after the March 4 family announcement; 20240307 is the snapshot identifier. Nominal total and active parameter counts remain undisclosed; no named, numerically sized base supports a lineage inference. The circulating 20B figure is an unofficial estimate. Dense/MoE routing is undisclosed. Transformer classification is supported by Anthropic's exact-model interpretability research. PDF is retained as a provider-documented input format from Google's Haiku 3 Documents row; native PDF-block support on the Anthropic API is not established, and AWS's June 2025 PDF launch listed newer models only. Claude API retirement was April 20, 2026; historical capability documentation does not establish current partner endpoint availability.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T10:20:18.112972Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/093.json" }, { "id": "01-ai/yi-9b", "slug": "01-ai--yi-9b", "name": "Yi-9B", "developer": "01.AI", "developer_id": "01-ai", "family": "Yi", "description": null, "release_date": "2024-03-06", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 8.829407232, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/01-ai/Yi", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/01-ai/Yi", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/01-ai/Yi-9B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/01-ai/Yi-9B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "anthropic/claude-3-opus-20240229", "slug": "anthropic--claude-3-opus-20240229", "name": "Claude Opus 3", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude", "description": "Anthropic's Claude 3 Opus model accepts text and images and generates text, with a 200,000-token standard context and tool calling.", "release_date": "2024-03-04", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 200000, "modalities": { "input": [ "text", "image", "pdf" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://www.anthropic.com/news/claude-3-family", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://www.anthropic.com/news/claude-3-family", "type": "announcement" }, { "label": "Claude 3 original model card (indexed text)", "url": "https://www-cdn.anthropic.com/f2986af8d052f26236f6251da62d16172cfabd6e/claude-3-model-card.pdf", "type": "model_card" }, { "label": "Exact-model interpretability research", "url": "https://transformer-circuits.pub/2025/introspection/index.html", "type": "paper" }, { "label": "Claude 3 tool-use general availability", "url": "https://claude.com/resources/articles/tool-use-ga", "type": "announcement" }, { "label": "Opus 3 post-retirement access update", "url": "https://www.anthropic.com/news/deprecation-updates-opus-3", "type": "announcement" }, { "label": "Retained provider specifications", "url": "https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/deprecations/partner-models", "type": "documentation" }, { "label": "Complete Claude 3 model card with later addenda", "url": "https://www-cdn.anthropic.com/c6a80a657af445f40e31afac050f3bf76d3b1404.pdf", "type": "model_card" }, { "label": "Claude 3.7 later comparison card", "url": "https://www-cdn.anthropic.com/9ff93dfa8f445c932415d335c88852ef47f1201e.pdf", "type": "model_card" }, { "label": "Claude 4 later Opus 3 comparison tables", "url": "https://www.anthropic.com/claude-4-system-card", "type": "model_card" }, { "label": "Claude 3 Opus RSP evaluations report", "url": "https://cdn.sanity.io/files/4zrzovbb/website/210523b8e11b09c704c5e185fd362fe9e648d457.pdf", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "Nominal total and active parameter counts remain undisclosed, with no identified sized-base lineage; generic Transformer structure is documented, but dense/MoE classification is undisclosed. Standard context is 200K; the launch offered 1M for specific uses by inquiry. Formal Anthropic API retirement was January 5, 2026, but the February 25 update documents continued paid claude.ai and by-request API access. No endpoint was tested. Dedicated extended thinking is distinct from the ordinary thinking tags documented for Opus 3 tool use. PDF is a provider-documented input format from Google; native Anthropic PDF-block support is not established. Proprietary denotes service access, not a model-weight redistribution license.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T10:20:57.682832Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/094.json" }, { "id": "anthropic/claude-3-sonnet-20240229", "slug": "anthropic--claude-3-sonnet-20240229", "name": "Claude Sonnet 3", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude", "description": "Historical Anthropic Claude model with text and image inputs, text output, a 200,000-token context window, and tool calling.", "release_date": "2024-03-04", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 200000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://www.anthropic.com/news/claude-3-family", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://www.anthropic.com/news/claude-3-family", "type": "announcement" }, { "label": "Exact-model architecture and disclosure policy", "url": "https://transformer-circuits.pub/2024/scaling-monosemanticity/index.html", "type": "technical_report" }, { "label": "Tool use general availability", "url": "https://claude.com/resources/articles/tool-use-ga", "type": "announcement" }, { "label": "Introduction of hybrid reasoning", "url": "https://www.anthropic.com/news/claude-3-7-sonnet", "type": "announcement" }, { "label": "Exact API model retirement", "url": "https://docs.anthropic.com/en/docs/about-claude/model-deprecations", "type": "documentation" }, { "label": "Claude 3 model card", "url": "https://assets.anthropic.com/m/61e7d27f8c8f5919/original/Claude-3-Model-Card.pdf", "type": "model_card" }, { "label": "Complete Claude 3 model card with later addenda", "url": "https://www-cdn.anthropic.com/c6a80a657af445f40e31afac050f3bf76d3b1404.pdf", "type": "model_card" }, { "label": "Later exact Sonnet 3 comparison tables", "url": "https://www-cdn.anthropic.com/9ff93dfa8f445c932415d335c88852ef47f1201e.pdf", "type": "model_card" }, { "label": "Later exact-model introspection and base-lineage research", "url": "https://transformer-circuits.pub/2025/introspection/index.html", "type": "technical_report" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "Anthropic explicitly withholds Claude 3 Sonnet's size; total and active parameters remain undisclosed, with no published sized-base lineage. Its exact-model interpretability research confirms Transformer structure without identifying dense versus MoE. Public launch was March 4, 2024; the API snapshot suffix is 20240229. Tool use became generally available May 30, 2024. The reasoning flag denotes dedicated extended thinking, not ordinary reasoning ability. Anthropic records this API model as retired July 21, 2025.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed" }, "audited_at": "2026-10-27T10:23:19.231300Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/095.json" }, { "id": "mistral/mistral-large", "slug": "mistral--mistral-large", "name": "Mistral Large", "developer": "Mistral", "developer_id": "mistral", "family": "mistral-large", "description": "Mistral Large alias, documented by OpenRouter as the dense 123B Mistral Large 2 (2407) text model with function calling.", "release_date": "2024-02-26", "date_precision": "day", "open_weights": true, "license": "Mistral Research License", "parameters_b": 123, "active_parameters_b": 123, "parameters_source": "primary_source", "architecture": "dense", "context_window": 128000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://openrouter.ai/api/v1/models/mistralai/mistral-large/endpoints", "source_type": "primary", "sources": [ { "label": "OpenRouter exact alias and documented 2407 mapping", "url": "https://openrouter.ai/api/v1/models/mistralai/mistral-large/endpoints", "type": "primary" }, { "label": "Original Mistral Large launch and alias history", "url": "https://mistral.ai/news/mistral-large/", "type": "primary" }, { "label": "Mistral Large 2.0 exact total and active parameters", "url": "https://docs.mistral.ai/models/mistral-large-2-0-24-07", "type": "primary" }, { "label": "Mistral Large 2407 official dense model card", "url": "https://huggingface.co/mistralai/Mistral-Large-Instruct-2407/blob/a286006d554cb37a61d13c7ae61bc90cc1d372fc/README.md", "type": "primary" }, { "label": "Mistral Large 2 release and weight license", "url": "https://mistral.ai/news/mistral-large-2407/", "type": "primary" }, { "label": "Mistral Large 2 native text modalities", "url": "https://docs.mistral.ai/models/mistral-large-2-0-24-07.md", "type": "primary" }, { "label": "OpenRouter provider PDF preprocessing", "url": "https://openrouter.ai/docs/guides/overview/multimodal/pdfs", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/mistral-large.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Undated OpenRouter-derived alias: February 26, 2024 is its original-name/listing history. OpenRouter currently describes revision 2407, released July 24, 2024; dense architecture and nominal 123B total/123B active, open weights and MRL refer to that documented revision, not original 2402. Actual backend weights were not verified; OpenRouter's later knowledge cutoff and Mistral's retired 2407 endpoint conflict with a literal live-revision reading. Native input is text; PDF is provider parsing. Dedicated reasoning-mode support remains unverified.", "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-26T23:15:48.738Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/332.json" }, { "id": "google/gemma-2b", "slug": "google--gemma-2b", "name": "gemma-2b", "developer": "Google", "developer_id": "google", "family": "Gemma", "description": null, "release_date": "2024-02-21", "date_precision": "day", "open_weights": true, "license": "gemma", "parameters_b": 2.506172416, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://cloud.google.com/blog/products/ai-machine-learning/gemma-model-available-in-vertex-ai-and-via-gke", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://cloud.google.com/blog/products/ai-machine-learning/gemma-model-available-in-vertex-ai-and-via-gke", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/google/gemma-2b", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/google/gemma-2b", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "google/gemma-2b-it", "slug": "google--gemma-2b-it", "name": "gemma-2b-it", "developer": "Google", "developer_id": "google", "family": "Gemma", "description": null, "release_date": "2024-02-21", "date_precision": "day", "open_weights": true, "license": "gemma", "parameters_b": 2.506172416, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://cloud.google.com/blog/products/ai-machine-learning/gemma-model-available-in-vertex-ai-and-via-gke", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://cloud.google.com/blog/products/ai-machine-learning/gemma-model-available-in-vertex-ai-and-via-gke", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/google/gemma-2b-it", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/google/gemma-2b-it", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "google/gemma-7b", "slug": "google--gemma-7b", "name": "gemma-7b", "developer": "Google", "developer_id": "google", "family": "Gemma", "description": null, "release_date": "2024-02-21", "date_precision": "day", "open_weights": true, "license": "gemma", "parameters_b": 8.537680896, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://cloud.google.com/blog/products/ai-machine-learning/gemma-model-available-in-vertex-ai-and-via-gke", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://cloud.google.com/blog/products/ai-machine-learning/gemma-model-available-in-vertex-ai-and-via-gke", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/google/gemma-7b", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/google/gemma-7b", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "google/gemma-7b-it", "slug": "google--gemma-7b-it", "name": "gemma-7b-it", "developer": "Google", "developer_id": "google", "family": "Gemma", "description": null, "release_date": "2024-02-21", "date_precision": "day", "open_weights": true, "license": "gemma", "parameters_b": 8.537680896, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://cloud.google.com/blog/products/ai-machine-learning/gemma-model-available-in-vertex-ai-and-via-gke", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://cloud.google.com/blog/products/ai-machine-learning/gemma-model-available-in-vertex-ai-and-via-gke", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/google/gemma-7b-it", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/google/gemma-7b-it", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "internlm/internlm2-chat-1_8b", "slug": "internlm--internlm2-chat-1_8b", "name": "internlm2-chat-1_8b", "developer": "Shanghai AI Laboratory", "developer_id": "internlm", "family": "InternLM2", "description": null, "release_date": "2024-02-19", "date_precision": "day", "open_weights": true, "license": null, "parameters_b": 1.889110016, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/InternLM/InternLM", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/InternLM/InternLM", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/internlm/internlm2-chat-1_8b", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/internlm/internlm2-chat-1_8b", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer card describes separate model-use terms and code licensing but gives no named model license; no license label is inferred." }, { "id": "google/gemini-1.0-ultra", "slug": "google--gemini-1.0-ultra", "name": "Gemini 1.0 Ultra", "developer": "Google", "developer_id": "google", "family": "gemini", "description": null, "release_date": "2024-02-08", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "unknown", "context_window": 32768, "modalities": { "input": [ "text", "image", "audio", "video" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://blog.google/innovation-and-ai/technology/ai/google-gemini-update-sundar-pichai-2024/", "source_type": "primary", "sources": [ { "label": "First verified public access through GeminiAdvanced", "url": "https://blog.google/innovation-and-ai/technology/ai/google-gemini-update-sundar-pichai-2024/", "type": "announcement" }, { "label": "Gemini Advanced release details", "url": "https://blog.google/intl/en-africa/products/explore-get-answers/bard-becomes-gemini-try-ultra-10-and-a-new-mobile-app-today/", "type": "announcement" }, { "label": "Original Gemini1.0 technical report", "url": "https://arxiv.org/abs/2312.11805v1", "type": "paper" }, { "label": "Original family announcement", "url": "https://blog.google/innovation-and-ai/technology/ai/google-gemini-ai/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:26:25.003Z", "last_seen": "2026-10-29T10:26:25.003Z", "upstream_present": false, "announcement_date": "2023-12-06", "availability": "public", "notes": "Ultra 1.0 was announced with the Gemini family in December 2023 and became publicly usable through Gemini Advanced on February 8, 2024. Google did not disclose its parameter count. The 32768-token context and multimodal inputs describe the model in the primary technical report; consumer product features and limits may differ." }, { "id": "allenai/olmo-1b", "slug": "allenai--olmo-1b", "name": "OLMo-1B", "developer": "Ai2", "developer_id": "allenai", "family": "OLMo", "description": null, "release_date": "2024-02-01", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 1.176764416, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://docs.allenai.org/release_notes/olmo-release-notes", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://docs.allenai.org/release_notes/olmo-release-notes", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/allenai/OLMo-1B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/allenai/OLMo-1B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "allenai/olmo-7b", "slug": "allenai--olmo-7b", "name": "OLMo-7B", "developer": "Ai2", "developer_id": "allenai", "family": "OLMo", "description": null, "release_date": "2024-02-01", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": 6.888095744, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://docs.allenai.org/release_notes/olmo-release-notes", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://docs.allenai.org/release_notes/olmo-release-notes", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/allenai/OLMo-7B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/allenai/OLMo-7B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "openbmb/minicpm-2b-sft", "slug": "openbmb--minicpm-2b-sft", "name": "MiniCPM-2B-sft", "developer": "OpenBMB", "developer_id": "openbmb", "family": "MiniCPM", "description": null, "release_date": "2024-02-01", "date_precision": "day", "open_weights": true, "license": "General Model License Agreement - Source Notes - Publicity Restrictions - Commercial License", "parameters_b": 2.4, "active_parameters_b": null, "parameters_source": "primary", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/OpenBMB/MiniCPM", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/OpenBMB/MiniCPM", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/openbmb/MiniCPM-2B-sft-bf16", "type": "model_card" }, { "label": "Model license", "url": "https://github.com/OpenBMB/General-Model-License/blob/main/%E9%80%9A%E7%94%A8%E6%A8%A1%E5%9E%8B%E8%AE%B8%E5%8F%AF%E5%8D%8F%E8%AE%AE-%E6%9D%A5%E6%BA%90%E8%AF%B4%E6%98%8E-%E5%AE%A3%E4%BC%A0%E9%99%90%E5%88%B6-%E5%95%86%E4%B8%9A%E6%8E%88%E6%9D%83.md", "type": "license" } ], "aliases": [ "MiniCPM-2B-sft-bf16" ], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "parameters_scope": "non_embedding", "parameter_components_b": { "non_embedding": 2.4 }, "notes": "The developer parameter count excludes embeddings; it is stored as a scoped component rather than a total." }, { "id": "openbmb/minicpm-v", "slug": "openbmb--minicpm-v", "name": "MiniCPM-V", "developer": "OpenBMB", "developer_id": "openbmb", "family": "MiniCPM-V", "description": null, "release_date": "2024-02-01", "date_precision": "day", "open_weights": true, "license": null, "parameters_b": 3.434965792, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/OpenBMB/MiniCPM-V/blob/main/docs/changelog_en.md", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/OpenBMB/MiniCPM-V/blob/main/docs/changelog_en.md", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/openbmb/MiniCPM-V", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/openbmb/MiniCPM-V", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "openbmb/omnilmm-12b", "slug": "openbmb--omnilmm-12b", "name": "OmniLMM 12B", "developer": "OpenBMB", "developer_id": "openbmb", "family": "omnilmm", "description": null, "release_date": "2024-02-01", "date_precision": "day", "open_weights": true, "license": "General Model License — Attribution, Publicity Restriction and Commercial Authorization", "parameters_b": 11.6, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/openbmb/OmniLMM-12B", "source_type": "primary", "sources": [ { "label": "Publisher dated release log", "url": "https://github.com/OpenBMB/MiniCPM-V/blob/main/docs/changelog_en.md", "type": "announcement" }, { "label": "Publisher model card", "url": "https://huggingface.co/openbmb/OmniLMM-12B/blob/d2e9c9d619a0b34f06f0b65d2eae9ed1af1124a9/README.md", "type": "model_card" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/openbmb/OmniLMM-12B/blob/d2e9c9d619a0b34f06f0b65d2eae9ed1af1124a9/config.json", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T11:22:26.610Z", "last_seen": "2026-10-29T11:22:26.610Z", "upstream_present": false, "availability": "public", "notes": "OpenBMB trained multimodal adaptation of Zephyr-7B-beta plus EVA02-5B and a perceiver resampler. The model card reports 11.6B total parameters; the 12B name is rounded. This record describes the trained image-language model, not its optional demonstration assistant that also calls GPT-3.5." }, { "id": "alibaba/qwen1.5-0.5b", "slug": "alibaba--qwen1.5-0.5b", "name": "Qwen1.5-0.5B", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen1.5", "description": null, "release_date": "2024-02", "date_precision": "month", "open_weights": true, "license": "Tongyi Qianwen RESEARCH LICENSE AGREEMENT", "parameters_b": 0.619570176, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen1.5/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen1.5/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen1.5-0.5B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen1.5-0.5B", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/Qwen/Qwen1.5-0.5B/blob/main/LICENSE", "type": "license" }, { "label": "Developer release changelog (February 5)", "url": "https://github.com/QwenLM/Qwen3", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The original Qwen1.5 release blog is dated February 4, while the developer repository changelog dates the release February 5. The release month is recorded because the earliest external availability day is unresolved. Later-added model sizes are not assigned this family date." }, { "id": "alibaba/qwen1.5-0.5b-chat", "slug": "alibaba--qwen1.5-0.5b-chat", "name": "Qwen1.5-0.5B-Chat", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen1.5", "description": null, "release_date": "2024-02", "date_precision": "month", "open_weights": true, "license": "Tongyi Qianwen RESEARCH LICENSE AGREEMENT", "parameters_b": 0.619570176, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen1.5/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen1.5/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen1.5-0.5B-Chat", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen1.5-0.5B-Chat", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/Qwen/Qwen1.5-0.5B-Chat/blob/main/LICENSE", "type": "license" }, { "label": "Developer release changelog (February 5)", "url": "https://github.com/QwenLM/Qwen3", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The original Qwen1.5 release blog is dated February 4, while the developer repository changelog dates the release February 5. The release month is recorded because the earliest external availability day is unresolved. Later-added model sizes are not assigned this family date." }, { "id": "alibaba/qwen1.5-1.8b", "slug": "alibaba--qwen1.5-1.8b", "name": "Qwen1.5-1.8B", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen1.5", "description": null, "release_date": "2024-02", "date_precision": "month", "open_weights": true, "license": "Tongyi Qianwen RESEARCH LICENSE AGREEMENT", "parameters_b": 1.836828672, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen1.5/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen1.5/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen1.5-1.8B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen1.5-1.8B", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/Qwen/Qwen1.5-1.8B/blob/main/LICENSE", "type": "license" }, { "label": "Developer release changelog (February 5)", "url": "https://github.com/QwenLM/Qwen3", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The original Qwen1.5 release blog is dated February 4, while the developer repository changelog dates the release February 5. The release month is recorded because the earliest external availability day is unresolved. Later-added model sizes are not assigned this family date." }, { "id": "alibaba/qwen1.5-1.8b-chat", "slug": "alibaba--qwen1.5-1.8b-chat", "name": "Qwen1.5-1.8B-Chat", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen1.5", "description": null, "release_date": "2024-02", "date_precision": "month", "open_weights": true, "license": "Tongyi Qianwen RESEARCH LICENSE AGREEMENT", "parameters_b": 1.836828672, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen1.5/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen1.5/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen1.5-1.8B-Chat", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen1.5-1.8B-Chat", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/Qwen/Qwen1.5-1.8B-Chat/blob/main/LICENSE", "type": "license" }, { "label": "Developer release changelog (February 5)", "url": "https://github.com/QwenLM/Qwen3", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The original Qwen1.5 release blog is dated February 4, while the developer repository changelog dates the release February 5. The release month is recorded because the earliest external availability day is unresolved. Later-added model sizes are not assigned this family date." }, { "id": "alibaba/qwen1.5-14b", "slug": "alibaba--qwen1.5-14b", "name": "Qwen1.5-14B", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen1.5", "description": null, "release_date": "2024-02", "date_precision": "month", "open_weights": true, "license": "Tongyi Qianwen LICENSE AGREEMENT", "parameters_b": 14.16729088, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen1.5/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen1.5/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen1.5-14B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen1.5-14B", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/Qwen/Qwen1.5-14B/blob/main/LICENSE", "type": "license" }, { "label": "Developer release changelog (February 5)", "url": "https://github.com/QwenLM/Qwen3", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The original Qwen1.5 release blog is dated February 4, while the developer repository changelog dates the release February 5. The release month is recorded because the earliest external availability day is unresolved. Later-added model sizes are not assigned this family date." }, { "id": "alibaba/qwen1.5-14b-chat", "slug": "alibaba--qwen1.5-14b-chat", "name": "Qwen1.5-14B-Chat", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen1.5", "description": null, "release_date": "2024-02", "date_precision": "month", "open_weights": true, "license": "Tongyi Qianwen LICENSE AGREEMENT", "parameters_b": 14.16729088, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen1.5/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen1.5/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen1.5-14B-Chat", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen1.5-14B-Chat", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/Qwen/Qwen1.5-14B-Chat/blob/main/LICENSE", "type": "license" }, { "label": "Developer release changelog (February 5)", "url": "https://github.com/QwenLM/Qwen3", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The original Qwen1.5 release blog is dated February 4, while the developer repository changelog dates the release February 5. The release month is recorded because the earliest external availability day is unresolved. Later-added model sizes are not assigned this family date." }, { "id": "alibaba/qwen1.5-4b", "slug": "alibaba--qwen1.5-4b", "name": "Qwen1.5-4B", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen1.5", "description": null, "release_date": "2024-02", "date_precision": "month", "open_weights": true, "license": "Tongyi Qianwen RESEARCH LICENSE AGREEMENT", "parameters_b": 3.95036928, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen1.5/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen1.5/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen1.5-4B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen1.5-4B", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/Qwen/Qwen1.5-4B/blob/main/LICENSE", "type": "license" }, { "label": "Developer release changelog (February 5)", "url": "https://github.com/QwenLM/Qwen3", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The original Qwen1.5 release blog is dated February 4, while the developer repository changelog dates the release February 5. The release month is recorded because the earliest external availability day is unresolved. Later-added model sizes are not assigned this family date." }, { "id": "alibaba/qwen1.5-4b-chat", "slug": "alibaba--qwen1.5-4b-chat", "name": "Qwen1.5-4B-Chat", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen1.5", "description": null, "release_date": "2024-02", "date_precision": "month", "open_weights": true, "license": "Tongyi Qianwen RESEARCH LICENSE AGREEMENT", "parameters_b": 3.95036928, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen1.5/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen1.5/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen1.5-4B-Chat", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen1.5-4B-Chat", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/Qwen/Qwen1.5-4B-Chat/blob/main/LICENSE", "type": "license" }, { "label": "Developer release changelog (February 5)", "url": "https://github.com/QwenLM/Qwen3", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The original Qwen1.5 release blog is dated February 4, while the developer repository changelog dates the release February 5. The release month is recorded because the earliest external availability day is unresolved. Later-added model sizes are not assigned this family date." }, { "id": "alibaba/qwen1.5-72b", "slug": "alibaba--qwen1.5-72b", "name": "Qwen1.5-72B", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen1.5", "description": null, "release_date": "2024-02", "date_precision": "month", "open_weights": true, "license": "Tongyi Qianwen LICENSE AGREEMENT", "parameters_b": 72.287920128, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen1.5/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen1.5/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen1.5-72B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen1.5-72B", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/Qwen/Qwen1.5-72B/blob/main/LICENSE", "type": "license" }, { "label": "Developer release changelog (February 5)", "url": "https://github.com/QwenLM/Qwen3", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The original Qwen1.5 release blog is dated February 4, while the developer repository changelog dates the release February 5. The release month is recorded because the earliest external availability day is unresolved. Later-added model sizes are not assigned this family date." }, { "id": "alibaba/qwen1.5-72b-chat", "slug": "alibaba--qwen1.5-72b-chat", "name": "Qwen1.5-72B-Chat", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen1.5", "description": null, "release_date": "2024-02", "date_precision": "month", "open_weights": true, "license": "Tongyi Qianwen LICENSE AGREEMENT", "parameters_b": 72.287920128, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen1.5/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen1.5/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen1.5-72B-Chat", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen1.5-72B-Chat", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/Qwen/Qwen1.5-72B-Chat/blob/main/LICENSE", "type": "license" }, { "label": "Developer release changelog (February 5)", "url": "https://github.com/QwenLM/Qwen3", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The original Qwen1.5 release blog is dated February 4, while the developer repository changelog dates the release February 5. The release month is recorded because the earliest external availability day is unresolved. Later-added model sizes are not assigned this family date." }, { "id": "alibaba/qwen1.5-7b", "slug": "alibaba--qwen1.5-7b", "name": "Qwen1.5-7B", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen1.5", "description": null, "release_date": "2024-02", "date_precision": "month", "open_weights": true, "license": "Tongyi Qianwen LICENSE AGREEMENT", "parameters_b": 7.721324544, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen1.5/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen1.5/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen1.5-7B", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen1.5-7B", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/Qwen/Qwen1.5-7B/blob/main/LICENSE", "type": "license" }, { "label": "Developer release changelog (February 5)", "url": "https://github.com/QwenLM/Qwen3", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The original Qwen1.5 release blog is dated February 4, while the developer repository changelog dates the release February 5. The release month is recorded because the earliest external availability day is unresolved. Later-added model sizes are not assigned this family date." }, { "id": "alibaba/qwen1.5-7b-chat", "slug": "alibaba--qwen1.5-7b-chat", "name": "Qwen1.5-7B-Chat", "developer": "Alibaba", "developer_id": "alibaba", "family": "Qwen1.5", "description": null, "release_date": "2024-02", "date_precision": "month", "open_weights": true, "license": "Tongyi Qianwen LICENSE AGREEMENT", "parameters_b": 7.721324544, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://qwenlm.github.io/blog/qwen1.5/", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://qwenlm.github.io/blog/qwen1.5/", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/Qwen/Qwen1.5-7B-Chat", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/Qwen/Qwen1.5-7B-Chat", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/Qwen/Qwen1.5-7B-Chat/blob/main/LICENSE", "type": "license" }, { "label": "Developer release changelog (February 5)", "url": "https://github.com/QwenLM/Qwen3", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The original Qwen1.5 release blog is dated February 4, while the developer repository changelog dates the release February 5. The release month is recorded because the earliest external availability day is unresolved. Later-added model sizes are not assigned this family date." }, { "id": "internlm/internlm2-1_8b", "slug": "internlm--internlm2-1_8b", "name": "internlm2-1_8b", "developer": "Shanghai AI Laboratory", "developer_id": "internlm", "family": "InternLM2", "description": null, "release_date": "2024-01-31", "date_precision": "day", "open_weights": true, "license": null, "parameters_b": 1.8, "active_parameters_b": null, "parameters_source": "model_name", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/InternLM/InternLM", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/InternLM/InternLM", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/internlm/internlm2-1_8b", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer card describes separate model-use terms and code licensing but gives no named model license; no license label is inferred." }, { "id": "internlm/internlm2-chat-1_8b-sft", "slug": "internlm--internlm2-chat-1_8b-sft", "name": "internlm2-chat-1_8b-sft", "developer": "Shanghai AI Laboratory", "developer_id": "internlm", "family": "InternLM2", "description": null, "release_date": "2024-01-31", "date_precision": "day", "open_weights": true, "license": null, "parameters_b": 1.889110016, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/InternLM/InternLM", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/InternLM/InternLM", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/internlm/internlm2-chat-1_8b-sft", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/internlm/internlm2-chat-1_8b-sft", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer card describes separate model-use terms and code licensing but gives no named model license; no license label is inferred." }, { "id": "meta/code-llama-70b", "slug": "meta--code-llama-70b", "name": "Code Llama 70B", "developer": "Meta", "developer_id": "meta", "family": "code-llama", "description": null, "release_date": "2024-01-29", "date_precision": "day", "open_weights": true, "license": "Llama 2 Community License", "parameters_b": 70, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/?trk=article-ssr-frontend-pulse_little-text-block", "source_type": "primary", "sources": [ { "label": "January 29 update naming all three 70B variants", "url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/?trk=article-ssr-frontend-pulse_little-text-block", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/codellama/CodeLlama-70b-hf/blob/cc11b0f79c8a072bd1e1f32ec280e1ff8ec018f6/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/codellama/CodeLlama-70b-hf/blob/cc11b0f79c8a072bd1e1f32ec280e1ff8ec018f6/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:26:23.347Z", "last_seen": "2026-10-29T10:26:23.347Z", "upstream_present": false, "availability": "public", "notes": "Meta’s continued training of its own Llama 2 checkpoint. Nominal parameter count; context follows the published configuration." }, { "id": "meta/code-llama-70b-instruct", "slug": "meta--code-llama-70b-instruct", "name": "Code Llama 70B Instruct", "developer": "Meta", "developer_id": "meta", "family": "code-llama", "description": null, "release_date": "2024-01-29", "date_precision": "day", "open_weights": true, "license": "Llama 2 Community License", "parameters_b": 70, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/?trk=article-ssr-frontend-pulse_little-text-block", "source_type": "primary", "sources": [ { "label": "January 29 update naming all three 70B variants", "url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/?trk=article-ssr-frontend-pulse_little-text-block", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/codellama/CodeLlama-70b-Instruct-hf/blob/397cae981dffaf5d5c9c90e89a0a75a850528b70/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/codellama/CodeLlama-70b-Instruct-hf/blob/397cae981dffaf5d5c9c90e89a0a75a850528b70/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:26:23.347Z", "last_seen": "2026-10-29T10:26:23.347Z", "upstream_present": false, "availability": "public", "notes": "Meta’s continued training of its own Llama 2 checkpoint. Nominal parameter count; context follows the published configuration." }, { "id": "meta/code-llama-70b-python", "slug": "meta--code-llama-70b-python", "name": "Code Llama 70B Python", "developer": "Meta", "developer_id": "meta", "family": "code-llama", "description": null, "release_date": "2024-01-29", "date_precision": "day", "open_weights": true, "license": "Llama 2 Community License", "parameters_b": 70, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/?trk=article-ssr-frontend-pulse_little-text-block", "source_type": "primary", "sources": [ { "label": "January 29 update naming all three 70B variants", "url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/?trk=article-ssr-frontend-pulse_little-text-block", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/codellama/CodeLlama-70b-Python-hf/blob/e037619532ec8449a01ecd5b253d065ac81637b6/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/codellama/CodeLlama-70b-Python-hf/blob/e037619532ec8449a01ecd5b253d065ac81637b6/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:26:23.347Z", "last_seen": "2026-10-29T10:26:23.347Z", "upstream_present": false, "availability": "public", "notes": "Meta’s continued training of its own Llama 2 checkpoint. Nominal parameter count; context follows the published configuration." }, { "id": "01-ai/yi-vl-34b", "slug": "01-ai--yi-vl-34b", "name": "Yi-VL-34B", "developer": "01.AI", "developer_id": "01-ai", "family": "Yi-VL", "description": null, "release_date": "2024-01-23", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 4096, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/01-ai/Yi", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/01-ai/Yi", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/01-ai/Yi-VL-34B", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The checkpoint name describes its text-backbone size. A complete multimodal parameter total was not established from the checked developer card and API metadata." }, { "id": "01-ai/yi-vl-6b", "slug": "01-ai--yi-vl-6b", "name": "Yi-VL-6B", "developer": "01.AI", "developer_id": "01-ai", "family": "Yi-VL", "description": null, "release_date": "2024-01-23", "date_precision": "day", "open_weights": true, "license": "apache-2.0", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 4096, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/01-ai/Yi", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/01-ai/Yi", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/01-ai/Yi-VL-6B", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The checkpoint name describes its text-backbone size. A complete multimodal parameter total was not established from the checked developer card and API metadata." }, { "id": "stabilityai/stablelm-2-1_6b", "slug": "stabilityai--stablelm-2-1_6b", "name": "stablelm-2-1_6b", "developer": "Stability AI", "developer_id": "stabilityai", "family": "Stable LM 2", "description": null, "release_date": "2024-01-19", "date_precision": "day", "open_weights": true, "license": "STABILITY AI COMMUNITY LICENSE AGREEMENT", "parameters_b": 1.644515328, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://stability.ai/news-updates/introducing-stable-lm-2", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://stability.ai/news-updates/introducing-stable-lm-2", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/stabilityai/stablelm-2-1_6b", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/stabilityai/stablelm-2-1_6b", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/stabilityai/stablelm-2-1_6b/blob/main/LICENSE.md", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "stabilityai/stablelm-2-zephyr-1_6b", "slug": "stabilityai--stablelm-2-zephyr-1_6b", "name": "stablelm-2-zephyr-1_6b", "developer": "Stability AI", "developer_id": "stabilityai", "family": "Stable LM 2", "description": null, "release_date": "2024-01-19", "date_precision": "day", "open_weights": true, "license": "STABILITY AI NON-COMMERCIAL RESEARCH COMMUNITY LICENSE AGREEMENT", "parameters_b": 1.644515328, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://stability.ai/news-updates/introducing-stable-lm-2", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://stability.ai/news-updates/introducing-stable-lm-2", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/stabilityai/stablelm-2-zephyr-1_6b", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/stabilityai/stablelm-2-zephyr-1_6b", "type": "weights" }, { "label": "Model license", "url": "https://huggingface.co/stabilityai/stablelm-2-zephyr-1_6b/blob/main/LICENSE", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public" }, { "id": "alibaba/qwen-vl-max", "slug": "alibaba--qwen-vl-max", "name": "Qwen-VL Max", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Hosted Qwen vision-language model for image, document, chart, and video understanding with text output.", "release_date": "2024-01-18", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 131072, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": false, "tool_call": false, "source_url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen-vl-max.toml", "source_type": "catalog", "sources": [ { "label": "Qwen official repository release history", "url": "https://github.com/QwenLM/Qwen-VL/blob/master/README.md", "type": "primary" }, { "label": "Qwen-VL announcement", "url": "https://qwenlm.github.io/blog/qwen-vl/", "type": "primary" }, { "label": "Alibaba model card (indexed official content)", "url": "https://docs.bailian.console.aliyun.com/zh/model-studio/qwen-vl-max", "type": "docs" }, { "label": "Alibaba alias update notice", "url": "https://www.alibabacloud.com/en/notice/model_studio_update_of_qwen_vlmax_4de", "type": "primary" }, { "label": "Qwen2-VL: To See the World More Clearly", "url": "https://qwenlm.github.io/blog/qwen2-vl/", "type": "primary" }, { "label": "Qwen2-VL: Enhancing Vision-Language Model’s Perception of the World at Any Resolution", "url": "https://arxiv.org/html/2409.12191v1", "type": "paper" }, { "label": "Qwen-VL-Max and Qwen-VL-Plus model update notice", "url": "https://www.alibabacloud.com/zh/notice/model_studio_update_of_qwenvlmax_and_qwenvlplus_587?_p_lc=1", "type": "primary" }, { "label": "Update of Qwen-VL-Max-Latest", "url": "https://www.alibabacloud.com/en/notice/model_studio_update_of_qwenvlmaxlatest_55b?_p_lc=1", "type": "primary" }, { "label": "Introduction to model fine-tuning: image and video token calculation", "url": "https://www.alibabacloud.com/help/en/model-studio/model-training-overview", "type": "docs" }, { "label": "Qwen2.5 VL! Qwen2.5 VL! Qwen2.5 VL!", "url": "https://qwenlm.github.io/blog/qwen2.5-vl/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen-vl-max.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Hosted rolling alias documented as qwen-vl-max-2025-08-13. The historical qwen-vl-max-0809 API was explicitly Qwen2-VL-72B (nominal size); later snapshot size, active count and dense/MoE topology remain undisclosed. Initial Max availability: 2024-01-18. The official card says native Function Calling is unsupported.", "field_status": { "license": "not_applicable", "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-27T07:40:02.374747Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/018.json" }, { "id": "internlm/internlm2-20b", "slug": "internlm--internlm2-20b", "name": "internlm2-20b", "developer": "Shanghai AI Laboratory", "developer_id": "internlm", "family": "InternLM2", "description": null, "release_date": "2024-01-17", "date_precision": "day", "open_weights": true, "license": null, "parameters_b": 20, "active_parameters_b": null, "parameters_source": "model_name", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/InternLM/InternLM", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/InternLM/InternLM", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/internlm/internlm2-20b", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer card describes separate model-use terms and code licensing but gives no named model license; no license label is inferred." }, { "id": "internlm/internlm2-7b", "slug": "internlm--internlm2-7b", "name": "internlm2-7b", "developer": "Shanghai AI Laboratory", "developer_id": "internlm", "family": "InternLM2", "description": null, "release_date": "2024-01-17", "date_precision": "day", "open_weights": true, "license": null, "parameters_b": 7, "active_parameters_b": null, "parameters_source": "model_name", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/InternLM/InternLM", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/InternLM/InternLM", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/internlm/internlm2-7b", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer card describes separate model-use terms and code licensing but gives no named model license; no license label is inferred." }, { "id": "internlm/internlm2-base-20b", "slug": "internlm--internlm2-base-20b", "name": "internlm2-base-20b", "developer": "Shanghai AI Laboratory", "developer_id": "internlm", "family": "InternLM2", "description": null, "release_date": "2024-01-17", "date_precision": "day", "open_weights": true, "license": null, "parameters_b": 20, "active_parameters_b": null, "parameters_source": "model_name", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/InternLM/InternLM", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/InternLM/InternLM", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/internlm/internlm2-base-20b", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer card describes separate model-use terms and code licensing but gives no named model license; no license label is inferred." }, { "id": "internlm/internlm2-base-7b", "slug": "internlm--internlm2-base-7b", "name": "internlm2-base-7b", "developer": "Shanghai AI Laboratory", "developer_id": "internlm", "family": "InternLM2", "description": null, "release_date": "2024-01-17", "date_precision": "day", "open_weights": true, "license": null, "parameters_b": 7, "active_parameters_b": null, "parameters_source": "model_name", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/InternLM/InternLM", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/InternLM/InternLM", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/internlm/internlm2-base-7b", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer card describes separate model-use terms and code licensing but gives no named model license; no license label is inferred." }, { "id": "internlm/internlm2-chat-20b", "slug": "internlm--internlm2-chat-20b", "name": "internlm2-chat-20b", "developer": "Shanghai AI Laboratory", "developer_id": "internlm", "family": "InternLM2", "description": null, "release_date": "2024-01-17", "date_precision": "day", "open_weights": true, "license": null, "parameters_b": 19.861149696, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/InternLM/InternLM", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/InternLM/InternLM", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/internlm/internlm2-chat-20b", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/internlm/internlm2-chat-20b", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer card describes separate model-use terms and code licensing but gives no named model license; no license label is inferred." }, { "id": "internlm/internlm2-chat-20b-sft", "slug": "internlm--internlm2-chat-20b-sft", "name": "internlm2-chat-20b-sft", "developer": "Shanghai AI Laboratory", "developer_id": "internlm", "family": "InternLM2", "description": null, "release_date": "2024-01-17", "date_precision": "day", "open_weights": true, "license": null, "parameters_b": 19.861149696, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/InternLM/InternLM", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/InternLM/InternLM", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/internlm/internlm2-chat-20b-sft", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/internlm/internlm2-chat-20b-sft", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer card describes separate model-use terms and code licensing but gives no named model license; no license label is inferred." }, { "id": "internlm/internlm2-chat-7b", "slug": "internlm--internlm2-chat-7b", "name": "internlm2-chat-7b", "developer": "Shanghai AI Laboratory", "developer_id": "internlm", "family": "InternLM2", "description": null, "release_date": "2024-01-17", "date_precision": "day", "open_weights": true, "license": null, "parameters_b": 7.737708544, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/InternLM/InternLM", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/InternLM/InternLM", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/internlm/internlm2-chat-7b", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/internlm/internlm2-chat-7b", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer card describes separate model-use terms and code licensing but gives no named model license; no license label is inferred." }, { "id": "internlm/internlm2-chat-7b-sft", "slug": "internlm--internlm2-chat-7b-sft", "name": "internlm2-chat-7b-sft", "developer": "Shanghai AI Laboratory", "developer_id": "internlm", "family": "InternLM2", "description": null, "release_date": "2024-01-17", "date_precision": "day", "open_weights": true, "license": null, "parameters_b": 7.737708544, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/InternLM/InternLM", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://github.com/InternLM/InternLM", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/internlm/internlm2-chat-7b-sft", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/internlm/internlm2-chat-7b-sft", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The developer card describes separate model-use terms and code licensing but gives no named model license; no license label is inferred." }, { "id": "deepseek/deepseek-chat", "slug": "deepseek--deepseek-chat", "name": "DeepSeek Chat", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek", "description": "Legacy DeepSeek chat API alias, last explicitly mapped to DeepSeek-V4-Flash preview in non-thinking mode; announced for retirement in July 2026.", "release_date": "2024-01", "date_precision": "month", "open_weights": true, "license": "MIT", "parameters_b": 284, "active_parameters_b": 13, "parameters_source": "primary_source", "architecture": "moe", "context_window": 1000000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://github.com/deepseek-ai/awesome-deepseek-integration/commit/f27a35cd49998183e5bd08618026307d36fe80c4", "source_type": "primary", "sources": [ { "label": "DeepSeek alias history", "url": "https://api-docs.deepseek.com/updates/", "type": "primary" }, { "label": "V4 preview and legacy alias mapping", "url": "https://api-docs.deepseek.com/news/news260424/", "type": "primary" }, { "label": "Last documented backing weights: V4-Flash preview", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash", "type": "weights" }, { "label": "DeepSeek-V4 technical report", "url": "https://arxiv.org/html/2606.19348v1", "type": "primary" }, { "label": "DeepSeek current API models", "url": "https://api-docs.deepseek.com/quick_start/pricing/", "type": "primary" }, { "label": "DeepSeek alias function calling announcement", "url": "https://www.deepseek.com/en/news/api-upgrade/", "type": "primary" }, { "label": "Immutable May 6, 2024 DeepSeek-V2 launch README", "url": "https://raw.githubusercontent.com/deepseek-ai/DeepSeek-V2/e23eeb51a8eaef96601f4074a0008c20c926d762/README.md", "type": "primary" }, { "label": "Original DeepSeek news archive inventory", "url": "https://www.deepseek.com/sitemap.xml", "type": "primary" }, { "label": "Original DeepSeek-LLM repository README checked", "url": "https://raw.githubusercontent.com/deepseek-ai/DeepSeek-LLM/main/README.md", "type": "primary" }, { "label": "Original January 11 DeepSeek API integration publication", "url": "https://github.com/deepseek-ai/awesome-deepseek-integration/commit/f27a35cd49998183e5bd08618026307d36fe80c4", "type": "primary" }, { "label": "Original DeepSeek Chatbox configuration image", "url": "https://github.com/deepseek-ai/awesome-deepseek-integration/assets/13600976/926a87dc-ba2c-4929-9310-c4f1a18c0ce3", "type": "primary" }, { "label": "January 18 DeepSeek API registration and integration guide", "url": "https://raw.githubusercontent.com/deepseek-ai/awesome-deepseek-integration/074784a8b52cf2163cc12c2d9220de062cdc0248/README_cn.md", "type": "primary" }, { "label": "Original January 18 integration guide commit metadata", "url": "https://github.com/deepseek-ai/awesome-deepseek-integration/commit/074784a8b52cf2163cc12c2d9220de062cdc0248", "type": "primary" }, { "label": "Immutable January 11 DeepSeek Chatbox integration guide", "url": "https://raw.githubusercontent.com/deepseek-ai/awesome-deepseek-integration/608616d698d9a51b6ac1493a247c0783f69ad70c/docs/chatbox/README.md", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/deepseek/deepseek-chat.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Specifications describe the explicit April 24, 2026 V4-Flash preview mapping. The original chat API rollout is dated to January 2024 at month precision using DeepSeek’s native integration guide and exact-alias configuration image; the public launch day remains unverified. The former December 1, 2025 date describes its V3.2 upgrade. DeepSeek announced inaccessibility after July 24, 2026, 15:59 UTC; no authenticated endpoint test was performed. Counts are developer nominal model parameters, not Hugging Face storage totals. MIT/open weights refer to the backing checkpoint.", "field_status": {}, "audited_at": "2026-10-28T08:04:32.447553Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/166.json" }, { "id": "h2oai/h2o-danube-1.8b-base", "slug": "h2oai--h2o-danube-1.8b-base", "name": "h2o-danube-1.8b-base", "developer": "H2O.ai", "developer_id": "h2oai", "family": "H2O Danube", "description": null, "release_date": "2024-01", "date_precision": "month", "open_weights": true, "license": "apache-2.0", "parameters_b": 1.83120128, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://arxiv.org/abs/2401.16818v1", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://arxiv.org/abs/2401.16818v1", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/h2oai/h2o-danube-1.8b-base", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/h2oai/h2o-danube-1.8b-base", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The January 30 developer paper explicitly makes the model and chat variant openly available; month precision avoids treating its submission timestamp as the exact public-access time." }, { "id": "h2oai/h2o-danube-1.8b-chat", "slug": "h2oai--h2o-danube-1.8b-chat", "name": "h2o-danube-1.8b-chat", "developer": "H2O.ai", "developer_id": "h2oai", "family": "H2O Danube", "description": null, "release_date": "2024-01", "date_precision": "month", "open_weights": true, "license": "apache-2.0", "parameters_b": 1.83120128, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://arxiv.org/abs/2401.16818v1", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://arxiv.org/abs/2401.16818v1", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/h2oai/h2o-danube-1.8b-chat", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/h2oai/h2o-danube-1.8b-chat", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The January 30 developer paper explicitly makes the model and chat variant openly available; month precision avoids treating its submission timestamp as the exact public-access time." }, { "id": "h2oai/h2o-danube-1.8b-sft", "slug": "h2oai--h2o-danube-1.8b-sft", "name": "h2o-danube-1.8b-sft", "developer": "H2O.ai", "developer_id": "h2oai", "family": "H2O Danube", "description": null, "release_date": "2024-01", "date_precision": "month", "open_weights": true, "license": "apache-2.0", "parameters_b": 1.83120128, "active_parameters_b": null, "parameters_source": "checkpoint_metadata", "architecture": null, "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://arxiv.org/abs/2401.16818v1", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://arxiv.org/abs/2401.16818v1", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/h2oai/h2o-danube-1.8b-sft", "type": "model_card" }, { "label": "Published checkpoint parameter metadata", "url": "https://huggingface.co/api/models/h2oai/h2o-danube-1.8b-sft", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:07:54.209Z", "last_seen": "2026-10-29T10:07:54.209Z", "upstream_present": false, "availability": "public", "notes": "The January 30 developer paper explicitly makes the model and chat variant openly available; month precision avoids treating its submission timestamp as the exact public-access time." }, { "id": "google/gemini-1.0-pro-vision", "slug": "google--gemini-1.0-pro-vision", "name": "Gemini 1.0 Pro Vision", "developer": "Google", "developer_id": "google", "family": "gemini", "description": null, "release_date": "2023-12-13", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 32768, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://docs.cloud.google.com/vertex-ai/docs/core-release-notes", "source_type": "primary", "sources": [ { "label": "Publisher historical API release notes", "url": "https://docs.cloud.google.com/vertex-ai/docs/core-release-notes", "type": "documentation" }, { "label": "Developer technical report", "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_1_report.pdf", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:26:10.259Z", "last_seen": "2026-10-29T10:26:10.259Z", "upstream_present": false, "availability": "public", "notes": "The December 13 Vertex AI note names Pro Vision separately from Pro. Multimodal API preview; later general availability is not a separate model release." }, { "id": "upstage/solar-10.7b-instruct-v1.0", "slug": "upstage--solar-10.7b-instruct-v1.0", "name": "SOLAR 10.7B Instruct v1.0", "developer": "Upstage", "developer_id": "upstage", "family": "solar", "description": "Instruction-tuned SOLAR text model built with depth up-scaling and optimized for single-turn conversation.", "release_date": "2023-12-13", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-4.0", "parameters_b": 10.731524096, "active_parameters_b": 10.7, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://docs.api.nvidia.com/nim/reference/upstage-solar-10_7b-instruct", "source_type": "secondary", "sources": [ { "label": "Exact Upstage Instruct model card", "url": "https://huggingface.co/upstage/SOLAR-10.7B-Instruct-v1.0/blob/090d405e6fe7b4c7b0b999a67f5c6dee0514a7ad/README.md", "type": "primary" }, { "label": "Pinned model configuration", "url": "https://huggingface.co/upstage/SOLAR-10.7B-Instruct-v1.0/blob/090d405e6fe7b4c7b0b999a67f5c6dee0514a7ad/config.json", "type": "primary" }, { "label": "Pinned original tensor index", "url": "https://huggingface.co/upstage/SOLAR-10.7B-Instruct-v1.0/blob/090d405e6fe7b4c7b0b999a67f5c6dee0514a7ad/model.safetensors.index.json", "type": "primary" }, { "label": "SOLAR technical report", "url": "https://arxiv.org/html/2312.15166v3", "type": "primary" }, { "label": "NVIDIA documented original release date", "url": "https://docs.api.nvidia.com/nim/reference/upstage-solar-10_7b-instruct", "type": "secondary" }, { "label": "Upstage developer license clarification", "url": "https://huggingface.co/upstage/SOLAR-10.7B-Instruct-v1.0/discussions/16", "type": "primary" }, { "label": "Watt Counts exact-checkpoint nominal active-count convention", "url": "https://arxiv.org/html/2604.09048v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/upstage/solar-10.7b-instruct-v1.0.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/upstage/SOLAR-10.7B-Instruct-v1.0", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Exact original F16 checkpoint contains 10,731,524,096 learned parameters across 435 tensors; 10.7B is the nominal name. Watt Counts Table 11 reports this exact checkpoint at 10.7B active under the authors’ nominal dense-model benchmark convention, not an Upstage per-token disclosure. CC-BY-NC-4.0 applies to Instruct; the base model uses Apache-2.0. December 13, 2023 is the explicit model-release date reported by NVIDIA and corroborated by the mirrored creator announcement; December 12 is only an upload timestamp, and direct access to the original announcement was unavailable. Dedicated reasoning/tool-calling support is unverified.", "field_status": { "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T18:13:04.841177+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/514.json" }, { "id": "microsoft/phi-2", "slug": "microsoft--phi-2", "name": "Phi-2", "developer": "Microsoft", "developer_id": "microsoft", "family": "phi", "description": null, "release_date": "2023-12-12", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 2.7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.microsoft.com/en-us/research/blog/phi-2-the-surprising-power-of-small-language-models/", "source_type": "primary", "sources": [ { "label": "Dated availability announcement", "url": "https://www.microsoft.com/en-us/research/blog/phi-2-the-surprising-power-of-small-language-models/", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/microsoft/phi-2/blob/810d367871c1d460086d9f82db8696f2e0a0fcd0/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/microsoft/phi-2/blob/810d367871c1d460086d9f82db8696f2e0a0fcd0/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "Previously announced at Ignite; this date is the publisher’s public availability announcement." }, { "id": "mistral/mixtral-8x7b-instruct-v0.1", "slug": "mistral--mixtral-8x7b-instruct-v0.1", "name": "Mixtral 8x7B Instruct v0.1", "developer": "Mistral", "developer_id": "mistral", "family": "mixtral", "description": "Open sparse mixture-of-experts Mistral instruction model", "release_date": "2023-12-11", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 46.7, "active_parameters_b": 12.9, "parameters_source": "primary_source", "architecture": "moe", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": false, "source_url": "https://mistral.ai/news/mixtral-of-experts/", "source_type": "primary", "sources": [ { "label": "Official model parameters", "url": "https://mistral.ai/news/mixtral-of-experts/", "type": "weights" }, { "label": "Exact Instruct checkpoint card", "url": "https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1/raw/eba92302a2861cdc0098cc54bc9f17cb2c47eb61/README.md", "type": "weights" }, { "label": "Exact Instruct configuration", "url": "https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1/raw/eba92302a2861cdc0098cc54bc9f17cb2c47eb61/config.json", "type": "weights" }, { "label": "Official Mixtral technical report", "url": "https://arxiv.org/html/2401.04088v1", "type": "primary" }, { "label": "Official Instruct weights specifications", "url": "https://docs.mistral.ai/models/mixtral-8x7b-0-1", "type": "primary" }, { "label": "Pinned official model specification", "url": "https://raw.githubusercontent.com/mistralai/platform-docs-public/cd2192ef7d0e96388a4261dbb6d417869e4bb965/src/schema/models/models/mixtral-8x7b-0-1.ts", "type": "primary" }, { "label": "Official V1 tool-protocol implementation", "url": "https://raw.githubusercontent.com/mistralai/mistral-common/d4364d04c684051a7a246e49bc699b9b1ed46420/src/mistral_common/tokens/tokenizers/instruct.py", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/mistral/mixtral-8x7b-instruct-v0.1.toml", "type": "catalog" }, { "label": "Hugging Face", "url": "https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Developer-disclosed nominal counts are 46.7B total and 12.9B active per token; the exact Instruct weights specification rounds these to 47B and 13B. Native tool calling is unsupported by the official V1 tokenizer; external wrappers are separate. Dedicated reasoning classification remains unverified.", "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-26T23:37:46.182030Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/348.json" }, { "id": "mistral/mixtral-8x7b-v0.1", "slug": "mistral--mixtral-8x7b-v0.1", "name": "Mixtral 8x7B v0.1", "developer": "Mistral AI", "developer_id": "mistral", "family": "mixtral", "description": null, "release_date": "2023-12-08", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 46.7, "active_parameters_b": 12.9, "parameters_source": "primary_source", "architecture": "moe", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://x.com/MistralAI/status/1733150512395038967", "source_type": "primary", "sources": [ { "label": "Original publisher torrent release", "url": "https://x.com/MistralAI/status/1733150512395038967", "type": "announcement" }, { "label": "Developer technical release and parameter counts", "url": "https://mistral.ai/news/mixtral-of-experts/", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/mistralai/Mixtral-8x7B-v0.1/blob/fc7ac94680e38d7348cfa806e51218e6273104b0/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/mistralai/Mixtral-8x7B-v0.1/blob/fc7ac94680e38d7348cfa806e51218e6273104b0/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "Base checkpoint released by the publisher through a torrent on December 8; technical post followed December 11. Publisher reports 46.7B total and 12.9B active parameters." }, { "id": "stabilityai/stablelm-zephyr-3b", "slug": "stabilityai--stablelm-zephyr-3b", "name": "StableLM Zephyr 3B", "developer": "Stability AI", "developer_id": "stabilityai", "family": "stablelm", "description": null, "release_date": "2023-12-07", "date_precision": "day", "open_weights": true, "license": "Stability AI Community License", "parameters_b": 3, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://stability.ai/news-updates/stablelm-zephyr-3b-stability-llm", "source_type": "primary", "sources": [ { "label": "Dated developer release", "url": "https://stability.ai/news-updates/stablelm-zephyr-3b-stability-llm", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/stabilityai/stablelm-zephyr-3b/blob/fe1fd5e36ccb8cb2cb1dc225ab6d2962692b9837/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/stabilityai/stablelm-zephyr-3b/blob/fe1fd5e36ccb8cb2cb1dc225ab6d2962692b9837/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "Publisher fine-tune of its own StableLM-3B-4E1T base; separate from third-party Zephyr models based on Mistral." }, { "id": "google/chat-bison-002", "slug": "google--chat-bison-002", "name": "PaLM 2 chat-bison 002", "developer": "Google", "developer_id": "google", "family": "palm-2", "description": null, "release_date": "2023-12-06", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://docs.cloud.google.com/vertex-ai/docs/core-release-notes", "source_type": "primary", "sources": [ { "label": "Publisher historical API release notes", "url": "https://docs.cloud.google.com/vertex-ai/docs/core-release-notes", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:26:10.259Z", "last_seen": "2026-10-29T10:26:10.259Z", "upstream_present": false, "availability": "public", "notes": "The December 6 release note names the exact @002 checkpoint and distinguishes it from @001. Context is not copied from another version." }, { "id": "google/code-bison-002", "slug": "google--code-bison-002", "name": "Codey code-bison 002", "developer": "Google", "developer_id": "google", "family": "codey", "description": null, "release_date": "2023-12-06", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://docs.cloud.google.com/vertex-ai/docs/core-release-notes", "source_type": "primary", "sources": [ { "label": "Publisher historical API release notes", "url": "https://docs.cloud.google.com/vertex-ai/docs/core-release-notes", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:26:10.259Z", "last_seen": "2026-10-29T10:26:10.259Z", "upstream_present": false, "availability": "public", "notes": "The December 6 release note names the exact @002 checkpoint and distinguishes it from @001. Context is not copied from another version." }, { "id": "google/code-gecko-002", "slug": "google--code-gecko-002", "name": "Codey code-gecko 002", "developer": "Google", "developer_id": "google", "family": "codey", "description": null, "release_date": "2023-12-06", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://docs.cloud.google.com/vertex-ai/docs/core-release-notes", "source_type": "primary", "sources": [ { "label": "Publisher historical API release notes", "url": "https://docs.cloud.google.com/vertex-ai/docs/core-release-notes", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:26:10.259Z", "last_seen": "2026-10-29T10:26:10.259Z", "upstream_present": false, "availability": "public", "notes": "The December 6 release note names the exact @002 checkpoint and distinguishes it from @001. Context is not copied from another version." }, { "id": "google/codechat-bison-002", "slug": "google--codechat-bison-002", "name": "Codey codechat-bison 002", "developer": "Google", "developer_id": "google", "family": "codey", "description": null, "release_date": "2023-12-06", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://docs.cloud.google.com/vertex-ai/docs/core-release-notes", "source_type": "primary", "sources": [ { "label": "Publisher historical API release notes", "url": "https://docs.cloud.google.com/vertex-ai/docs/core-release-notes", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:26:10.259Z", "last_seen": "2026-10-29T10:26:10.259Z", "upstream_present": false, "availability": "public", "notes": "The December 6 release note names the exact @002 checkpoint and distinguishes it from @001. Context is not copied from another version." }, { "id": "google/gemini-1.0-pro", "slug": "google--gemini-1.0-pro", "name": "Gemini 1.0 Pro", "developer": "Google", "developer_id": "google", "family": "gemini", "description": null, "release_date": "2023-12-06", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://blog.google/technology/ai/google-gemini-ai/", "source_type": "primary", "sources": [ { "label": "Dated launch and product availability", "url": "https://blog.google/technology/ai/google-gemini-ai/", "type": "announcement" }, { "label": "Publisher historical API release notes", "url": "https://docs.cloud.google.com/vertex-ai/docs/core-release-notes", "type": "documentation" }, { "label": "Developer technical report", "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_1_report.pdf", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:26:10.259Z", "last_seen": "2026-10-29T10:26:10.259Z", "upstream_present": false, "availability": "public", "notes": "First externally available in Bard on December 6; the public API preview followed December 13." }, { "id": "google/text-bison-002", "slug": "google--text-bison-002", "name": "PaLM 2 text-bison 002", "developer": "Google", "developer_id": "google", "family": "palm-2", "description": null, "release_date": "2023-12-06", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://docs.cloud.google.com/vertex-ai/docs/core-release-notes", "source_type": "primary", "sources": [ { "label": "Publisher historical API release notes", "url": "https://docs.cloud.google.com/vertex-ai/docs/core-release-notes", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:26:10.259Z", "last_seen": "2026-10-29T10:26:10.259Z", "upstream_present": false, "availability": "public", "notes": "The December 6 release note names the exact @002 checkpoint and distinguishes it from @001. Context is not copied from another version." }, { "id": "alibaba/qwen-1.8b", "slug": "alibaba--qwen-1.8b", "name": "Qwen-1.8B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": null, "release_date": "2023-11-30", "date_precision": "day", "open_weights": true, "license": "Tongyi Qianwen Research License", "parameters_b": 1.8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/QwenLM/Qwen/blob/2df8e8ac450fa185c421a08b0090ef81826caa6e/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/QwenLM/Qwen/blob/2df8e8ac450fa185c421a08b0090ef81826caa6e/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Qwen/Qwen-1_8B/blob/fa6e214ccbbc6a55235c26ef406355b6bfdf5eed/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/Qwen/Qwen-1_8B/blob/fa6e214ccbbc6a55235c26ef406355b6bfdf5eed/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Nominal parameter count. " }, { "id": "alibaba/qwen-1.8b-chat", "slug": "alibaba--qwen-1.8b-chat", "name": "Qwen-1.8B-Chat", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": null, "release_date": "2023-11-30", "date_precision": "day", "open_weights": true, "license": "Tongyi Qianwen Research License", "parameters_b": 1.8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/QwenLM/Qwen/blob/2df8e8ac450fa185c421a08b0090ef81826caa6e/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/QwenLM/Qwen/blob/2df8e8ac450fa185c421a08b0090ef81826caa6e/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Qwen/Qwen-1_8B-Chat", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Nominal parameter count. " }, { "id": "alibaba/qwen-72b", "slug": "alibaba--qwen-72b", "name": "Qwen-72B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": null, "release_date": "2023-11-30", "date_precision": "day", "open_weights": true, "license": "Tongyi Qianwen License", "parameters_b": 72, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/QwenLM/Qwen/blob/2df8e8ac450fa185c421a08b0090ef81826caa6e/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/QwenLM/Qwen/blob/2df8e8ac450fa185c421a08b0090ef81826caa6e/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Qwen/Qwen-72B/blob/b8e18ac61df64d35308695769ff46b976b6a00f4/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/Qwen/Qwen-72B/blob/b8e18ac61df64d35308695769ff46b976b6a00f4/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Nominal parameter count. " }, { "id": "alibaba/qwen-72b-chat", "slug": "alibaba--qwen-72b-chat", "name": "Qwen-72B-Chat", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": null, "release_date": "2023-11-30", "date_precision": "day", "open_weights": true, "license": "Tongyi Qianwen License", "parameters_b": 72, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/QwenLM/Qwen/blob/2df8e8ac450fa185c421a08b0090ef81826caa6e/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/QwenLM/Qwen/blob/2df8e8ac450fa185c421a08b0090ef81826caa6e/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Qwen/Qwen-72B-Chat/blob/307fccc1a45211b087ab294a8d291f7632259322/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/Qwen/Qwen-72B-Chat/blob/307fccc1a45211b087ab294a8d291f7632259322/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Nominal parameter count. " }, { "id": "alibaba/qwen-audio", "slug": "alibaba--qwen-audio", "name": "Qwen-Audio", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": null, "release_date": "2023-11-30", "date_precision": "day", "open_weights": true, "license": "Tongyi Qianwen License", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text", "audio" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/QwenLM/Qwen-Audio/blob/b50fb958438081d36e1a14e93dbbc2f329c7f10e/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/QwenLM/Qwen-Audio/blob/b50fb958438081d36e1a14e93dbbc2f329c7f10e/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Qwen/Qwen-Audio/blob/83eba550f821f38a12d5003582d9098fe66820c0/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/Qwen/Qwen-Audio/blob/83eba550f821f38a12d5003582d9098fe66820c0/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "The published 7B figure describes the language backbone; a primary total including the modality encoder has not yet been verified. November 15 was the paper announcement; the publisher dates checkpoint availability November 30." }, { "id": "alibaba/qwen-audio-chat", "slug": "alibaba--qwen-audio-chat", "name": "Qwen-Audio-Chat", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": null, "release_date": "2023-11-30", "date_precision": "day", "open_weights": true, "license": "Tongyi Qianwen License", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text", "audio" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/QwenLM/Qwen-Audio/blob/b50fb958438081d36e1a14e93dbbc2f329c7f10e/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/QwenLM/Qwen-Audio/blob/b50fb958438081d36e1a14e93dbbc2f329c7f10e/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Qwen/Qwen-Audio-Chat/blob/8b1c0dc720d34da5498f93535f416e3590bf3a71/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/Qwen/Qwen-Audio-Chat/blob/8b1c0dc720d34da5498f93535f416e3590bf3a71/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "The published 7B figure describes the language backbone; a primary total including the modality encoder has not yet been verified. November 15 was the paper announcement; the publisher dates checkpoint availability November 30." }, { "id": "google/text-unicorn-001", "slug": "google--text-unicorn-001", "name": "PaLM 2 Text Unicorn", "developer": "Google", "developer_id": "google", "family": "palm-2", "description": null, "release_date": "2023-11-30", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://docs.cloud.google.com/vertex-ai/docs/core-release-notes", "source_type": "primary", "sources": [ { "label": "Publisher historical API release notes", "url": "https://docs.cloud.google.com/vertex-ai/docs/core-release-notes", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:26:10.259Z", "last_seen": "2026-10-29T10:26:10.259Z", "upstream_present": false, "availability": "public", "notes": "The publisher dates public API availability November 30; its family size was announced May 10.", "announcement_date": "2023-05-10" }, { "id": "alibaba/qwen-vl-plus", "slug": "alibaba--qwen-vl-plus", "name": "Qwen-VL Plus", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Proprietary Qwen vision-language model for high-resolution image understanding, text recognition, and visual question answering.", "release_date": "2023-11-28", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 131072, "modalities": { "input": [ "text", "image", "video" ], "output": [ "text" ] }, "reasoning": null, "tool_call": false, "source_url": "https://github.com/QwenLM/Qwen-VL/blob/eeba51a04932c16a2c8cea998e6742f4ddc8303d/README.md", "source_type": "primary", "sources": [ { "label": "Qwen official repository and dated news", "url": "https://github.com/QwenLM/Qwen-VL", "type": "primary" }, { "label": "Qwen-VL overview", "url": "https://qwenlm.github.io/blog/qwen-vl/", "type": "primary" }, { "label": "Qwen proprietary model classification", "url": "https://qwen.readthedocs.io/en/stable/getting_started/concepts.html", "type": "primary" }, { "label": "Alibaba Cloud snapshot update notice", "url": "https://www.alibabacloud.com/zh/notice/model_studio_update_of_qwenvlmax_and_qwenvlplus_587?_p_lc=1", "type": "primary" }, { "label": "DashScope tools support", "url": "https://help.aliyun.com/en/model-studio/qwen-api-via-dashscope", "type": "primary" }, { "label": "Official specification (last indexed August 2026)", "url": "https://www.alibabacloud.com/help/zh/model-studio/qwen-vl-plus", "type": "primary" }, { "label": "Qwen-VL-Plus official indexed model card", "url": "https://docs.bailian.console.aliyun.com/zh/model-studio/qwen-vl-plus", "type": "primary" }, { "label": "Plus snapshot and Qwen2.5-VL family relationship", "url": "https://help.aliyun.com/en/model-studio/application-release-notes", "type": "primary" }, { "label": "Qwen2-VL launch and dated Max API mapping", "url": "https://qwenlm.github.io/blog/qwen2-vl/", "type": "primary" }, { "label": "Qwen2.5-VL launch sizes and architecture changes", "url": "https://qwenlm.github.io/blog/qwen2.5-vl/", "type": "primary" }, { "label": "Qwen-VL original November 28, 2023 public-availability news (December 12 README revision)", "url": "https://github.com/QwenLM/Qwen-VL/blob/eeba51a04932c16a2c8cea998e6742f4ddc8303d/README.md", "type": "primary" }, { "label": "Original Qwen-VL README commit metadata", "url": "https://api.github.com/repos/QwenLM/Qwen-VL/commits/eeba51a04932c16a2c8cea998e6742f4ddc8303d", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen-vl-plus.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Qwen-VL-Plus was publicly available on November 28, 2023 according to the original dated repository news, which linked its web interface. This is the original model release, distinct from later alias revisions. The last indexed official card maps the alias to 2025-08-15 and reports 131072 context with video input. Parameter counts and architecture remain undisclosed: the Qwen2.5-VL family relationship identifies no sized base model. Dedicated thinking mode remains unverified; endpoint availability was not tested.", "field_status": { "license": "not_applicable", "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed", "reasoning": "unverified" }, "audited_at": "2026-10-27T19:27:26.318060Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/019.json" }, { "id": "01-ai/yi-34b-chat", "slug": "01-ai--yi-34b-chat", "name": "Yi-34B-Chat", "developer": "01.AI", "developer_id": "01-ai", "family": "yi", "description": null, "release_date": "2023-11-23", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 34, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/01-ai/Yi/blob/f73194cc16e82df438bc01240b8e4b622209f3eb/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/01-ai/Yi/blob/f73194cc16e82df438bc01240b8e4b622209f3eb/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/01-ai/Yi-34B-Chat/blob/cf02cb50f2a03dead2fe205766a1c5598a90bf80/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/01-ai/Yi-34B-Chat/blob/cf02cb50f2a03dead2fe205766a1c5598a90bf80/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Publisher nominal parameter count. Current license is Apache-2.0; original November release used the Yi community license. " }, { "id": "01-ai/yi-6b-chat", "slug": "01-ai--yi-6b-chat", "name": "Yi-6B-Chat", "developer": "01.AI", "developer_id": "01-ai", "family": "yi", "description": null, "release_date": "2023-11-23", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 6, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/01-ai/Yi/blob/f73194cc16e82df438bc01240b8e4b622209f3eb/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/01-ai/Yi/blob/f73194cc16e82df438bc01240b8e4b622209f3eb/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/01-ai/Yi-6B-Chat/blob/2dbf63b0cb7bc493c0243502c6e6111a36e3a093/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/01-ai/Yi-6B-Chat/blob/2dbf63b0cb7bc493c0243502c6e6111a36e3a093/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Publisher nominal parameter count. Current license is Apache-2.0; original November release used the Yi community license. " }, { "id": "anthropic/claude-2.1", "slug": "anthropic--claude-2.1", "name": "Claude 2.1", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude", "description": "Anthropic Claude 2 revision with a 200K-token context window, system prompts and beta tool use.", "release_date": "2023-11-21", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 200000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://www.anthropic.com/news/claude-2-1", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://www.anthropic.com/news/claude-2-1", "type": "announcement" }, { "label": "Claude 2 model card and 2.1 appendix", "url": "https://www-cdn.anthropic.com/files/4zrzovbb/website/5c49cc247484cecf107c699baf29250302e5da70.pdf", "type": "model_card" }, { "label": "Claude 2.1 Messages API context", "url": "https://www.anthropic.com/news/expanded-legal-protections-api-improvements", "type": "announcement" }, { "label": "Anthropic historical commercial terms", "url": "https://www-cdn.anthropic.com/6b68a6508f0210c5fe08f0199caa05c4ee6fb4dc/Anthropic-on-Bedrock-Commercial-Terms-of-Service_Dec_2023.pdf", "type": "documentation" }, { "label": "Anthropic model deprecations", "url": "https://docs.anthropic.com/en/docs/about-claude/model-deprecations", "type": "documentation" }, { "label": "Claude 3 family card with later addenda", "url": "https://www-cdn.anthropic.com/c6a80a657af445f40e31afac050f3bf76d3b1404.pdf", "type": "primary" }, { "label": "Responsible Scaling Policy, Version 1.0", "url": "https://www-cdn.anthropic.com/1adf000c8f675958c2ee23805d91aaade1cd4613/responsible-scaling-policy.pdf", "type": "primary" }, { "label": "Claude 2.1 appendix, complete official copy", "url": "https://www-cdn.anthropic.com/5c49cc247484cecf107c699baf29250302e5da70/ModelCardClaudev2_with_appendix_v1.pdf", "type": "model_card" }, { "label": "Claude 3.7 Sonnet system card", "url": "https://www-cdn.anthropic.com/9ff93dfa8f445c932415d335c88852ef47f1201e.pdf", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "No nominal total or active count is disclosed for Claude 2.1 or its explicitly identified Claude 2.0 base in the reviewed primary sources. Transformer is documented, but dense/MoE routing is not established. Tool use was beta; Messages API function calling was still forthcoming in December 2023. Weight access is restricted under Claude 2 containment policy and proprietary terms; the 2.1 appendix identifies the same base model. Anthropic retired claude-2.1 on July 21, 2025; historical capabilities do not establish a currently served endpoint.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "reasoning": "unverified" }, "audited_at": "2026-10-27T10:13:48.373Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/088.json" }, { "id": "openai/gpt-3.5-turbo-1106", "slug": "openai--gpt-3.5-turbo-1106", "name": "GPT-3.5 Turbo 1106", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-3.5", "description": null, "release_date": "2023-11-06", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 16385, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://openai.com/index/new-models-and-developer-products-announced-at-devday/", "source_type": "primary", "sources": [ { "label": "Dated exact-checkpoint release", "url": "https://openai.com/index/new-models-and-developer-products-announced-at-devday/", "type": "announcement" }, { "label": "Publisher checkpoint specification", "url": "https://platform.openai.com/docs/models/gpt-3.5-turbo", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "" }, { "id": "openai/gpt-4-turbo", "slug": "openai--gpt-4-turbo", "name": "GPT-4 Turbo", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-4", "description": "GPT-4 model with a 128,000-token context window; its April 2024 stable snapshot accepts text and images, generates text, and supports function calling.", "release_date": "2023-11-06", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 128000, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://openai.com/index/new-models-and-developer-products-announced-at-devday/", "source_type": "primary", "sources": [ { "label": "New models and developer products announced at DevDay", "url": "https://openai.com/index/new-models-and-developer-products-announced-at-devday/", "type": "primary" }, { "label": "GPT-4 Turbo model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-4-turbo", "type": "primary" }, { "label": "OpenAI API changelog, April 9, 2024", "url": "https://developers.openai.com/api/docs/changelog", "type": "primary" }, { "label": "Deprecations: legacy GPT model snapshots", "url": "https://developers.openai.com/api/docs/deprecations#2026-04-22-legacy-gpt-model-snapshots", "type": "primary" }, { "label": "Introducing gpt-oss", "url": "https://openai.com/index/introducing-gpt-oss/", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "Scaling and evaluating sparse autoencoders", "url": "https://arxiv.org/html/2406.04093v1", "type": "primary" }, { "label": "GPT-4o System Card", "url": "https://openai.com/index/gpt-4o-system-card/", "type": "primary" }, { "label": "DeepSeek-Coder-V2: Breaking the Barrier of Closed-Source Models in Code Intelligence", "url": "https://arxiv.org/html/2406.11931v1", "type": "primary" }, { "label": "Incompressible Knowledge Probes: Estimating Black-Box LLM Parameter Counts via Factual Capacity (v2)", "url": "https://arxiv.org/html/2604.24827v2", "type": "primary" }, { "label": "EcoLogits proprietary-model estimation methodology", "url": "https://ecologits.ai/0.10/methodology/proprietary_models/", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-4-turbo.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The release date is the November 6, 2023 Turbo preview; listed capabilities describe the stable gpt-4-turbo-2024-04-09 snapshot with vision. OpenAI documents its October 23, 2026 API shutdown. Total and active parameters remain undisclosed: exact-model research comparison tables supply neither count. Pricing-based and effective-knowledge estimates are not model specifications. Architecture and dedicated reasoning-mode support remain unverified.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "unverified", "reasoning": "unverified" }, "audited_at": "2026-10-27T16:21:25.756876Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/404.json" }, { "id": "01-ai/yi-34b-200k", "slug": "01-ai--yi-34b-200k", "name": "Yi-34B-200K", "developer": "01.AI", "developer_id": "01-ai", "family": "yi", "description": null, "release_date": "2023-11-05", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 34, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 200000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/01-ai/Yi/blob/f73194cc16e82df438bc01240b8e4b622209f3eb/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/01-ai/Yi/blob/f73194cc16e82df438bc01240b8e4b622209f3eb/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/01-ai/Yi-34B-200K/blob/e0ae1afac6b69f604556efd441ab4befafb2a835/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/01-ai/Yi-34B-200K/blob/e0ae1afac6b69f604556efd441ab4befafb2a835/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Publisher nominal parameter count. Current license is Apache-2.0; original November release used the Yi community license. The publisher retains the original November 5 weights at revision 069cd341d60f4ce4b07ec394e82b79e94f656cf; main was updated in 2024." }, { "id": "01-ai/yi-6b-200k", "slug": "01-ai--yi-6b-200k", "name": "Yi-6B-200K", "developer": "01.AI", "developer_id": "01-ai", "family": "yi", "description": null, "release_date": "2023-11-05", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 6, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 200000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/01-ai/Yi/blob/f73194cc16e82df438bc01240b8e4b622209f3eb/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/01-ai/Yi/blob/f73194cc16e82df438bc01240b8e4b622209f3eb/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/01-ai/Yi-6B-200K/blob/46b2762d8a78344292598f3ced6af746cec84bec/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/01-ai/Yi-6B-200K/blob/46b2762d8a78344292598f3ced6af746cec84bec/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Publisher nominal parameter count. Current license is Apache-2.0; original November release used the Yi community license. " }, { "id": "01-ai/yi-34b", "slug": "01-ai--yi-34b", "name": "Yi-34B", "developer": "01.AI", "developer_id": "01-ai", "family": "yi", "description": null, "release_date": "2023-11-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 34, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/01-ai/Yi/blob/f73194cc16e82df438bc01240b8e4b622209f3eb/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/01-ai/Yi/blob/f73194cc16e82df438bc01240b8e4b622209f3eb/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/01-ai/Yi-34B/blob/e1fa7c83283f2e1d5a12cb4adea37c3f829e28f8/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/01-ai/Yi-34B/blob/e1fa7c83283f2e1d5a12cb4adea37c3f829e28f8/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Publisher nominal parameter count. Current license is Apache-2.0; original November release used the Yi community license. " }, { "id": "01-ai/yi-6b", "slug": "01-ai--yi-6b", "name": "Yi-6B", "developer": "01.AI", "developer_id": "01-ai", "family": "yi", "description": null, "release_date": "2023-11-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 6, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/01-ai/Yi/blob/f73194cc16e82df438bc01240b8e4b622209f3eb/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/01-ai/Yi/blob/f73194cc16e82df438bc01240b8e4b622209f3eb/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/01-ai/Yi-6B/blob/80080be87ec5a0103f643195f2d9003b8068941b/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/01-ai/Yi-6B/blob/80080be87ec5a0103f643195f2d9003b8068941b/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Publisher nominal parameter count. Current license is Apache-2.0; original November release used the Yi community license. " }, { "id": "eleutherai/pythia-14m", "slug": "eleutherai--pythia-14m", "name": "Pythia 14M", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2023-11-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.014067712, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "source_type": "primary", "sources": [ { "label": "Dated release of additional small models", "url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/EleutherAI/pythia-14m/blob/cf967c0a9a04383db6f7b1108d86b2962634b4ac/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/EleutherAI/pythia-14m/blob/cf967c0a9a04383db6f7b1108d86b2962634b4ac/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Total includes embedding and output matrices.", "availability": "public" }, { "id": "eleutherai/pythia-31m", "slug": "eleutherai--pythia-31m", "name": "Pythia 31M", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2023-11-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.03049472, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "source_type": "primary", "sources": [ { "label": "Dated release of additional small models", "url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/EleutherAI/pythia-31m/blob/e556ace21b489575e94e9d50b6dad2fcc7419679/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/EleutherAI/pythia-31m/blob/e556ace21b489575e94e9d50b6dad2fcc7419679/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Total includes embedding and output matrices.", "availability": "public" }, { "id": "deepseek/deepseek-coder-1.3b-base", "slug": "deepseek--deepseek-coder-1.3b-base", "name": "DeepSeek Coder 1.3B Base", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-coder", "description": null, "release_date": "2023-11", "date_precision": "month", "open_weights": true, "license": "DeepSeek Model License", "parameters_b": 1.3, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/deepseek-ai/DeepSeek-Coder/blob/1af85c9ad616a6b9c33d812f6db90409953c7bc6/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/deepseek-ai/DeepSeek-Coder/blob/1af85c9ad616a6b9c33d812f6db90409953c7bc6/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/deepseek-ai/deepseek-coder-1.3b-base/blob/c919139c3a9b4070729c8b2cca4847ab29ca8d94/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/deepseek-ai/deepseek-coder-1.3b-base/blob/c919139c3a9b4070729c8b2cca4847ab29ca8d94/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "The November 2023 publisher release documentation names these model sizes; exact public launch day remains unresolved. Publisher states these models were trained from scratch." }, { "id": "deepseek/deepseek-coder-1.3b-instruct", "slug": "deepseek--deepseek-coder-1.3b-instruct", "name": "DeepSeek Coder 1.3B Instruct", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-coder", "description": null, "release_date": "2023-11", "date_precision": "month", "open_weights": true, "license": "DeepSeek Model License", "parameters_b": 1.3, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/deepseek-ai/DeepSeek-Coder/blob/1af85c9ad616a6b9c33d812f6db90409953c7bc6/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/deepseek-ai/DeepSeek-Coder/blob/1af85c9ad616a6b9c33d812f6db90409953c7bc6/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/deepseek-ai/deepseek-coder-1.3b-instruct/blob/e063262dac8366fc1f28a4da0ff3c50ea66259ca/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/deepseek-ai/deepseek-coder-1.3b-instruct/blob/e063262dac8366fc1f28a4da0ff3c50ea66259ca/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "The November 2023 publisher release documentation names these model sizes; exact public launch day remains unresolved. Publisher states these models were trained from scratch." }, { "id": "deepseek/deepseek-coder-33b-base", "slug": "deepseek--deepseek-coder-33b-base", "name": "DeepSeek Coder 33B Base", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-coder", "description": null, "release_date": "2023-11", "date_precision": "month", "open_weights": true, "license": "DeepSeek Model License", "parameters_b": 33, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/deepseek-ai/DeepSeek-Coder/blob/1af85c9ad616a6b9c33d812f6db90409953c7bc6/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/deepseek-ai/DeepSeek-Coder/blob/1af85c9ad616a6b9c33d812f6db90409953c7bc6/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/deepseek-ai/deepseek-coder-33b-base/blob/45c85cadf3720ef3e85a492e24fd4b8c5d21d8ac/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/deepseek-ai/deepseek-coder-33b-base/blob/45c85cadf3720ef3e85a492e24fd4b8c5d21d8ac/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "The November 2023 publisher release documentation names these model sizes; exact public launch day remains unresolved. Publisher states these models were trained from scratch." }, { "id": "deepseek/deepseek-coder-33b-instruct", "slug": "deepseek--deepseek-coder-33b-instruct", "name": "DeepSeek Coder 33B Instruct", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-coder", "description": null, "release_date": "2023-11", "date_precision": "month", "open_weights": true, "license": "DeepSeek Model License", "parameters_b": 33, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/deepseek-ai/DeepSeek-Coder/blob/1af85c9ad616a6b9c33d812f6db90409953c7bc6/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/deepseek-ai/DeepSeek-Coder/blob/1af85c9ad616a6b9c33d812f6db90409953c7bc6/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/deepseek-ai/deepseek-coder-33b-instruct/blob/61dc97b922b13995e7f83b7c8397701dbf9cfd4c/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/deepseek-ai/deepseek-coder-33b-instruct/blob/61dc97b922b13995e7f83b7c8397701dbf9cfd4c/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "The November 2023 publisher release documentation names these model sizes; exact public launch day remains unresolved. Publisher states these models were trained from scratch." }, { "id": "deepseek/deepseek-coder-5.7bmqa-base", "slug": "deepseek--deepseek-coder-5.7bmqa-base", "name": "DeepSeek Coder 5.7B Base", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-coder", "description": null, "release_date": "2023-11", "date_precision": "month", "open_weights": true, "license": "DeepSeek Model License", "parameters_b": 5.7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/deepseek-ai/DeepSeek-Coder/blob/1af85c9ad616a6b9c33d812f6db90409953c7bc6/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/deepseek-ai/DeepSeek-Coder/blob/1af85c9ad616a6b9c33d812f6db90409953c7bc6/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/deepseek-ai/deepseek-coder-5.7bmqa-base/blob/3746ee18367858733aa17c944755df6df2bb1d14/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/deepseek-ai/deepseek-coder-5.7bmqa-base/blob/3746ee18367858733aa17c944755df6df2bb1d14/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "The November 2023 publisher release documentation names these model sizes; exact public launch day remains unresolved. Publisher states these models were trained from scratch." }, { "id": "deepseek/deepseek-coder-6.7b-base", "slug": "deepseek--deepseek-coder-6.7b-base", "name": "DeepSeek Coder 6.7B Base", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-coder", "description": null, "release_date": "2023-11", "date_precision": "month", "open_weights": true, "license": "DeepSeek Model License", "parameters_b": 6.7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/deepseek-ai/DeepSeek-Coder/blob/1af85c9ad616a6b9c33d812f6db90409953c7bc6/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/deepseek-ai/DeepSeek-Coder/blob/1af85c9ad616a6b9c33d812f6db90409953c7bc6/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/deepseek-ai/deepseek-coder-6.7b-base/blob/ce2207a8bfef3ee92bd7dd4cc31c52cfa0046912/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/deepseek-ai/deepseek-coder-6.7b-base/blob/ce2207a8bfef3ee92bd7dd4cc31c52cfa0046912/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "The November 2023 publisher release documentation names these model sizes; exact public launch day remains unresolved. Publisher states these models were trained from scratch." }, { "id": "deepseek/deepseek-coder-6.7b-instruct", "slug": "deepseek--deepseek-coder-6.7b-instruct", "name": "DeepSeek Coder 6.7B Instruct", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-coder", "description": null, "release_date": "2023-11", "date_precision": "month", "open_weights": true, "license": "DeepSeek Model License", "parameters_b": 6.7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/deepseek-ai/DeepSeek-Coder/blob/1af85c9ad616a6b9c33d812f6db90409953c7bc6/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/deepseek-ai/DeepSeek-Coder/blob/1af85c9ad616a6b9c33d812f6db90409953c7bc6/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/deepseek-ai/deepseek-coder-6.7b-instruct/blob/e5d64addd26a6a1db0f9b863abf6ee3141936807/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/deepseek-ai/deepseek-coder-6.7b-instruct/blob/e5d64addd26a6a1db0f9b863abf6ee3141936807/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "The November 2023 publisher release documentation names these model sizes; exact public launch day remains unresolved. Publisher states these models were trained from scratch." }, { "id": "deepseek/deepseek-llm-67b-base", "slug": "deepseek--deepseek-llm-67b-base", "name": "DeepSeek LLM 67B Base", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-llm", "description": null, "release_date": "2023-11", "date_precision": "month", "open_weights": true, "license": "DeepSeek Model License", "parameters_b": 67, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/deepseek-ai/DeepSeek-LLM/blob/20d316b940944a77ef914cc9bcabe604fe51dc69/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/deepseek-ai/DeepSeek-LLM/blob/20d316b940944a77ef914cc9bcabe604fe51dc69/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/deepseek-ai/deepseek-llm-67b-base/blob/c3f813a1121c95488a20132d3a4da89f4a46452f/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/deepseek-ai/deepseek-llm-67b-base/blob/c3f813a1121c95488a20132d3a4da89f4a46452f/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "Publisher release documentation from November 2023 lists this exact checkpoint. Exact public launch day remains unresolved." }, { "id": "deepseek/deepseek-llm-67b-chat", "slug": "deepseek--deepseek-llm-67b-chat", "name": "DeepSeek LLM 67B Chat", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-llm", "description": null, "release_date": "2023-11", "date_precision": "month", "open_weights": true, "license": "DeepSeek Model License", "parameters_b": 67, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/deepseek-ai/DeepSeek-LLM/blob/20d316b940944a77ef914cc9bcabe604fe51dc69/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/deepseek-ai/DeepSeek-LLM/blob/20d316b940944a77ef914cc9bcabe604fe51dc69/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/deepseek-ai/deepseek-llm-67b-chat/blob/79648bef7658bb824e4630740f6e1484c1b0620b/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/deepseek-ai/deepseek-llm-67b-chat/blob/79648bef7658bb824e4630740f6e1484c1b0620b/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "Publisher release documentation from November 2023 lists this exact checkpoint. Exact public launch day remains unresolved." }, { "id": "deepseek/deepseek-llm-7b-base", "slug": "deepseek--deepseek-llm-7b-base", "name": "DeepSeek LLM 7B Base", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-llm", "description": null, "release_date": "2023-11", "date_precision": "month", "open_weights": true, "license": "DeepSeek Model License", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/deepseek-ai/DeepSeek-LLM/blob/20d316b940944a77ef914cc9bcabe604fe51dc69/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/deepseek-ai/DeepSeek-LLM/blob/20d316b940944a77ef914cc9bcabe604fe51dc69/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/deepseek-ai/deepseek-llm-7b-base/blob/7683fea62db869066ddaff6a41d032262c490d4f/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/deepseek-ai/deepseek-llm-7b-base/blob/7683fea62db869066ddaff6a41d032262c490d4f/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "Publisher release documentation from November 2023 lists this exact checkpoint. Exact public launch day remains unresolved." }, { "id": "deepseek/deepseek-llm-7b-chat", "slug": "deepseek--deepseek-llm-7b-chat", "name": "DeepSeek LLM 7B Chat", "developer": "DeepSeek", "developer_id": "deepseek", "family": "deepseek-llm", "description": null, "release_date": "2023-11", "date_precision": "month", "open_weights": true, "license": "DeepSeek Model License", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/deepseek-ai/DeepSeek-LLM/blob/20d316b940944a77ef914cc9bcabe604fe51dc69/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/deepseek-ai/DeepSeek-LLM/blob/20d316b940944a77ef914cc9bcabe604fe51dc69/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/deepseek-ai/deepseek-llm-7b-chat/blob/afbda8b347ec881666061fa67447046fc5164ec8/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/deepseek-ai/deepseek-llm-7b-chat/blob/afbda8b347ec881666061fa67447046fc5164ec8/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "Publisher release documentation from November 2023 lists this exact checkpoint. Exact public launch day remains unresolved." }, { "id": "alibaba/qwen-max", "slug": "alibaba--qwen-max", "name": "Qwen Max", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Hosted Qwen text model for multilingual chat and coding, currently mapped to the Qwen2.5-Max snapshot.", "release_date": "2023-10-31", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "moe", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": true, "source_url": "https://www.alibabacloud.com/blog/600526", "source_type": "primary", "sources": [ { "label": "qwen-max Model Info", "url": "https://help.aliyun.com/en/model-studio/qwen-max", "type": "primary" }, { "label": "Qwen2.5-Max: Exploring the Intelligence of Large-scale MoE Model", "url": "https://qwenlm.github.io/blog/qwen2.5-max/", "type": "primary" }, { "label": "Alibaba Cloud Model Studio model pricing", "url": "https://www.alibabacloud.com/help/en/model-studio/model-pricing", "type": "primary" }, { "label": "Text generation models", "url": "https://www.alibabacloud.com/help/en/model-studio/text-generation-model/", "type": "primary" }, { "label": "Model lifecycle and updates", "url": "https://help.aliyun.com/en/model-studio/newly-released-models", "type": "primary" }, { "label": "Alibaba Cloud Launches First Cloud Region in Brazil to Accelerate Cloud and AI Transformation", "url": "https://www.alibabacloud.com/en/press-room/alibaba-cloud-launch-first-cloud-region-in-brazil?_p_lc=1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen-max.toml", "type": "catalog" }, { "label": "qwen-max Model Info (international documentation)", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen-max", "type": "primary" }, { "label": "qwen-max model information (Chinese documentation)", "url": "https://help.aliyun.com/zh/model-studio/qwen-max", "type": "primary" }, { "label": "Qwen Series Model Free Quota Validity Period Adjustment Notice", "url": "https://help.aliyun.com/en/model-studio/qwen-model-free-quota-adjustment", "type": "primary" }, { "label": "Qwen2.5 Technical Report, revision 2", "url": "https://arxiv.org/html/2412.15115v2", "type": "primary" }, { "label": "Qwen3: Think Deeper, Act Faster", "url": "https://qwenlm.github.io/blog/qwen3/", "type": "primary" }, { "label": "Qwen2.5-Max: Exploring large-scale MoE intelligence (Chinese announcement)", "url": "https://qwenlm.github.io/zh/blog/qwen2.5-max/", "type": "primary" }, { "label": "Alibaba Cloud launches Tongyi Qianwen 2.0 (October 31, 2023)", "url": "https://www.alibabacloud.com/blog/600526", "type": "primary" }, { "label": "Qwen original README identifying qwen-max as Tongyi Qianwen 2.0", "url": "https://github.com/QwenLM/Qwen/blob/b1d80a9385a73b89e3c3b002fb54b9c432da2925/README_CN.md", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Rolling hosted alias: MoE architecture refers to the documented qwen-max-2025-01-25 (Qwen2.5-Max) mapping. Alibaba describes a hundred-billion-scale model but publishes no nominal total or active parameter count for this snapshot; later Qwen3/Qwen3.8 sizes do not apply. Current context is unresolved: exact model cards state 32768, including the October 11, 2026 international card; the October 15 overview says 128k. No region-specific context difference or upgrade announcement was verified. Function calling is supported in Beijing but marked unsupported in Singapore. Original Max release: October 31, 2023. Alibaba announced public/API availability of Tongyi Qianwen 2.0 on that day; Qwen’s original README explicitly identifies qwen-max as that model. January 28, 2025 announces the later Qwen2.5-Max snapshot.", "field_status": { "license": "not_applicable", "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "context_window": "unverified" }, "audited_at": "2026-10-27T19:27:26.384816Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/013.json" }, { "id": "replit/replit-code-v1.5-3b", "slug": "replit--replit-code-v1.5-3b", "name": "Replit Code v1.5 3B", "developer": "Replit", "developer_id": "replit", "family": "replit-code", "description": null, "release_date": "2023-10-10", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 3, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://replit.com/blog/replit-code-v1_5", "source_type": "primary", "sources": [ { "label": "Dated exact-model release", "url": "https://replit.com/blog/replit-code-v1_5", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/replit/replit-code-v1_5-3b/blob/b85a5dd28ef6053ad095bf5113dd740a86a67597/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/replit/replit-code-v1_5-3b/blob/b85a5dd28ef6053ad095bf5113dd740a86a67597/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "October 9 announced free Replit AI; the developer dates the downloadable v1.5 checkpoint announcement October 10." }, { "id": "zhipuai/chatglm3-6b", "slug": "zhipuai--chatglm3-6b", "name": "ChatGLM3-6B", "developer": "Zhipu AI / Tsinghua University", "developer_id": "zhipuai", "family": "chatglm", "description": null, "release_date": "2023-10", "date_precision": "month", "open_weights": true, "license": "ChatGLM3 Model License", "parameters_b": 6, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://arxiv.org/html/2406.12793v1", "source_type": "primary", "sources": [ { "label": "Model-team technical history", "url": "https://arxiv.org/html/2406.12793v1", "type": "paper" }, { "label": "Publisher release chronology", "url": "https://github.com/THUDM/ChatGLM3/blob/732294f42e2625509000daa87a9c3748d0bbb957/README_en.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/THUDM/chatglm3-6b/blob/e9e0406d062cdb887444fe5bd546833920abd4ac/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/THUDM/chatglm3-6b/blob/e9e0406d062cdb887444fe5bd546833920abd4ac/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "The model-team timeline dates the third generation to October 2023; an exact launch day has not yet been verified." }, { "id": "zhipuai/chatglm3-6b-32k", "slug": "zhipuai--chatglm3-6b-32k", "name": "ChatGLM3-6B-32K", "developer": "Zhipu AI / Tsinghua University", "developer_id": "zhipuai", "family": "chatglm", "description": null, "release_date": "2023-10", "date_precision": "month", "open_weights": true, "license": "ChatGLM3 Model License", "parameters_b": 6, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://arxiv.org/html/2406.12793v1", "source_type": "primary", "sources": [ { "label": "Model-team technical history", "url": "https://arxiv.org/html/2406.12793v1", "type": "paper" }, { "label": "Publisher release chronology", "url": "https://github.com/THUDM/ChatGLM3/blob/732294f42e2625509000daa87a9c3748d0bbb957/README_en.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/THUDM/chatglm3-6b-32k/blob/5e12a231fe1640a5bac73363de0044be6e20b54a/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/THUDM/chatglm3-6b-32k/blob/5e12a231fe1640a5bac73363de0044be6e20b54a/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "The model-team timeline dates the third generation to October 2023; an exact launch day has not yet been verified." }, { "id": "zhipuai/chatglm3-6b-base", "slug": "zhipuai--chatglm3-6b-base", "name": "ChatGLM3-6B-Base", "developer": "Zhipu AI / Tsinghua University", "developer_id": "zhipuai", "family": "chatglm", "description": null, "release_date": "2023-10", "date_precision": "month", "open_weights": true, "license": "ChatGLM3 Model License", "parameters_b": 6, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://arxiv.org/html/2406.12793v1", "source_type": "primary", "sources": [ { "label": "Model-team technical history", "url": "https://arxiv.org/html/2406.12793v1", "type": "paper" }, { "label": "Publisher release chronology", "url": "https://github.com/THUDM/ChatGLM3/blob/732294f42e2625509000daa87a9c3748d0bbb957/README_en.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/THUDM/chatglm3-6b-base/blob/9680e216e62c3a69bc9eec480c35d59c325079bf/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/THUDM/chatglm3-6b-base/blob/9680e216e62c3a69bc9eec480c35d59c325079bf/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "The model-team timeline dates the third generation to October 2023; an exact launch day has not yet been verified." }, { "id": "stabilityai/stablelm-3b-4e1t", "slug": "stabilityai--stablelm-3b-4e1t", "name": "StableLM 3B 4E1T", "developer": "Stability AI", "developer_id": "stabilityai", "family": "stablelm", "description": null, "release_date": "2023-09-29", "date_precision": "day", "open_weights": true, "license": "CC-BY-SA-4.0", "parameters_b": 2.7954432, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/Stability-AI/StableLM/blob/main/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release changelog", "url": "https://github.com/Stability-AI/StableLM/blob/main/README.md", "type": "repository" }, { "label": "Subsequent availability announcement", "url": "https://stability.ai/news-updates/stable-lm-3b-sustainable-high-performance-language-models-smart-devices", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/stabilityai/stablelm-3b-4e1t/blob/fa4a6a92fca83c3b4223a3c9bf792887090ebfba/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/stabilityai/stablelm-3b-4e1t/blob/fa4a6a92fca83c3b4223a3c9bf792887090ebfba/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "The repository dates the checkpoint September 29; the corresponding blog post was published October 2." }, { "id": "mistral/mistral-7b-instruct-v0.1", "slug": "mistral--mistral-7b-instruct-v0.1", "name": "Mistral 7B Instruct v0.1", "developer": "Mistral AI", "developer_id": "mistral", "family": "mistral", "description": "Original Mistral 7B instruction-tuned checkpoint.", "release_date": "2023-09-27", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 7.241732096, "active_parameters_b": 7, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": false, "source_url": "https://mistral.ai/news/announcing-mistral-7b/", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://mistral.ai/news/announcing-mistral-7b/", "type": "announcement" }, { "label": "Official model documentation", "url": "https://docs.mistral.ai/models/mistral-7b-0-1", "type": "docs" }, { "label": "Mistral-7B-Instruct-v0.1 model card at pinned revision", "url": "https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1/blob/ec5deb64f2c6e6fa90c1abf74a91d5c93a9669ca/README.md", "type": "weights" }, { "label": "Exact v0.1 checkpoint configuration", "url": "https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1/blob/ec5deb64f2c6e6fa90c1abf74a91d5c93a9669ca/config.json", "type": "weights" }, { "label": "Complete v0.1 Safetensors index", "url": "https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1/blob/ec5deb64f2c6e6fa90c1abf74a91d5c93a9669ca/model.safetensors.index.json", "type": "weights" }, { "label": "v0.1 Safetensors shard 1 header", "url": "https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1/resolve/ec5deb64f2c6e6fa90c1abf74a91d5c93a9669ca/model-00001-of-00002.safetensors", "type": "weights" }, { "label": "v0.1 Safetensors shard 2 header", "url": "https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1/resolve/ec5deb64f2c6e6fa90c1abf74a91d5c93a9669ca/model-00002-of-00002.safetensors", "type": "weights" }, { "label": "Mistral 7B original technical report, arXiv v1", "url": "https://arxiv.org/html/2310.06825v1", "type": "primary" }, { "label": "Mistral official InstructTokenizerV1 implementation", "url": "https://github.com/mistralai/mistral-common/blob/d4364d04c684051a7a246e49bc699b9b1ed46420/src/mistral_common/tokens/tokenizers/instruct.py#L297", "type": "primary" }, { "label": "Mistral official tokenizer version factory", "url": "https://github.com/mistralai/mistral-common/blob/d4364d04c684051a7a246e49bc699b9b1ed46420/src/mistral_common/tokens/tokenizers/mistral.py#L193", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "Total is the complete 7,241,732,096-parameter BF16 checkpoint count; active 7B is separately published and nominal. The documented context is 8,192 tokens, distinct from the 4,096-token sliding attention window and 32,768 configuration position ceiling. Native tool calls are unsupported by the official V1 tokenizer. Dedicated reasoning mode remains unverified; structured thinking-chunk rejection alone cannot establish its absence.", "field_status": { "reasoning": "unverified" }, "audited_at": "2026-10-26T23:08:11.596900Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/329.json" }, { "id": "mistral/mistral-7b-v0.1", "slug": "mistral--mistral-7b-v0.1", "name": "Mistral 7B v0.1", "developer": "Mistral AI", "developer_id": "mistral", "family": "mistral", "description": "Original Mistral 7B pretrained checkpoint.", "release_date": "2023-09-27", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 7.241732096, "active_parameters_b": 7, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://mistral.ai/news/announcing-mistral-7b/", "source_type": "primary", "sources": [ { "label": "Official model documentation", "url": "https://docs.mistral.ai/models/mistral-7b-0-1", "type": "docs" }, { "label": "Exact pretrained checkpoint model card", "url": "https://huggingface.co/mistralai/Mistral-7B-v0.1/blob/27d67f1b5f57dc0953326b2601d68371d40ea8da/README.md", "type": "weights" }, { "label": "Exact checkpoint configuration", "url": "https://huggingface.co/mistralai/Mistral-7B-v0.1/blob/27d67f1b5f57dc0953326b2601d68371d40ea8da/config.json", "type": "weights" }, { "label": "Complete checkpoint tensor index", "url": "https://huggingface.co/mistralai/Mistral-7B-v0.1/blob/27d67f1b5f57dc0953326b2601d68371d40ea8da/model.safetensors.index.json", "type": "weights" }, { "label": "Checkpoint shard 1 tensor header", "url": "https://huggingface.co/mistralai/Mistral-7B-v0.1/resolve/27d67f1b5f57dc0953326b2601d68371d40ea8da/model-00001-of-00002.safetensors", "type": "weights" }, { "label": "Checkpoint shard 2 tensor header", "url": "https://huggingface.co/mistralai/Mistral-7B-v0.1/resolve/27d67f1b5f57dc0953326b2601d68371d40ea8da/model-00002-of-00002.safetensors", "type": "weights" }, { "label": "Original Mistral 7B architecture implementation", "url": "https://github.com/mistralai/mistral-inference/blob/85448190c8cecaa746e6153020fc7f80bc65b160/mistral/model.py", "type": "primary" }, { "label": "Mistral 7B technical report", "url": "https://arxiv.org/html/2310.06825v1", "type": "primary" }, { "label": "Original release source", "url": "https://mistral.ai/news/announcing-mistral-7b/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "Complete BF16 checkpoint: 7,241,732,096 learned parameters (7.241732096B), verified across all 291 indexed tensors. Official v0.1 documentation separately reports nominal Active (B)=7; this is not an exact active count. The release announcement uses the nominal 7.3B label. Context is the documented 8192 tokens; HF max_position_embeddings=32768 is a distinct configuration limit. Dedicated reasoning and native tool calling are not disclosed for this base checkpoint.", "field_status": { "reasoning": "undisclosed", "tool_call": "undisclosed" }, "audited_at": "2026-10-26T23:13:22.076465Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/331.json" }, { "id": "alibaba/qwen-14b", "slug": "alibaba--qwen-14b", "name": "Qwen-14B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": null, "release_date": "2023-09-25", "date_precision": "day", "open_weights": true, "license": "Tongyi Qianwen License", "parameters_b": 14, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/QwenLM/Qwen/blob/2df8e8ac450fa185c421a08b0090ef81826caa6e/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/QwenLM/Qwen/blob/2df8e8ac450fa185c421a08b0090ef81826caa6e/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Qwen/Qwen-14B/blob/c4051215126d906ac22bb67fe5edb39a921cd831/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/Qwen/Qwen-14B/blob/c4051215126d906ac22bb67fe5edb39a921cd831/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Nominal parameter count. " }, { "id": "alibaba/qwen-14b-chat", "slug": "alibaba--qwen-14b-chat", "name": "Qwen-14B-Chat", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": null, "release_date": "2023-09-25", "date_precision": "day", "open_weights": true, "license": "Tongyi Qianwen License", "parameters_b": 14, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/QwenLM/Qwen/blob/2df8e8ac450fa185c421a08b0090ef81826caa6e/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/QwenLM/Qwen/blob/2df8e8ac450fa185c421a08b0090ef81826caa6e/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Qwen/Qwen-14B-Chat/blob/cdaff792392504e679496a9f386acf3c1e4333a5/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/Qwen/Qwen-14B-Chat/blob/cdaff792392504e679496a9f386acf3c1e4333a5/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Nominal parameter count. " }, { "id": "internlm/internlm-20b", "slug": "internlm--internlm-20b", "name": "InternLM-20B", "developer": "Shanghai AI Laboratory", "developer_id": "internlm", "family": "internlm", "description": null, "release_date": "2023-09-20", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 20, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/InternLM/InternLM/blob/e6118174422224334bb14c5523caea6391fcaaf8/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/InternLM/InternLM/blob/e6118174422224334bb14c5523caea6391fcaaf8/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/internlm/internlm-20b/blob/80729bcf52fbc4553d965926b27304ac5e156d98/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/internlm/internlm-20b/blob/80729bcf52fbc4553d965926b27304ac5e156d98/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Publisher release table names the exact checkpoint and date. " }, { "id": "internlm/internlm-chat-20b", "slug": "internlm--internlm-chat-20b", "name": "InternLM-chat-20B", "developer": "Shanghai AI Laboratory", "developer_id": "internlm", "family": "internlm", "description": null, "release_date": "2023-09-20", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 20, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/InternLM/InternLM/blob/e6118174422224334bb14c5523caea6391fcaaf8/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/InternLM/InternLM/blob/e6118174422224334bb14c5523caea6391fcaaf8/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/internlm/internlm-chat-20b/blob/ce6e0150bc4b525c44f6e450569a05705dcb4e72/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/internlm/internlm-chat-20b/blob/ce6e0150bc4b525c44f6e450569a05705dcb4e72/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Publisher release table names the exact checkpoint and date. " }, { "id": "alibaba/qwen-plus", "slug": "alibaba--qwen-plus", "name": "Qwen Plus", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Proprietary Qwen text model for multilingual chat and thinking, with function calling availability varying by deployment region.", "release_date": "2023-09-13", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 1000000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": true, "source_url": "https://developer.aliyun.com/article/1326924", "source_type": "primary", "sources": [ { "label": "Alibaba Cloud Qwen Plus model documentation", "url": "https://www.alibabacloud.com/help/en/model-studio/qwen-plus", "type": "documentation" }, { "label": "Qwen proprietary versus open-weight model definitions", "url": "https://qwen.readthedocs.io/en/v2.5/getting_started/concepts.html", "type": "documentation" }, { "label": "Alibaba Cloud text generation capability matrix", "url": "https://www.alibabacloud.com/help/en/model-studio/text-generation-model", "type": "documentation" }, { "label": "Alibaba Cloud Qwen Plus snapshot release history", "url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "type": "documentation" }, { "label": "Original Qwen DashScope documentation listing qwen-plus (October 16, 2023)", "url": "https://github.com/QwenLM/Qwen/blob/235aa8f71efd5218323e53a98c065554aeb41272/README_CN.md", "type": "primary" }, { "label": "Qwen original API documentation commit metadata", "url": "https://api.github.com/repos/QwenLM/Qwen/commits/235aa8f71efd5218323e53a98c065554aeb41272", "type": "primary" }, { "label": "Alibaba Cloud opens Tongyi Qianwen to the public and API customers (September 13, 2023)", "url": "https://developer.aliyun.com/article/1326924", "type": "primary" }, { "label": "Alibaba DashScope SDK 1.10.0: published API identifiers", "url": "https://pypi.org/pypi/dashscope/1.10.0/json", "type": "primary" }, { "label": "Alibaba DashScope SDK 1.9.1: published API identifiers", "url": "https://pypi.org/pypi/dashscope/1.9.1/json", "type": "primary" }, { "label": "Alibaba DashScope SDK 1.11.0: published API identifiers", "url": "https://pypi.org/pypi/dashscope/1.11.0/json", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen-plus.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The qwen-plus alias maps to qwen-plus-2025-12-01. Total and active parameters remain undisclosed: no nominal size or named sized-base lineage was found. Architecture remains unresolved. Function calling availability varies by deployment region. The public qwen-plus API launched on September 13, 2023: Alibaba’s public-opening announcement and same-day SDK identify the released alias, replacing the earlier invitation-test qwen-plus-v1 name.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-28T07:57:33.380048Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/016.json" }, { "id": "alibaba/qwen-turbo", "slug": "alibaba--qwen-turbo", "name": "Qwen Turbo", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": "Proprietary Qwen3 text model with thinking and non-thinking modes; the final qwen-turbo alias matched the 2025-04-28 snapshot.", "release_date": "2023-09-13", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 131072, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": false, "source_url": "https://developer.aliyun.com/article/1326924", "source_type": "primary", "sources": [ { "label": "qwen-turbo model information", "url": "https://help.aliyun.com/en/model-studio/qwen-turbo", "type": "docs" }, { "label": "Qwen Key Concepts", "url": "https://qwen.readthedocs.io/en/stable/getting_started/concepts.html", "type": "docs" }, { "label": "Qwen2.5: A Party of Foundation Models!", "url": "https://qwenlm.github.io/blog/qwen2.5/", "type": "announcement" }, { "label": "Extending the Context Length to 1M Tokens!", "url": "https://qwenlm.github.io/blog/qwen2.5-turbo/", "type": "announcement" }, { "label": "Notice on the Decommissioning of Certain Historical Mainline Models", "url": "https://www.alibabacloud.com/en/notice/model_studionotice_on_the_decommissioning_of_certain_historical_mainline_models_731?_p_lc=1", "type": "announcement" }, { "label": "Output price of Qwen-Plus and Qwen-Turbo reduced under Qwen3 thinking mode", "url": "https://www.alibabacloud.com/en/notice/output_price_of_qwenplus_and_qwenturbo_reduced_by_50_under_qwen3s_thinking_mode_521?_p_lc=1", "type": "announcement" }, { "label": "qwen-turbo 模型信息", "url": "https://help.aliyun.com/zh/model-studio/qwen-turbo", "type": "docs" }, { "label": "Qwen3: Think Deeper, Act Faster", "url": "https://qwenlm.github.io/blog/qwen3/", "type": "announcement" }, { "label": "Alibaba Cloud Model Studio: Model lifecycle and updates", "url": "https://www.alibabacloud.com/help/en/model-studio/newly-released-models", "type": "docs" }, { "label": "Original Qwen DashScope documentation listing qwen-turbo (October 16, 2023)", "url": "https://github.com/QwenLM/Qwen/blob/235aa8f71efd5218323e53a98c065554aeb41272/README_CN.md", "type": "primary" }, { "label": "Qwen original API documentation commit metadata", "url": "https://api.github.com/repos/QwenLM/Qwen/commits/235aa8f71efd5218323e53a98c065554aeb41272", "type": "primary" }, { "label": "Alibaba Cloud opens Tongyi Qianwen to the public and API customers (September 13, 2023)", "url": "https://developer.aliyun.com/article/1326924", "type": "primary" }, { "label": "Alibaba DashScope SDK 1.10.0: published API identifiers", "url": "https://pypi.org/pypi/dashscope/1.10.0/json", "type": "primary" }, { "label": "Alibaba DashScope SDK 1.9.1: published API identifiers", "url": "https://pypi.org/pypi/dashscope/1.9.1/json", "type": "primary" }, { "label": "Alibaba DashScope SDK 1.11.0: published API identifiers", "url": "https://pypi.org/pypi/dashscope/1.11.0/json", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/alibaba/qwen-turbo.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The documented alias matches qwen-turbo-2025-04-28. Nominal total and active parameter counts remain undisclosed; no sized base-model lineage is identified. Dense/MoE architecture remains unresolved. Alibaba scheduled retirement in Beijing and Singapore for 2026-10-10; endpoint availability was not tested. The public qwen-turbo API launched on September 13, 2023: Alibaba’s public-opening announcement and same-day SDK identify the released alias, replacing the earlier invitation-test qwen-v1 name.", "field_status": { "license": "not_applicable", "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "undisclosed" }, "audited_at": "2026-10-28T07:57:33.597049Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/017.json" }, { "id": "microsoft/phi-1.5", "slug": "microsoft--phi-1.5", "name": "Phi-1.5", "developer": "Microsoft", "developer_id": "microsoft", "family": "phi", "description": null, "release_date": "2023-09-11", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 1.3, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://arxiv.org/abs/2309.05463v1", "source_type": "primary", "sources": [ { "label": "Dated model-team report announcing the open release", "url": "https://arxiv.org/abs/2309.05463v1", "type": "paper" }, { "label": "Official model card", "url": "https://huggingface.co/microsoft/phi-1_5/blob/77aa61eeac94fbf33d492b9f2744c98b42d5b5eb/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/microsoft/phi-1_5/blob/77aa61eeac94fbf33d492b9f2744c98b42d5b5eb/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "" }, { "id": "baichuan/baichuan2-13b-base", "slug": "baichuan--baichuan2-13b-base", "name": "Baichuan2-13B-Base", "developer": "Baichuan AI", "developer_id": "baichuan", "family": "baichuan", "description": null, "release_date": "2023-09-06", "date_precision": "day", "open_weights": true, "license": "Baichuan 2 Community License", "parameters_b": 13, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/baichuan-inc/Baichuan-7B/blob/6f3ef4633a90c2d8a3e0763d0dec1b8dc11588f5/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/baichuan-inc/Baichuan-7B/blob/6f3ef4633a90c2d8a3e0763d0dec1b8dc11588f5/README.md", "type": "repository" }, { "label": "Publisher release chronology", "url": "https://github.com/baichuan-inc/Baichuan2/blob/ed8b1ae415c62b65a7d41b491ebeb854c54f0653/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/baichuan-inc/Baichuan2-13B-Base/blob/c6f590cab590cf33e78ad834dbd5f9bd6df34a94/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/baichuan-inc/Baichuan2-13B-Base/blob/c6f590cab590cf33e78ad834dbd5f9bd6df34a94/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Nominal parameter count. Dated publisher news names both sizes; the linked release repository names the base and chat checkpoints." }, { "id": "baichuan/baichuan2-13b-chat", "slug": "baichuan--baichuan2-13b-chat", "name": "Baichuan2-13B-Chat", "developer": "Baichuan AI", "developer_id": "baichuan", "family": "baichuan", "description": null, "release_date": "2023-09-06", "date_precision": "day", "open_weights": true, "license": "Baichuan 2 Community License", "parameters_b": 13, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/baichuan-inc/Baichuan-7B/blob/6f3ef4633a90c2d8a3e0763d0dec1b8dc11588f5/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/baichuan-inc/Baichuan-7B/blob/6f3ef4633a90c2d8a3e0763d0dec1b8dc11588f5/README.md", "type": "repository" }, { "label": "Publisher release chronology", "url": "https://github.com/baichuan-inc/Baichuan2/blob/ed8b1ae415c62b65a7d41b491ebeb854c54f0653/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/baichuan-inc/Baichuan2-13B-Chat/blob/c8d877c7ca596d9aeff429d43bff06e288684f45/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/baichuan-inc/Baichuan2-13B-Chat/blob/c8d877c7ca596d9aeff429d43bff06e288684f45/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Nominal parameter count. Dated publisher news names both sizes; the linked release repository names the base and chat checkpoints." }, { "id": "baichuan/baichuan2-7b-base", "slug": "baichuan--baichuan2-7b-base", "name": "Baichuan2-7B-Base", "developer": "Baichuan AI", "developer_id": "baichuan", "family": "baichuan", "description": null, "release_date": "2023-09-06", "date_precision": "day", "open_weights": true, "license": "Baichuan 2 Community License", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/baichuan-inc/Baichuan-7B/blob/6f3ef4633a90c2d8a3e0763d0dec1b8dc11588f5/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/baichuan-inc/Baichuan-7B/blob/6f3ef4633a90c2d8a3e0763d0dec1b8dc11588f5/README.md", "type": "repository" }, { "label": "Publisher release chronology", "url": "https://github.com/baichuan-inc/Baichuan2/blob/ed8b1ae415c62b65a7d41b491ebeb854c54f0653/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/baichuan-inc/Baichuan2-7B-Base/blob/f9d4d8dd2f7a3dbede3bda3b0cf0224e9272bbe5/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/baichuan-inc/Baichuan2-7B-Base/blob/f9d4d8dd2f7a3dbede3bda3b0cf0224e9272bbe5/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Nominal parameter count. Dated publisher news names both sizes; the linked release repository names the base and chat checkpoints." }, { "id": "baichuan/baichuan2-7b-chat", "slug": "baichuan--baichuan2-7b-chat", "name": "Baichuan2-7B-Chat", "developer": "Baichuan AI", "developer_id": "baichuan", "family": "baichuan", "description": null, "release_date": "2023-09-06", "date_precision": "day", "open_weights": true, "license": "Baichuan 2 Community License", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/baichuan-inc/Baichuan-7B/blob/6f3ef4633a90c2d8a3e0763d0dec1b8dc11588f5/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/baichuan-inc/Baichuan-7B/blob/6f3ef4633a90c2d8a3e0763d0dec1b8dc11588f5/README.md", "type": "repository" }, { "label": "Publisher release chronology", "url": "https://github.com/baichuan-inc/Baichuan2/blob/ed8b1ae415c62b65a7d41b491ebeb854c54f0653/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/baichuan-inc/Baichuan2-7B-Chat/blob/ea66ced17780ca3db39bc9f8aa601d8463db3da5/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/baichuan-inc/Baichuan2-7B-Chat/blob/ea66ced17780ca3db39bc9f8aa601d8463db3da5/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Nominal parameter count. Dated publisher news names both sizes; the linked release repository names the base and chat checkpoints." }, { "id": "tiiuae/falcon-180b", "slug": "tiiuae--falcon-180b", "name": "Falcon 180B", "developer": "Technology Innovation Institute", "developer_id": "tiiuae", "family": "falcon", "description": null, "release_date": "2023-09-06", "date_precision": "day", "open_weights": true, "license": "Falcon-180B TII License", "parameters_b": 180, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.tii.ae/news/technology-innovation-institute-introduces-worlds-most-powerful-open-llm-falcon-180b", "source_type": "primary", "sources": [ { "label": "Dated developer release", "url": "https://www.tii.ae/news/technology-innovation-institute-introduces-worlds-most-powerful-open-llm-falcon-180b", "type": "announcement" }, { "label": "Launch documentation coauthored by the Falcon team", "url": "https://huggingface.co/blog/falcon-180b", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/tiiuae/falcon-180B", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Weights require accepting the publisher license; hosted use has additional terms.", "availability": "public" }, { "id": "tiiuae/falcon-180b-chat", "slug": "tiiuae--falcon-180b-chat", "name": "Falcon 180B Chat", "developer": "Technology Innovation Institute", "developer_id": "tiiuae", "family": "falcon", "description": null, "release_date": "2023-09-06", "date_precision": "day", "open_weights": true, "license": "Falcon-180B TII License", "parameters_b": 180, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.tii.ae/news/technology-innovation-institute-introduces-worlds-most-powerful-open-llm-falcon-180b", "source_type": "primary", "sources": [ { "label": "Dated developer release", "url": "https://www.tii.ae/news/technology-innovation-institute-introduces-worlds-most-powerful-open-llm-falcon-180b", "type": "announcement" }, { "label": "Launch documentation coauthored by the Falcon team", "url": "https://huggingface.co/blog/falcon-180b", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/tiiuae/falcon-180B-chat", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Weights require accepting the publisher license; hosted use has additional terms.", "availability": "public" }, { "id": "meta/code-llama-13b", "slug": "meta--code-llama-13b", "name": "Code Llama 13B", "developer": "Meta", "developer_id": "meta", "family": "code-llama", "description": null, "release_date": "2023-08-24", "date_precision": "day", "open_weights": true, "license": "Llama 2 Community License", "parameters_b": 13, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/", "source_type": "primary", "sources": [ { "label": "Dated release naming all trained sizes and variants", "url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/codellama/CodeLlama-13b-hf/blob/8da65ff4ee20f74ecd107ca9d54f9f121b279860/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/codellama/CodeLlama-13b-hf/blob/8da65ff4ee20f74ecd107ca9d54f9f121b279860/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:26:10.259Z", "last_seen": "2026-10-29T10:26:10.259Z", "upstream_present": false, "availability": "public", "notes": "Meta’s continued training of its own Llama 2 checkpoint. Context is the published configuration/training window; the paper evaluates extrapolation to 100K tokens." }, { "id": "meta/code-llama-13b-instruct", "slug": "meta--code-llama-13b-instruct", "name": "Code Llama 13B Instruct", "developer": "Meta", "developer_id": "meta", "family": "code-llama", "description": null, "release_date": "2023-08-24", "date_precision": "day", "open_weights": true, "license": "Llama 2 Community License", "parameters_b": 13, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/", "source_type": "primary", "sources": [ { "label": "Dated release naming all trained sizes and variants", "url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/codellama/CodeLlama-13b-Instruct-hf/blob/745795438019e47e4dad1347a0093e11deee4c68/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/codellama/CodeLlama-13b-Instruct-hf/blob/745795438019e47e4dad1347a0093e11deee4c68/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:26:10.259Z", "last_seen": "2026-10-29T10:26:10.259Z", "upstream_present": false, "availability": "public", "notes": "Meta’s continued training of its own Llama 2 checkpoint. Context is the published configuration/training window; the paper evaluates extrapolation to 100K tokens." }, { "id": "meta/code-llama-13b-python", "slug": "meta--code-llama-13b-python", "name": "Code Llama 13B Python", "developer": "Meta", "developer_id": "meta", "family": "code-llama", "description": null, "release_date": "2023-08-24", "date_precision": "day", "open_weights": true, "license": "Llama 2 Community License", "parameters_b": 13, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/", "source_type": "primary", "sources": [ { "label": "Dated release naming all trained sizes and variants", "url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/codellama/CodeLlama-13b-Python-hf/blob/bced38894a8c7198df0486bc444c203f3159d266/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/codellama/CodeLlama-13b-Python-hf/blob/bced38894a8c7198df0486bc444c203f3159d266/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:26:10.259Z", "last_seen": "2026-10-29T10:26:10.259Z", "upstream_present": false, "availability": "public", "notes": "Meta’s continued training of its own Llama 2 checkpoint. Context is the published configuration/training window; the paper evaluates extrapolation to 100K tokens." }, { "id": "meta/code-llama-34b", "slug": "meta--code-llama-34b", "name": "Code Llama 34B", "developer": "Meta", "developer_id": "meta", "family": "code-llama", "description": null, "release_date": "2023-08-24", "date_precision": "day", "open_weights": true, "license": "Llama 2 Community License", "parameters_b": 34, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/", "source_type": "primary", "sources": [ { "label": "Dated release naming all trained sizes and variants", "url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/codellama/CodeLlama-34b-hf/blob/6008b9656730b71c7d19a15370c7ff6d2902f4ef/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/codellama/CodeLlama-34b-hf/blob/6008b9656730b71c7d19a15370c7ff6d2902f4ef/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:26:10.259Z", "last_seen": "2026-10-29T10:26:10.259Z", "upstream_present": false, "availability": "public", "notes": "Meta’s continued training of its own Llama 2 checkpoint. Context is the published configuration/training window; the paper evaluates extrapolation to 100K tokens." }, { "id": "meta/code-llama-34b-instruct", "slug": "meta--code-llama-34b-instruct", "name": "Code Llama 34B Instruct", "developer": "Meta", "developer_id": "meta", "family": "code-llama", "description": null, "release_date": "2023-08-24", "date_precision": "day", "open_weights": true, "license": "Llama 2 Community License", "parameters_b": 34, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/", "source_type": "primary", "sources": [ { "label": "Dated release naming all trained sizes and variants", "url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/codellama/CodeLlama-34b-Instruct-hf/blob/d4c1c474abcacd32d2a6eda45f9811d38c83e93d/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/codellama/CodeLlama-34b-Instruct-hf/blob/d4c1c474abcacd32d2a6eda45f9811d38c83e93d/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:26:10.259Z", "last_seen": "2026-10-29T10:26:10.259Z", "upstream_present": false, "availability": "public", "notes": "Meta’s continued training of its own Llama 2 checkpoint. Context is the published configuration/training window; the paper evaluates extrapolation to 100K tokens." }, { "id": "meta/code-llama-34b-python", "slug": "meta--code-llama-34b-python", "name": "Code Llama 34B Python", "developer": "Meta", "developer_id": "meta", "family": "code-llama", "description": null, "release_date": "2023-08-24", "date_precision": "day", "open_weights": true, "license": "Llama 2 Community License", "parameters_b": 34, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/", "source_type": "primary", "sources": [ { "label": "Dated release naming all trained sizes and variants", "url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/codellama/CodeLlama-34b-Python-hf/blob/7daa5d7ae0be52aea6c5523df2c887853f1e7035/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/codellama/CodeLlama-34b-Python-hf/blob/7daa5d7ae0be52aea6c5523df2c887853f1e7035/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:26:10.259Z", "last_seen": "2026-10-29T10:26:10.259Z", "upstream_present": false, "availability": "public", "notes": "Meta’s continued training of its own Llama 2 checkpoint. Context is the published configuration/training window; the paper evaluates extrapolation to 100K tokens." }, { "id": "meta/code-llama-7b", "slug": "meta--code-llama-7b", "name": "Code Llama 7B", "developer": "Meta", "developer_id": "meta", "family": "code-llama", "description": null, "release_date": "2023-08-24", "date_precision": "day", "open_weights": true, "license": "Llama 2 Community License", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/", "source_type": "primary", "sources": [ { "label": "Dated release naming all trained sizes and variants", "url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/codellama/CodeLlama-7b-hf/blob/6c284d1468fe6c413cf56183e69b194dcfa27fe6/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/codellama/CodeLlama-7b-hf/blob/6c284d1468fe6c413cf56183e69b194dcfa27fe6/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:26:10.259Z", "last_seen": "2026-10-29T10:26:10.259Z", "upstream_present": false, "availability": "public", "notes": "Meta’s continued training of its own Llama 2 checkpoint. Context is the published configuration/training window; the paper evaluates extrapolation to 100K tokens." }, { "id": "meta/code-llama-7b-instruct", "slug": "meta--code-llama-7b-instruct", "name": "Code Llama 7B Instruct", "developer": "Meta", "developer_id": "meta", "family": "code-llama", "description": null, "release_date": "2023-08-24", "date_precision": "day", "open_weights": true, "license": "Llama 2 Community License", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/", "source_type": "primary", "sources": [ { "label": "Dated release naming all trained sizes and variants", "url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/codellama/CodeLlama-7b-Instruct-hf/blob/22cb240e0292b0b5ab4c17ccd97aa3a2f799cbed/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/codellama/CodeLlama-7b-Instruct-hf/blob/22cb240e0292b0b5ab4c17ccd97aa3a2f799cbed/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:26:10.259Z", "last_seen": "2026-10-29T10:26:10.259Z", "upstream_present": false, "availability": "public", "notes": "Meta’s continued training of its own Llama 2 checkpoint. Context is the published configuration/training window; the paper evaluates extrapolation to 100K tokens." }, { "id": "meta/code-llama-7b-python", "slug": "meta--code-llama-7b-python", "name": "Code Llama 7B Python", "developer": "Meta", "developer_id": "meta", "family": "code-llama", "description": null, "release_date": "2023-08-24", "date_precision": "day", "open_weights": true, "license": "Llama 2 Community License", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/", "source_type": "primary", "sources": [ { "label": "Dated release naming all trained sizes and variants", "url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/codellama/CodeLlama-7b-Python-hf/blob/d4178f5d2eead875e627ec487b23679266319b7f/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/codellama/CodeLlama-7b-Python-hf/blob/d4178f5d2eead875e627ec487b23679266319b7f/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:26:10.259Z", "last_seen": "2026-10-29T10:26:10.259Z", "upstream_present": false, "availability": "public", "notes": "Meta’s continued training of its own Llama 2 checkpoint. Context is the published configuration/training window; the paper evaluates extrapolation to 100K tokens." }, { "id": "alibaba/qwen-vl", "slug": "alibaba--qwen-vl", "name": "Qwen-VL", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": null, "release_date": "2023-08-22", "date_precision": "day", "open_weights": true, "license": "Tongyi Qianwen License", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/QwenLM/Qwen-VL/blob/aa00ed04091eea5fcdd32985e7915f1c53e7d599/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/QwenLM/Qwen-VL/blob/aa00ed04091eea5fcdd32985e7915f1c53e7d599/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Qwen/Qwen-VL/blob/0547ed36a86561e2e42fecec8fd0c4f6953e33c4/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/Qwen/Qwen-VL/blob/0547ed36a86561e2e42fecec8fd0c4f6953e33c4/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "The published 7B figure describes the language backbone; a primary total including the modality encoder has not yet been verified. " }, { "id": "alibaba/qwen-vl-chat", "slug": "alibaba--qwen-vl-chat", "name": "Qwen-VL-Chat", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": null, "release_date": "2023-08-22", "date_precision": "day", "open_weights": true, "license": "Tongyi Qianwen License", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/QwenLM/Qwen-VL/blob/aa00ed04091eea5fcdd32985e7915f1c53e7d599/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/QwenLM/Qwen-VL/blob/aa00ed04091eea5fcdd32985e7915f1c53e7d599/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Qwen/Qwen-VL-Chat/blob/f57cfbd358cb56b710d963669ad1bcfb44cdcdd8/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/Qwen/Qwen-VL-Chat/blob/f57cfbd358cb56b710d963669ad1bcfb44cdcdd8/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "The published 7B figure describes the language backbone; a primary total including the modality encoder has not yet been verified. " }, { "id": "internlm/internlm-chat-7b-v1.1", "slug": "internlm--internlm-chat-7b-v1.1", "name": "InternLM-chat-7B-v1.1", "developer": "Shanghai AI Laboratory", "developer_id": "internlm", "family": "internlm", "description": null, "release_date": "2023-08-22", "date_precision": "day", "open_weights": true, "license": "InternLM Model License", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://github.com/InternLM/InternLM/blob/e6118174422224334bb14c5523caea6391fcaaf8/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/InternLM/InternLM/blob/e6118174422224334bb14c5523caea6391fcaaf8/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/internlm/internlm-chat-7b-v1.1", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Publisher release table names the exact checkpoint and date. The original checkpoint endpoint now returns 401; release and nominal size are retained in the dated official repository." }, { "id": "anthropic/claude-instant-1.2", "slug": "anthropic--claude-instant-1.2", "name": "Claude Instant 1.2", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude", "description": "Historical Anthropic text model in the Claude Instant line, released through the API on August 9, 2023, with a 100,000-token context window. The Anthropic API model was retired on November 6, 2024.", "release_date": "2023-08-09", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 100000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.anthropic.com/news/releasing-claude-instant-1-2", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://www.anthropic.com/news/releasing-claude-instant-1-2", "type": "announcement" }, { "label": "Exact-version long-context evaluation", "url": "https://www.anthropic.com/news/prompting-long-context", "type": "documentation" }, { "label": "Anthropic model retirement history", "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "type": "documentation" }, { "label": "Anthropic statement on Claude weight availability", "url": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities", "type": "research" }, { "label": "Claude 3 family card with later addenda — full accessible copy", "url": "https://www-cdn.anthropic.com/c6a80a657af445f40e31afac050f3bf76d3b1404.pdf", "type": "model_card" }, { "label": "Claude model pricing linked from the Instant 1.2 release", "url": "https://www-cdn.anthropic.com/90df03aed08b794ab03c5a7bf28b2ad9cf26cf3c/model_pricing_july2023.pdf", "type": "documentation" }, { "label": "Collective Constitutional AI same-size relationship", "url": "https://arxiv.org/html/2406.07814v1", "type": "research" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "Anthropic discloses neither a nominal total nor an active parameter count for Instant 1.2; its same-sized Collective Constitutional AI research models also have no published count. Architecture and native reasoning/tool-calling support remain unverified for this exact revision; the original Claude 2 model card discusses Instant 1.1. Anthropic retired claude-instant-1.2 on November 6, 2024; partner platforms may have separate lifecycles.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "unverified", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T10:31:04.208422Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/101.json" }, { "id": "stabilityai/stablecode-completion-alpha-3b", "slug": "stabilityai--stablecode-completion-alpha-3b", "name": "StableCode-Completion-Alpha-3B", "developer": "Stability AI", "developer_id": "stabilityai", "family": "stablecode", "description": null, "release_date": "2023-08-08", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 2.79643136, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://stability.ai/news-updates/stablecode-llm-generative-ai-coding", "source_type": "primary", "sources": [ { "label": "Dated release of all three variants", "url": "https://stability.ai/news-updates/stablecode-llm-generative-ai-coding", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/stabilityai/stablecode-completion-alpha-3b/blob/04fc3a2d85e4a12ccf00f01904222d3b4af25ac4/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/stabilityai/stablecode-completion-alpha-3b/blob/04fc3a2d85e4a12ccf00f01904222d3b4af25ac4/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "Exact total and context from the publisher model architecture table." }, { "id": "stabilityai/stablecode-completion-alpha-3b-4k", "slug": "stabilityai--stablecode-completion-alpha-3b-4k", "name": "StableCode-Completion-Alpha-3B-4K", "developer": "Stability AI", "developer_id": "stabilityai", "family": "stablecode", "description": null, "release_date": "2023-08-08", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 2.79643136, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://stability.ai/news-updates/stablecode-llm-generative-ai-coding", "source_type": "primary", "sources": [ { "label": "Dated release of all three variants", "url": "https://stability.ai/news-updates/stablecode-llm-generative-ai-coding", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/stabilityai/stablecode-completion-alpha-3b-4k/blob/7ccb1408b35c33832be9760138d621196e524866/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/stabilityai/stablecode-completion-alpha-3b-4k/blob/7ccb1408b35c33832be9760138d621196e524866/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "Exact total and context from the publisher model architecture table." }, { "id": "stabilityai/stablecode-instruct-alpha-3b", "slug": "stabilityai--stablecode-instruct-alpha-3b", "name": "StableCode-Instruct-Alpha-3B", "developer": "Stability AI", "developer_id": "stabilityai", "family": "stablecode", "description": null, "release_date": "2023-08-08", "date_precision": "day", "open_weights": true, "license": "StableCode Research License", "parameters_b": 2.79643136, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://stability.ai/news-updates/stablecode-llm-generative-ai-coding", "source_type": "primary", "sources": [ { "label": "Dated release of all three variants", "url": "https://stability.ai/news-updates/stablecode-llm-generative-ai-coding", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/stabilityai/stablecode-instruct-alpha-3b/blob/3e6a4ab529afd87112152f397fbbedcd6a437f78/README.md", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "Exact total and context from the publisher model architecture table." }, { "id": "stabilityai/stablelm-base-alpha-3b-v2", "slug": "stabilityai--stablelm-base-alpha-3b-v2", "name": "StableLM Base Alpha 3B v2", "developer": "Stability AI", "developer_id": "stabilityai", "family": "stablelm", "description": null, "release_date": "2023-08-05", "date_precision": "day", "open_weights": true, "license": "CC-BY-SA-4.0", "parameters_b": 2.79643136, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/Stability-AI/StableLM/blob/main/README.md", "source_type": "primary", "sources": [ { "label": "Dated release changelog", "url": "https://github.com/Stability-AI/StableLM/blob/main/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/stabilityai/stablelm-base-alpha-3b-v2/blob/997d95943eee0697ba9e9cd2b981b9fad6962d9e/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/stabilityai/stablelm-base-alpha-3b-v2/blob/997d95943eee0697ba9e9cd2b981b9fad6962d9e/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Exact total comes from the model card.", "availability": "public" }, { "id": "stabilityai/stablelm-base-alpha-7b-v2", "slug": "stabilityai--stablelm-base-alpha-7b-v2", "name": "StableLM Base Alpha 7B v2", "developer": "Stability AI", "developer_id": "stabilityai", "family": "stablelm", "description": null, "release_date": "2023-08-05", "date_precision": "day", "open_weights": true, "license": "CC-BY-SA-4.0", "parameters_b": 6.89020928, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/Stability-AI/StableLM/blob/main/README.md", "source_type": "primary", "sources": [ { "label": "Dated release changelog", "url": "https://github.com/Stability-AI/StableLM/blob/main/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/stabilityai/stablelm-base-alpha-7b-v2/blob/6d176503c1496b3c96a2bceb76a7d37695582618/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/stabilityai/stablelm-base-alpha-7b-v2/blob/6d176503c1496b3c96a2bceb76a7d37695582618/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Exact total comes from the model card.", "availability": "public" }, { "id": "alibaba/qwen-7b", "slug": "alibaba--qwen-7b", "name": "Qwen-7B", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": null, "release_date": "2023-08-03", "date_precision": "day", "open_weights": true, "license": "Tongyi Qianwen License", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/QwenLM/Qwen/blob/2df8e8ac450fa185c421a08b0090ef81826caa6e/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/QwenLM/Qwen/blob/2df8e8ac450fa185c421a08b0090ef81826caa6e/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Qwen/Qwen-7B/blob/ef3c5c9c57b252f3149c1408daf4d649ec8b6c85/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/Qwen/Qwen-7B/blob/ef3c5c9c57b252f3149c1408daf4d649ec8b6c85/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Nominal parameter count. The same repository received updated weights and a larger context window on September 25, 2023; the date records the first release." }, { "id": "alibaba/qwen-7b-chat", "slug": "alibaba--qwen-7b-chat", "name": "Qwen-7B-Chat", "developer": "Alibaba", "developer_id": "alibaba", "family": "qwen", "description": null, "release_date": "2023-08-03", "date_precision": "day", "open_weights": true, "license": "Tongyi Qianwen License", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/QwenLM/Qwen/blob/2df8e8ac450fa185c421a08b0090ef81826caa6e/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/QwenLM/Qwen/blob/2df8e8ac450fa185c421a08b0090ef81826caa6e/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Qwen/Qwen-7B-Chat/blob/93a65d34827a3cc269b727e67004743b723e2f83/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/Qwen/Qwen-7B-Chat/blob/93a65d34827a3cc269b727e67004743b723e2f83/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Nominal parameter count. The same repository received updated weights and a larger context window on September 25, 2023; the date records the first release." }, { "id": "lmsys/vicuna-13b-v1.5", "slug": "lmsys--vicuna-13b-v1.5", "name": "Vicuna 13B v1.5", "developer": "LMSYS", "developer_id": "lmsys", "family": "vicuna", "description": null, "release_date": "2023-08-01", "date_precision": "day", "open_weights": true, "license": "Llama 2 Community License", "parameters_b": 13, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/lmsys/vicuna-13b-v1.5", "source_type": "primary", "sources": [ { "label": "LMSYS release history", "url": "https://github.com/lm-sys/FastChat/blob/587d5cfa1609a43d192cedb8441cac3c17db105d/docs/vicuna_weights_version.md", "type": "documentation" }, { "label": "Official model card", "url": "https://huggingface.co/lmsys/vicuna-13b-v1.5/blob/c8327bf999adbd2efe2e75f6509fa01436100dc2/README.md", "type": "model_card" }, { "label": "Official configuration", "url": "https://huggingface.co/lmsys/vicuna-13b-v1.5/blob/c8327bf999adbd2efe2e75f6509fa01436100dc2/config.json", "type": "config" }, { "label": "LMSYS institutional provenance", "url": "https://www.lmsys.org/about/", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:09:10.442Z", "last_seen": "2026-10-30T07:09:10.442Z", "upstream_present": false, "availability": "public", "notes": "Institutional LMSYS release. Nominal publisher size; independently trained revision verified by its release notes and model card." }, { "id": "lmsys/vicuna-13b-v1.5-16k", "slug": "lmsys--vicuna-13b-v1.5-16k", "name": "Vicuna 13B v1.5 16k", "developer": "LMSYS", "developer_id": "lmsys", "family": "vicuna", "description": null, "release_date": "2023-08-01", "date_precision": "day", "open_weights": true, "license": "Llama 2 Community License", "parameters_b": 13, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/lmsys/vicuna-13b-v1.5-16k", "source_type": "primary", "sources": [ { "label": "LMSYS release history", "url": "https://github.com/lm-sys/FastChat/blob/587d5cfa1609a43d192cedb8441cac3c17db105d/docs/vicuna_weights_version.md", "type": "documentation" }, { "label": "Official model card", "url": "https://huggingface.co/lmsys/vicuna-13b-v1.5-16k/blob/17c61f9ca19f5a7a04e96b2cc0d9bcf2920cb8c2/README.md", "type": "model_card" }, { "label": "Official configuration", "url": "https://huggingface.co/lmsys/vicuna-13b-v1.5-16k/blob/17c61f9ca19f5a7a04e96b2cc0d9bcf2920cb8c2/config.json", "type": "config" }, { "label": "LMSYS institutional provenance", "url": "https://www.lmsys.org/about/", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:09:10.442Z", "last_seen": "2026-10-30T07:09:10.442Z", "upstream_present": false, "availability": "public", "notes": "Institutional LMSYS release. Nominal publisher size; independently trained revision verified by its release notes and model card. The model card confirms separate supervised fine-tuning on 16K sequences with linear RoPE scaling; this is a trained variant." }, { "id": "lmsys/vicuna-7b-v1.5", "slug": "lmsys--vicuna-7b-v1.5", "name": "Vicuna 7B v1.5", "developer": "LMSYS", "developer_id": "lmsys", "family": "vicuna", "description": null, "release_date": "2023-08-01", "date_precision": "day", "open_weights": true, "license": "Llama 2 Community License", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/lmsys/vicuna-7b-v1.5", "source_type": "primary", "sources": [ { "label": "LMSYS release history", "url": "https://github.com/lm-sys/FastChat/blob/587d5cfa1609a43d192cedb8441cac3c17db105d/docs/vicuna_weights_version.md", "type": "documentation" }, { "label": "Official model card", "url": "https://huggingface.co/lmsys/vicuna-7b-v1.5/blob/3321f76e3f527bd14065daf69dad9344000a201d/README.md", "type": "model_card" }, { "label": "Official configuration", "url": "https://huggingface.co/lmsys/vicuna-7b-v1.5/blob/3321f76e3f527bd14065daf69dad9344000a201d/config.json", "type": "config" }, { "label": "LMSYS institutional provenance", "url": "https://www.lmsys.org/about/", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:09:10.442Z", "last_seen": "2026-10-30T07:09:10.442Z", "upstream_present": false, "availability": "public", "notes": "Institutional LMSYS release. Nominal publisher size; independently trained revision verified by its release notes and model card." }, { "id": "lmsys/vicuna-7b-v1.5-16k", "slug": "lmsys--vicuna-7b-v1.5-16k", "name": "Vicuna 7B v1.5 16k", "developer": "LMSYS", "developer_id": "lmsys", "family": "vicuna", "description": null, "release_date": "2023-08-01", "date_precision": "day", "open_weights": true, "license": "Llama 2 Community License", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/lmsys/vicuna-7b-v1.5-16k", "source_type": "primary", "sources": [ { "label": "LMSYS release history", "url": "https://github.com/lm-sys/FastChat/blob/587d5cfa1609a43d192cedb8441cac3c17db105d/docs/vicuna_weights_version.md", "type": "documentation" }, { "label": "Official model card", "url": "https://huggingface.co/lmsys/vicuna-7b-v1.5-16k/blob/c8df3ca4436a3bce5c4b5877e0117032081852b4/README.md", "type": "model_card" }, { "label": "Official configuration", "url": "https://huggingface.co/lmsys/vicuna-7b-v1.5-16k/blob/c8df3ca4436a3bce5c4b5877e0117032081852b4/config.json", "type": "config" }, { "label": "LMSYS institutional provenance", "url": "https://www.lmsys.org/about/", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:09:10.442Z", "last_seen": "2026-10-30T07:09:10.442Z", "upstream_present": false, "availability": "public", "notes": "Institutional LMSYS release. Nominal publisher size; independently trained revision verified by its release notes and model card. The model card confirms separate supervised fine-tuning on 16K sequences with linear RoPE scaling; this is a trained variant." }, { "id": "zhipuai/chatglm2-6b-32k", "slug": "zhipuai--chatglm2-6b-32k", "name": "ChatGLM2-6B-32K", "developer": "Zhipu AI / Tsinghua University", "developer_id": "zhipuai", "family": "chatglm", "description": null, "release_date": "2023-07-31", "date_precision": "day", "open_weights": true, "license": "ChatGLM Model License", "parameters_b": 6, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/THUDM/ChatGLM2-6B/blob/cb8e8b43c0951b32614f25c03e1ab593a0603a1c/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/THUDM/ChatGLM2-6B/blob/cb8e8b43c0951b32614f25c03e1ab593a0603a1c/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/THUDM/chatglm2-6b-32k/blob/09eee0bd6ec97413c6fb7fed5071689fbbdc7a1e/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/THUDM/chatglm2-6b-32k/blob/09eee0bd6ec97413c6fb7fed5071689fbbdc7a1e/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Nominal parameter count from the publisher." }, { "id": "zhipuai/codegeex2-6b", "slug": "zhipuai--codegeex2-6b", "name": "CodeGeeX2-6B", "developer": "Zhipu AI / Tsinghua University", "developer_id": "zhipuai", "family": "codegeex", "description": null, "release_date": "2023-07-25", "date_precision": "day", "open_weights": true, "license": "CodeGeeX2 Model License", "parameters_b": 6, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/THUDM/ChatGLM-6B/blob/401bf3a8a7dd8a26fba189551dccfc61a7079b4e/README_en.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/THUDM/ChatGLM-6B/blob/401bf3a8a7dd8a26fba189551dccfc61a7079b4e/README_en.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/THUDM/codegeex2-6b/blob/341d8174042d97b618dc5a34b2a1e743be516375/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/THUDM/codegeex2-6b/blob/341d8174042d97b618dc5a34b2a1e743be516375/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Nominal parameter count from the publisher." }, { "id": "meta/llama-2-13b", "slug": "meta--llama-2-13b", "name": "Llama 2 13B", "developer": "Meta", "developer_id": "meta", "family": "llama-2", "description": "Meta's Llama 2 13B pretrained text-generation model, using a dense autoregressive Transformer with multi-head attention and a 4,096-token context.", "release_date": "2023-07-18", "date_precision": "day", "open_weights": true, "license": "Llama 2 Community License", "parameters_b": 13, "active_parameters_b": 13, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/llama-2/", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://ai.meta.com/blog/llama-2/", "type": "announcement" }, { "label": "Official model card and license", "url": "https://huggingface.co/meta-llama/Llama-2-13b-hf", "type": "weights" }, { "label": "Llama 2 technical paper", "url": "https://arxiv.org/html/2307.09288v2", "type": "primary" }, { "label": "Exact model license", "url": "https://huggingface.co/meta-llama/Llama-2-13b-hf/blob/main/LICENSE.txt", "type": "primary" }, { "label": "Llama 2 reference documentation", "url": "https://raw.githubusercontent.com/meta-llama/llama/llama_v2/README.md", "type": "primary" }, { "label": "Llama 2 reference architecture", "url": "https://raw.githubusercontent.com/meta-llama/llama/llama_v2/llama/model.py", "type": "primary" }, { "label": "Mistral comparison: nominal active parameters, Table 2", "url": "https://arxiv.org/html/2401.04088v1", "type": "primary" }, { "label": "DeepSeek comparison: activated parameters, Figure 1(a)", "url": "https://arxiv.org/html/2405.04434v1", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "Total size is Meta’s nominal 13B. Active size is nominally 13B under Mistral’s Mixtral benchmark convention, corroborated approximately by DeepSeek; neither is a tensor-exact count. The original base model uses multi-head attention and a 4096-token context. Released weights require approval and the Llama 2 Community License. Dedicated thinking and native tool calling are not documented for this base checkpoint.", "field_status": { "reasoning": "undisclosed", "tool_call": "undisclosed" }, "audited_at": "2026-10-27T13:53:09.566363Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/258.json" }, { "id": "meta/llama-2-13b-chat", "slug": "meta--llama-2-13b-chat", "name": "Llama 2 13B Chat", "developer": "Meta", "developer_id": "meta", "family": "llama-2", "description": "Meta's nominal 13B-parameter Llama 2 dialogue model, fine-tuned with supervised learning and RLHF, with text input/output and a 4,096-token context.", "release_date": "2023-07-18", "date_precision": "day", "open_weights": true, "license": "Llama 2 Community License", "parameters_b": 13, "active_parameters_b": 13, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/llama-2/", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://ai.meta.com/blog/llama-2/", "type": "announcement" }, { "label": "Official model card and license", "url": "https://huggingface.co/meta-llama/Llama-2-13b-chat-hf", "type": "weights" }, { "label": "Llama 2 technical report", "url": "https://arxiv.org/html/2307.09288v2", "type": "primary" }, { "label": "Llama 2 reference model code", "url": "https://github.com/meta-llama/llama/blob/7565eb6fee2175b2d4fe2cfb45067a61b35d7f5e/llama/model.py", "type": "primary" }, { "label": "Llama 2 reference README", "url": "https://github.com/meta-llama/llama/blob/7565eb6fee2175b2d4fe2cfb45067a61b35d7f5e/README.md", "type": "primary" }, { "label": "Llama 2 Community License", "url": "https://github.com/meta-llama/llama/blob/7565eb6fee2175b2d4fe2cfb45067a61b35d7f5e/LICENSE", "type": "primary" }, { "label": "Llama 2 reference chat interface", "url": "https://github.com/meta-llama/llama/blob/7565eb6fee2175b2d4fe2cfb45067a61b35d7f5e/llama/generation.py", "type": "primary" }, { "label": "Later Meta SKU utility (conflicts with original 13B attention specification)", "url": "https://github.com/meta-llama/llama-models/blob/main/models/sku_list.py", "type": "primary" }, { "label": "Mistral Mixtral paper: nominal active-count comparison for Llama 2 13B base", "url": "https://arxiv.org/html/2401.04088v1", "type": "primary" }, { "label": "Meta Llama 2 release manifest: shared tokenizer for base and Chat", "url": "https://github.com/meta-llama/llama/blob/7565eb6fee2175b2d4fe2cfb45067a61b35d7f5e/download.sh", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "Total 13B is Meta’s nominal Chat specification. Active 13B is a nominal inference from Mistral’s published Llama 2 13B base comparison and Meta’s documented Chat fine-tuning lineage with shared dense architecture and tokenizer. Neither is an exact tensor count. The original 13B release uses MHA despite a conflicting later Meta SKU utility. Dedicated reasoning/tool-call features are not separately disclosed.", "field_status": { "reasoning": "undisclosed", "tool_call": "undisclosed" }, "audited_at": "2026-10-27T13:53:47.517823Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/259.json" }, { "id": "meta/llama-2-70b", "slug": "meta--llama-2-70b", "name": "Llama 2 70B", "developer": "Meta", "developer_id": "meta", "family": "llama-2", "description": "Llama 2 pretrained checkpoint.", "release_date": "2023-07-18", "date_precision": "day", "open_weights": true, "license": "Llama 2 Community License", "parameters_b": 70, "active_parameters_b": 70, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/llama-2/", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://ai.meta.com/blog/llama-2/", "type": "announcement" }, { "label": "Official model card and license", "url": "https://huggingface.co/meta-llama/Llama-2-70b-hf", "type": "weights" }, { "label": "Original model card", "url": "https://raw.githubusercontent.com/meta-llama/llama/llama_v2/MODEL_CARD.md", "type": "primary" }, { "label": "Llama 2 technical paper", "url": "https://arxiv.org/html/2307.09288v2", "type": "primary" }, { "label": "Original model implementation", "url": "https://raw.githubusercontent.com/meta-llama/llama/llama_v2/llama/model.py", "type": "primary" }, { "label": "Llama 2 Community License", "url": "https://raw.githubusercontent.com/meta-llama/llama/llama_v2/LICENSE", "type": "primary" }, { "label": "Original inference README", "url": "https://raw.githubusercontent.com/meta-llama/llama/llama_v2/README.md", "type": "primary" }, { "label": "Mixtral paper: nominal active-parameter comparison", "url": "https://arxiv.org/html/2401.04088v1", "type": "primary" }, { "label": "DeepSeek-V2: activated-parameter comparison plot", "url": "https://arxiv.org/html/2405.04434v1", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "70B total is Meta's nominal specification. The nominal 70B active count follows Mistral's Mixtral comparison tables, corroborated approximately by DeepSeek-V2; it is not a Meta disclosure of exact per-token participation. This is the original pretrained base model with GQA and 4,096-token context under the Llama 2 Community License. Dedicated reasoning and native tool-call support remain undocumented.", "field_status": { "reasoning": "undisclosed", "tool_call": "undisclosed" }, "audited_at": "2026-10-27T13:56:00.652304Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/260.json" }, { "id": "meta/llama-2-70b-chat", "slug": "meta--llama-2-70b-chat", "name": "Llama 2 70B Chat", "developer": "Meta", "developer_id": "meta", "family": "llama-2", "description": "Meta’s nominal 70B Llama 2 text chat model, fine-tuned with SFT and RLHF; a dense Transformer using grouped-query attention and a 4,096-token context window.", "release_date": "2023-07-18", "date_precision": "day", "open_weights": true, "license": "Llama 2 Community License", "parameters_b": 70, "active_parameters_b": 70, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/llama-2/", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://ai.meta.com/blog/llama-2/", "type": "announcement" }, { "label": "Official model card and license", "url": "https://huggingface.co/meta-llama/Llama-2-70b-chat-hf", "type": "weights" }, { "label": "Llama 2 Community License Agreement", "url": "https://huggingface.co/meta-llama/Llama-2-70b-chat-hf/blob/main/LICENSE.txt", "type": "primary" }, { "label": "Original Llama 2 reference Transformer implementation", "url": "https://raw.githubusercontent.com/meta-llama/llama/llama_v2/llama/model.py", "type": "primary" }, { "label": "Original Llama 2 text and chat generation implementation", "url": "https://raw.githubusercontent.com/meta-llama/llama/llama_v2/llama/generation.py", "type": "primary" }, { "label": "Llama 2: Open Foundation and Fine-Tuned Chat Models", "url": "https://arxiv.org/html/2307.09288v2", "type": "primary" }, { "label": "Original Llama 2 release and inference README", "url": "https://raw.githubusercontent.com/meta-llama/llama/llama_v2/README.md", "type": "primary" }, { "label": "Mixtral of Experts — Table 2 active-parameter comparison", "url": "https://arxiv.org/html/2401.04088v1", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "70B is Meta’s nominal total. Nominal 70B active is inferred through Meta’s documented SFT/RLHF lineage from Mistral’s published active-count convention for base Llama 2 70B; it is not a separate Meta active-count disclosure. The original context is 4096 tokens. Reasoning and tool-call flags remain unknown: ordinary reasoning benchmarks and emergent tool-use text do not document native thinking or function-calling interfaces. Weights use the custom Llama 2 Community License and require license-conditioned access.", "field_status": { "reasoning": "undisclosed", "tool_call": "undisclosed" }, "audited_at": "2026-10-27T13:56:42.449770Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/261.json" }, { "id": "meta/llama-2-7b", "slug": "meta--llama-2-7b", "name": "Llama 2 7B", "developer": "Meta", "developer_id": "meta", "family": "llama-2", "description": "Llama 2 pretrained checkpoint.", "release_date": "2023-07-18", "date_precision": "day", "open_weights": true, "license": "Llama 2 Community License", "parameters_b": 7, "active_parameters_b": 7, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/llama-2/", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://ai.meta.com/blog/llama-2/", "type": "announcement" }, { "label": "Official model card and license", "url": "https://huggingface.co/meta-llama/Llama-2-7b-hf", "type": "weights" }, { "label": "Llama 2 technical report", "url": "https://arxiv.org/html/2307.09288v2", "type": "primary" }, { "label": "Llama 2 Community License", "url": "https://raw.githubusercontent.com/meta-llama/llama/7565eb6fee2175b2d4fe2cfb45067a61b35d7f5e/LICENSE", "type": "primary" }, { "label": "Original inference documentation", "url": "https://raw.githubusercontent.com/meta-llama/llama/7565eb6fee2175b2d4fe2cfb45067a61b35d7f5e/README.md", "type": "primary" }, { "label": "Reference Transformer implementation", "url": "https://raw.githubusercontent.com/meta-llama/llama/7565eb6fee2175b2d4fe2cfb45067a61b35d7f5e/llama/model.py", "type": "primary" }, { "label": "Reference generation implementation", "url": "https://raw.githubusercontent.com/meta-llama/llama/7565eb6fee2175b2d4fe2cfb45067a61b35d7f5e/llama/generation.py", "type": "primary" }, { "label": "Mistral original active-parameter comparison (Table 2)", "url": "https://arxiv.org/html/2401.04088v1", "type": "primary" }, { "label": "JetMoE original active-parameter comparison", "url": "https://research.myshell.ai/jetmoe", "type": "primary" }, { "label": "DeepSeekMoE activated-parameter plot (Figure 7)", "url": "https://aclanthology.org/2024.acl-long.70.pdf", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "7B is Meta's nominal total size. Nominal active 7B follows Mistral's and JetMoE's explicit comparison tables; DeepSeek plots approximately 6.7B. These research-team conventions do not establish an exact per-token count or a direct Meta active-count disclosure. This dense pretrained text model uses multi-head attention, 4096-token context and the custom Llama 2 Community License. Native thinking/tool-call support is not separately documented.", "field_status": { "reasoning": "undisclosed", "tool_call": "undisclosed" }, "audited_at": "2026-10-27T13:56:57.781886Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/262.json" }, { "id": "meta/llama-2-7b-chat", "slug": "meta--llama-2-7b-chat", "name": "Llama 2 7B Chat", "developer": "Meta", "developer_id": "meta", "family": "llama-2", "description": "Llama 2 chat-tuned checkpoint.", "release_date": "2023-07-18", "date_precision": "day", "open_weights": true, "license": "Llama 2 Community License", "parameters_b": 7, "active_parameters_b": 7, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/llama-2/", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://ai.meta.com/blog/llama-2/", "type": "announcement" }, { "label": "Official model card and license", "url": "https://huggingface.co/meta-llama/Llama-2-7b-chat-hf", "type": "weights" }, { "label": "Llama 2 technical report", "url": "https://arxiv.org/html/2307.09288v2", "type": "primary" }, { "label": "Original model implementation", "url": "https://raw.githubusercontent.com/meta-llama/llama/llama_v2/llama/model.py", "type": "primary" }, { "label": "Original generation interface", "url": "https://raw.githubusercontent.com/meta-llama/llama/llama_v2/llama/generation.py", "type": "primary" }, { "label": "Public repository metadata", "url": "https://huggingface.co/api/models/meta-llama/Llama-2-7b-chat-hf", "type": "primary" }, { "label": "Mistral active-parameter comparison (Table 2)", "url": "https://arxiv.org/html/2401.04088v1", "type": "primary" }, { "label": "Current Meta Llama 2 model card", "url": "https://raw.githubusercontent.com/meta-llama/llama-models/main/models/llama2/MODEL_CARD.md", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "Meta publishes a nominal 7B total for this Chat fine-tune. Nominal 7B active is inferred through its documented Llama 2 7B lineage using Mistral's Mixtral Table 2 active-count convention; Meta does not separately disclose a Chat active count. Exact per-token tensor participation is unspecified. Dense MHA architecture and a 4096-token trained context are documented. Dedicated reasoning and native tool-calling interfaces are not established.", "field_status": { "reasoning": "undisclosed", "tool_call": "undisclosed" }, "audited_at": "2026-10-27T14:02:29.180015Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/263.json" }, { "id": "mosaicml/mpt-7b-8k", "slug": "mosaicml--mpt-7b-8k", "name": "MPT-7B-8K", "developer": "MosaicML", "developer_id": "mosaicml", "family": "mpt", "description": null, "release_date": "2023-07-18", "date_precision": "day", "open_weights": true, "license": "CC-BY-SA-3.0", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.databricks.com/blog/long-context-mpt-7b-8k", "source_type": "primary", "sources": [ { "label": "Dated release naming this variant", "url": "https://www.databricks.com/blog/long-context-mpt-7b-8k", "type": "announcement" }, { "label": "Official context-length table", "url": "https://github.com/mosaicml/llm-foundry/blob/main/README.md", "type": "repository" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Context uses the published training window; ALiBi can extrapolate beyond it.", "availability": "public" }, { "id": "mosaicml/mpt-7b-8k-chat", "slug": "mosaicml--mpt-7b-8k-chat", "name": "MPT-7B-8K-Chat", "developer": "MosaicML", "developer_id": "mosaicml", "family": "mpt", "description": null, "release_date": "2023-07-18", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-SA-4.0", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.databricks.com/blog/long-context-mpt-7b-8k", "source_type": "primary", "sources": [ { "label": "Dated release naming this variant", "url": "https://www.databricks.com/blog/long-context-mpt-7b-8k", "type": "announcement" }, { "label": "Official context-length table", "url": "https://github.com/mosaicml/llm-foundry/blob/main/README.md", "type": "repository" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Context uses the published training window; ALiBi can extrapolate beyond it.", "availability": "public" }, { "id": "mosaicml/mpt-7b-8k-instruct", "slug": "mosaicml--mpt-7b-8k-instruct", "name": "MPT-7B-8K-Instruct", "developer": "MosaicML", "developer_id": "mosaicml", "family": "mpt", "description": null, "release_date": "2023-07-18", "date_precision": "day", "open_weights": true, "license": "CC-BY-SA-3.0", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.databricks.com/blog/long-context-mpt-7b-8k", "source_type": "primary", "sources": [ { "label": "Dated release naming this variant", "url": "https://www.databricks.com/blog/long-context-mpt-7b-8k", "type": "announcement" }, { "label": "Official context-length table", "url": "https://github.com/mosaicml/llm-foundry/blob/main/README.md", "type": "repository" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Context uses the published training window; ALiBi can extrapolate beyond it.", "availability": "public" }, { "id": "openlm/openllama-3b-v2", "slug": "openlm--openllama-3b-v2", "name": "OpenLLaMA 3B v2", "developer": "OpenLM Research", "developer_id": "openlm", "family": "openllama", "description": null, "release_date": "2023-07-15", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 3, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/openlm-research/open_llama/blob/6e7f73eab7e799e2464f38ed977e537bae02873e/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/openlm-research/open_llama/blob/6e7f73eab7e799e2464f38ed977e537bae02873e/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/openlm-research/open_llama_3b_v2/blob/bce5d60d3b0c68318862270ec4e794d83308d80a/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/openlm-research/open_llama_3b_v2/blob/bce5d60d3b0c68318862270ec4e794d83308d80a/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "Independently pretrained reproduction, not a fine-tune of Meta weights. The July 15 changelog calls it 3Bv3, but the linked released checkpoint is named open_llama_3b_v2." }, { "id": "anthropic/claude-2.0", "slug": "anthropic--claude-2.0", "name": "Claude 2", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude", "description": "Anthropic's July 2023 Claude text model, using a Transformer architecture and a 100K-token public context window. The claude-2.0 API model was retired on July 21, 2025.", "release_date": "2023-07-11", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 100000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.anthropic.com/news/claude-2", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://www.anthropic.com/news/claude-2", "type": "announcement" }, { "label": "Original Claude 2 model card", "url": "https://www-cdn.anthropic.com/bd2a28d2535bfb0494cc8e2a3bf135d2e7523226/Model-Card-Claude-2.pdf", "type": "model_card" }, { "label": "Claude 2 card with 2.1 appendix", "url": "https://www-cdn.anthropic.com/5c49cc247484cecf107c699baf29250302e5da70/ModelCardClaudev2_with_appendix_v1.pdf", "type": "model_card" }, { "label": "Later Claude 3 and 3.5 comparison card", "url": "https://www-cdn.anthropic.com/c6a80a657af445f40e31afac050f3bf76d3b1404.pdf", "type": "model_card" }, { "label": "Claude 2 weight-access policy", "url": "https://www-cdn.anthropic.com/1adf000c8f675958c2ee23805d91aaade1cd4613/responsible-scaling-policy.pdf", "type": "technical_report" }, { "label": "Anthropic model retirement history", "url": "https://platform.claude.com/docs/en/about-claude/model-deprecations", "type": "documentation" }, { "label": "Claude 2.1 API tool-use introduction", "url": "https://www.anthropic.com/news/claude-2-1", "type": "announcement" }, { "label": "Bedrock Agents Claude 2.0 parser", "url": "https://docs.aws.amazon.com/bedrock/latest/userguide/lambda-parser.html", "type": "documentation" }, { "label": "Claude 3.7 Sonnet System Card", "url": "https://www-cdn.anthropic.com/9ff93dfa8f445c932415d335c88852ef47f1201e.pdf", "type": "technical_report" }, { "label": "Many-shot Jailbreaking", "url": "https://www-cdn.anthropic.com/af5633c94ed2beb282f6a53c595eb437e8e7b630/Many_Shot_Jailbreaking2024_04_02_0936.pdf", "type": "technical_report" }, { "label": "Towards Monosemanticity: Decomposing Language Models With Dictionary Learning", "url": "https://transformer-circuits.pub/2023/monosemantic-features/index.html", "type": "technical_report" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "Nominal total and active parameter counts remain undisclosed; the 52B research-model count is not a Claude 2.0 specification. Transformer architecture is documented; dense versus MoE is not. Claude 2.0's publicly supported context was 100K, although the later model-card appendix discloses 200K training context; 2.1 enabled full 200K support. Claude 2.0 retired from the Anthropic API on July 21, 2025. Dedicated reasoning and native tool-call flags remain unresolved; Bedrock Agents' scaffolded function-call parser is distinct from Anthropic's later API tool-use feature.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T10:13:58.955285Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/087.json" }, { "id": "openlm/openllama-7b-v2", "slug": "openlm--openllama-7b-v2", "name": "OpenLLaMA 7B v2", "developer": "OpenLM Research", "developer_id": "openlm", "family": "openllama", "description": null, "release_date": "2023-07-07", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/openlm-research/open_llama/blob/6e7f73eab7e799e2464f38ed977e537bae02873e/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/openlm-research/open_llama/blob/6e7f73eab7e799e2464f38ed977e537bae02873e/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/openlm-research/open_llama_7b_v2/blob/e5961def23172a2384543940e773ab676033c963/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/openlm-research/open_llama_7b_v2/blob/e5961def23172a2384543940e773ab676033c963/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "Independently pretrained reproduction, not a fine-tune of Meta weights. " }, { "id": "internlm/internlm-7b", "slug": "internlm--internlm-7b", "name": "InternLM-7B", "developer": "Shanghai AI Laboratory", "developer_id": "internlm", "family": "internlm", "description": null, "release_date": "2023-07-06", "date_precision": "day", "open_weights": true, "license": "InternLM Model License", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/InternLM/InternLM/blob/e6118174422224334bb14c5523caea6391fcaaf8/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/InternLM/InternLM/blob/e6118174422224334bb14c5523caea6391fcaaf8/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/internlm/internlm-7b/blob/96e127d08d851a88cac736a9b091dd953ae1b873/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/internlm/internlm-7b/blob/96e127d08d851a88cac736a9b091dd953ae1b873/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Publisher release table names the exact checkpoint and date. " }, { "id": "internlm/internlm-chat-7b", "slug": "internlm--internlm-chat-7b", "name": "InternLM-chat-7B", "developer": "Shanghai AI Laboratory", "developer_id": "internlm", "family": "internlm", "description": null, "release_date": "2023-07-06", "date_precision": "day", "open_weights": true, "license": "InternLM Model License", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/InternLM/InternLM/blob/e6118174422224334bb14c5523caea6391fcaaf8/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/InternLM/InternLM/blob/e6118174422224334bb14c5523caea6391fcaaf8/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/internlm/internlm-chat-7b/blob/4e2d2e185058a4cbaa2377ba1bb993eb599ce492/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/internlm/internlm-chat-7b/blob/4e2d2e185058a4cbaa2377ba1bb993eb599ce492/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Publisher release table names the exact checkpoint and date. " }, { "id": "internlm/internlm-chat-7b-8k", "slug": "internlm--internlm-chat-7b-8k", "name": "InternLM-chat-7B-8k", "developer": "Shanghai AI Laboratory", "developer_id": "internlm", "family": "internlm", "description": null, "release_date": "2023-07-06", "date_precision": "day", "open_weights": true, "license": "InternLM Model License", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/InternLM/InternLM/blob/e6118174422224334bb14c5523caea6391fcaaf8/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/InternLM/InternLM/blob/e6118174422224334bb14c5523caea6391fcaaf8/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/internlm/internlm-chat-7b-8k", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Publisher release table names the exact checkpoint and date. The original checkpoint endpoint now returns 401; release and nominal size are retained in the dated official repository." }, { "id": "salesforce/codegen2.5-7b-instruct", "slug": "salesforce--codegen2.5-7b-instruct", "name": "CodeGen2.5 7B Instruct", "developer": "Salesforce", "developer_id": "salesforce", "family": "codegen", "description": null, "release_date": "2023-07-06", "date_precision": "day", "open_weights": true, "license": "CodeGen Research License", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.salesforce.com/blog/codegen25/", "source_type": "primary", "sources": [ { "label": "Dated publisher release and variant descriptions", "url": "https://www.salesforce.com/blog/codegen25/", "type": "announcement" }, { "label": "Publisher variant and license list", "url": "https://github.com/salesforce/CodeGen/blob/ec21c93bda19c087f419f552c196afa9a36ee890/codegen25/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Salesforce/codegen25-7b-instruct/blob/5c3821aa6090f1cada356c5c60fe96cf68b33945/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/Salesforce/codegen25-7b-instruct/blob/5c3821aa6090f1cada356c5c60fe96cf68b33945/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "The instruction checkpoint has a research-only license; the multilingual and Python checkpoints use Apache-2.0." }, { "id": "salesforce/codegen2.5-7b-mono", "slug": "salesforce--codegen2.5-7b-mono", "name": "CodeGen2.5 7B Mono", "developer": "Salesforce", "developer_id": "salesforce", "family": "codegen", "description": null, "release_date": "2023-07-06", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.salesforce.com/blog/codegen25/", "source_type": "primary", "sources": [ { "label": "Dated publisher release and variant descriptions", "url": "https://www.salesforce.com/blog/codegen25/", "type": "announcement" }, { "label": "Publisher variant and license list", "url": "https://github.com/salesforce/CodeGen/blob/ec21c93bda19c087f419f552c196afa9a36ee890/codegen25/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Salesforce/codegen25-7b-mono/blob/1c90c89671f4d700deabd60210d2cb7aa1df10d8/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/Salesforce/codegen25-7b-mono/blob/1c90c89671f4d700deabd60210d2cb7aa1df10d8/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "The instruction checkpoint has a research-only license; the multilingual and Python checkpoints use Apache-2.0." }, { "id": "salesforce/codegen2.5-7b-multi", "slug": "salesforce--codegen2.5-7b-multi", "name": "CodeGen2.5 7B Multi", "developer": "Salesforce", "developer_id": "salesforce", "family": "codegen", "description": null, "release_date": "2023-07-06", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.salesforce.com/blog/codegen25/", "source_type": "primary", "sources": [ { "label": "Dated publisher release and variant descriptions", "url": "https://www.salesforce.com/blog/codegen25/", "type": "announcement" }, { "label": "Publisher variant and license list", "url": "https://github.com/salesforce/CodeGen/blob/ec21c93bda19c087f419f552c196afa9a36ee890/codegen25/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Salesforce/codegen25-7b-multi/blob/3cfb2194ec55e4a229f2d2184623b747fa24ab94/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/Salesforce/codegen25-7b-multi/blob/3cfb2194ec55e4a229f2d2184623b747fa24ab94/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "The instruction checkpoint has a research-only license; the multilingual and Python checkpoints use Apache-2.0." }, { "id": "baichuan/baichuan-13b-base", "slug": "baichuan--baichuan-13b-base", "name": "Baichuan-13B-Base", "developer": "Baichuan AI", "developer_id": "baichuan", "family": "baichuan", "description": null, "release_date": "2023-07", "date_precision": "month", "open_weights": true, "license": "Baichuan-13B Model License", "parameters_b": 13.26490112, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/baichuan-inc/Baichuan-13B/blob/66b540d2c58aed798540b21386f4aedda0991ff2/README_EN.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/baichuan-inc/Baichuan-13B/blob/66b540d2c58aed798540b21386f4aedda0991ff2/README_EN.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/baichuan-inc/Baichuan-13B-Base/blob/0ef0739c7bdd34df954003ef76d80f3dabca2ff9/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/baichuan-inc/Baichuan-13B-Base/blob/0ef0739c7bdd34df954003ef76d80f3dabca2ff9/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Historical publisher release documentation verifies availability in this month; exact public launch day remains unresolved. Parameter count is from the publisher architecture table." }, { "id": "baichuan/baichuan-13b-chat", "slug": "baichuan--baichuan-13b-chat", "name": "Baichuan-13B-Chat", "developer": "Baichuan AI", "developer_id": "baichuan", "family": "baichuan", "description": null, "release_date": "2023-07", "date_precision": "month", "open_weights": true, "license": "Baichuan-13B Model License", "parameters_b": 13.26490112, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/baichuan-inc/Baichuan-13B/blob/66b540d2c58aed798540b21386f4aedda0991ff2/README_EN.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/baichuan-inc/Baichuan-13B/blob/66b540d2c58aed798540b21386f4aedda0991ff2/README_EN.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/baichuan-inc/Baichuan-13B-Chat/blob/e3e1498bb6b7fb0bcb5a65679c09b862d7c29301/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/baichuan-inc/Baichuan-13B-Chat/blob/e3e1498bb6b7fb0bcb5a65679c09b862d7c29301/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Historical publisher release documentation verifies availability in this month; exact public launch day remains unresolved. Parameter count is from the publisher architecture table." }, { "id": "zhipuai/chatglm2-6b", "slug": "zhipuai--chatglm2-6b", "name": "ChatGLM2-6B", "developer": "Zhipu AI / Tsinghua University", "developer_id": "zhipuai", "family": "chatglm", "description": null, "release_date": "2023-06-25", "date_precision": "day", "open_weights": true, "license": "ChatGLM Model License", "parameters_b": 6, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/THUDM/ChatGLM-6B/blob/401bf3a8a7dd8a26fba189551dccfc61a7079b4e/README_en.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/THUDM/ChatGLM-6B/blob/401bf3a8a7dd8a26fba189551dccfc61a7079b4e/README_en.md", "type": "repository" }, { "label": "Publisher release chronology", "url": "https://github.com/THUDM/ChatGLM2-6B/blob/cb8e8b43c0951b32614f25c03e1ab593a0603a1c/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/THUDM/chatglm2-6b/blob/d2e2d91789248536a747d9ce60642a336444186c/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/THUDM/chatglm2-6b/blob/d2e2d91789248536a747d9ce60642a336444186c/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Nominal parameter count from the publisher." }, { "id": "lmsys/vicuna-13b-v1.3", "slug": "lmsys--vicuna-13b-v1.3", "name": "Vicuna 13B v1.3", "developer": "LMSYS", "developer_id": "lmsys", "family": "vicuna", "description": null, "release_date": "2023-06-22", "date_precision": "day", "open_weights": true, "license": "LLaMA research license", "parameters_b": 13, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/lmsys/vicuna-13b-v1.3", "source_type": "primary", "sources": [ { "label": "LMSYS release history", "url": "https://github.com/lm-sys/FastChat/blob/587d5cfa1609a43d192cedb8441cac3c17db105d/docs/vicuna_weights_version.md", "type": "documentation" }, { "label": "Official model card", "url": "https://huggingface.co/lmsys/vicuna-13b-v1.3/blob/6566e9cb1787585d1147dcf4f9bc48f29e1328d2/README.md", "type": "model_card" }, { "label": "Official configuration", "url": "https://huggingface.co/lmsys/vicuna-13b-v1.3/blob/6566e9cb1787585d1147dcf4f9bc48f29e1328d2/config.json", "type": "config" }, { "label": "LMSYS institutional provenance", "url": "https://www.lmsys.org/about/", "type": "primary" }, { "label": "LMSYS v1.3 release announcement", "url": "https://www.lmsys.org/blog/2023-06-22-leaderboard/", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:09:10.442Z", "last_seen": "2026-10-30T07:09:10.442Z", "upstream_present": false, "availability": "public", "notes": "Institutional LMSYS release. Nominal publisher size; independently trained revision verified by its release notes and model card." }, { "id": "lmsys/vicuna-33b-v1.3", "slug": "lmsys--vicuna-33b-v1.3", "name": "Vicuna 33B v1.3", "developer": "LMSYS", "developer_id": "lmsys", "family": "vicuna", "description": null, "release_date": "2023-06-22", "date_precision": "day", "open_weights": true, "license": "LLaMA research license", "parameters_b": 33, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/lmsys/vicuna-33b-v1.3", "source_type": "primary", "sources": [ { "label": "LMSYS release history", "url": "https://github.com/lm-sys/FastChat/blob/587d5cfa1609a43d192cedb8441cac3c17db105d/docs/vicuna_weights_version.md", "type": "documentation" }, { "label": "Official model card", "url": "https://huggingface.co/lmsys/vicuna-33b-v1.3/blob/ef8d6becf883fb3ce52e3706885f761819477ab4/README.md", "type": "model_card" }, { "label": "Official configuration", "url": "https://huggingface.co/lmsys/vicuna-33b-v1.3/blob/ef8d6becf883fb3ce52e3706885f761819477ab4/config.json", "type": "config" }, { "label": "LMSYS institutional provenance", "url": "https://www.lmsys.org/about/", "type": "primary" }, { "label": "LMSYS v1.3 release announcement", "url": "https://www.lmsys.org/blog/2023-06-22-leaderboard/", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:09:10.442Z", "last_seen": "2026-10-30T07:09:10.442Z", "upstream_present": false, "availability": "public", "notes": "Institutional LMSYS release. Nominal publisher size; independently trained revision verified by its release notes and model card." }, { "id": "lmsys/vicuna-7b-v1.3", "slug": "lmsys--vicuna-7b-v1.3", "name": "Vicuna 7B v1.3", "developer": "LMSYS", "developer_id": "lmsys", "family": "vicuna", "description": null, "release_date": "2023-06-22", "date_precision": "day", "open_weights": true, "license": "LLaMA research license", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/lmsys/vicuna-7b-v1.3", "source_type": "primary", "sources": [ { "label": "LMSYS release history", "url": "https://github.com/lm-sys/FastChat/blob/587d5cfa1609a43d192cedb8441cac3c17db105d/docs/vicuna_weights_version.md", "type": "documentation" }, { "label": "Official model card", "url": "https://huggingface.co/lmsys/vicuna-7b-v1.3/blob/236eeeab96f0dc2e463f2bebb7bb49809279c6d6/README.md", "type": "model_card" }, { "label": "Official configuration", "url": "https://huggingface.co/lmsys/vicuna-7b-v1.3/blob/236eeeab96f0dc2e463f2bebb7bb49809279c6d6/config.json", "type": "config" }, { "label": "LMSYS institutional provenance", "url": "https://www.lmsys.org/about/", "type": "primary" }, { "label": "LMSYS v1.3 release announcement", "url": "https://www.lmsys.org/blog/2023-06-22-leaderboard/", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:09:10.442Z", "last_seen": "2026-10-30T07:09:10.442Z", "upstream_present": false, "availability": "public", "notes": "Institutional LMSYS release. Nominal publisher size; independently trained revision verified by its release notes and model card." }, { "id": "mosaicml/mpt-30b", "slug": "mosaicml--mpt-30b", "name": "MPT-30B", "developer": "MosaicML", "developer_id": "mosaicml", "family": "mpt", "description": null, "release_date": "2023-06-22", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 30, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.databricks.com/blog/mpt-30b", "source_type": "primary", "sources": [ { "label": "Dated release naming this variant", "url": "https://www.databricks.com/blog/mpt-30b", "type": "announcement" }, { "label": "Official context-length table", "url": "https://github.com/mosaicml/llm-foundry/blob/main/README.md", "type": "repository" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Context uses the published training window; ALiBi can extrapolate beyond it.", "availability": "public" }, { "id": "mosaicml/mpt-30b-chat", "slug": "mosaicml--mpt-30b-chat", "name": "MPT-30B-Chat", "developer": "MosaicML", "developer_id": "mosaicml", "family": "mpt", "description": null, "release_date": "2023-06-22", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-SA-4.0", "parameters_b": 30, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.databricks.com/blog/mpt-30b", "source_type": "primary", "sources": [ { "label": "Dated release naming this variant", "url": "https://www.databricks.com/blog/mpt-30b", "type": "announcement" }, { "label": "Official context-length table", "url": "https://github.com/mosaicml/llm-foundry/blob/main/README.md", "type": "repository" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Context uses the published training window; ALiBi can extrapolate beyond it.", "availability": "public" }, { "id": "mosaicml/mpt-30b-instruct", "slug": "mosaicml--mpt-30b-instruct", "name": "MPT-30B-Instruct", "developer": "MosaicML", "developer_id": "mosaicml", "family": "mpt", "description": null, "release_date": "2023-06-22", "date_precision": "day", "open_weights": true, "license": "CC-BY-SA-3.0", "parameters_b": 30, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.databricks.com/blog/mpt-30b", "source_type": "primary", "sources": [ { "label": "Dated release naming this variant", "url": "https://www.databricks.com/blog/mpt-30b", "type": "announcement" }, { "label": "Official context-length table", "url": "https://github.com/mosaicml/llm-foundry/blob/main/README.md", "type": "repository" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Context uses the published training window; ALiBi can extrapolate beyond it.", "availability": "public" }, { "id": "openlm/openllama-13b", "slug": "openlm--openllama-13b", "name": "OpenLLaMA 13B", "developer": "OpenLM Research", "developer_id": "openlm", "family": "openllama", "description": null, "release_date": "2023-06-15", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 13, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/openlm-research/open_llama/blob/6e7f73eab7e799e2464f38ed977e537bae02873e/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/openlm-research/open_llama/blob/6e7f73eab7e799e2464f38ed977e537bae02873e/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/openlm-research/open_llama_13b/blob/b6d7fde8392250730d24cc2fcfa3b7e5f9a03ce8/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/openlm-research/open_llama_13b/blob/b6d7fde8392250730d24cc2fcfa3b7e5f9a03ce8/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "Independently pretrained reproduction, not a fine-tune of Meta weights. Final 1T-token checkpoint; earlier partial-training previews are separate releases." }, { "id": "openai/gpt-3.5-turbo-0613", "slug": "openai--gpt-3.5-turbo-0613", "name": "GPT-3.5 Turbo 0613", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-3.5", "description": null, "release_date": "2023-06-13", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://openai.com/index/function-calling-and-other-api-updates/", "source_type": "primary", "sources": [ { "label": "Dated API model update", "url": "https://openai.com/index/function-calling-and-other-api-updates/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "Pinned API checkpoint, distinct from the original March release. Now retired; availability records the original public API release." }, { "id": "openai/gpt-3.5-turbo-16k-0613", "slug": "openai--gpt-3.5-turbo-16k-0613", "name": "GPT-3.5 Turbo 16K 0613", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-3.5", "description": null, "release_date": "2023-06-13", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 16384, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://openai.com/index/function-calling-and-other-api-updates/", "source_type": "primary", "sources": [ { "label": "Dated API model update", "url": "https://openai.com/index/function-calling-and-other-api-updates/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "Pinned API checkpoint, distinct from the original March release. Now retired; availability records the original public API release." }, { "id": "openai/gpt-4-0613", "slug": "openai--gpt-4-0613", "name": "GPT-4 0613", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-4", "description": null, "release_date": "2023-06-13", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://openai.com/index/function-calling-and-other-api-updates/", "source_type": "primary", "sources": [ { "label": "Dated API model update", "url": "https://openai.com/index/function-calling-and-other-api-updates/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "Pinned API checkpoint, distinct from the original March release. Now retired; availability records the original public API release." }, { "id": "openai/gpt-4-32k-0613", "slug": "openai--gpt-4-32k-0613", "name": "GPT-4 32K 0613", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-4", "description": null, "release_date": "2023-06-13", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://openai.com/index/function-calling-and-other-api-updates/", "source_type": "primary", "sources": [ { "label": "Dated API model update", "url": "https://openai.com/index/function-calling-and-other-api-updates/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "Pinned API checkpoint, distinct from the original March release. Now retired; availability records the original public API release." }, { "id": "openlm/openllama-3b", "slug": "openlm--openllama-3b", "name": "OpenLLaMA 3B", "developer": "OpenLM Research", "developer_id": "openlm", "family": "openllama", "description": null, "release_date": "2023-06-07", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 3, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/openlm-research/open_llama/blob/6e7f73eab7e799e2464f38ed977e537bae02873e/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/openlm-research/open_llama/blob/6e7f73eab7e799e2464f38ed977e537bae02873e/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/openlm-research/open_llama_3b/blob/141067009124b9c0aea62c76b3eb952174864057/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/openlm-research/open_llama_3b/blob/141067009124b9c0aea62c76b3eb952174864057/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "Independently pretrained reproduction, not a fine-tune of Meta weights. Final 1T-token checkpoint; earlier partial-training previews are separate releases." }, { "id": "openlm/openllama-7b", "slug": "openlm--openllama-7b", "name": "OpenLLaMA 7B", "developer": "OpenLM Research", "developer_id": "openlm", "family": "openllama", "description": null, "release_date": "2023-06-07", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/openlm-research/open_llama/blob/6e7f73eab7e799e2464f38ed977e537bae02873e/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/openlm-research/open_llama/blob/6e7f73eab7e799e2464f38ed977e537bae02873e/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/openlm-research/open_llama_7b/blob/6fb184ff23774c25bf84b3628e49c8b78372c7be/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/openlm-research/open_llama_7b/blob/6fb184ff23774c25bf84b3628e49c8b78372c7be/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "Independently pretrained reproduction, not a fine-tune of Meta weights. Final 1T-token checkpoint; earlier partial-training previews are separate releases." }, { "id": "together/redpajama-incite-7b-base-v1", "slug": "together--redpajama-incite-7b-base-v1", "name": "RedPajama-INCITE 7B Base v1", "developer": "Together AI", "developer_id": "together", "family": "redpajama", "description": null, "release_date": "2023-06-06", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 6.9, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.together.ai/blog/redpajama-7b", "source_type": "primary", "sources": [ { "label": "Dated v1 release", "url": "https://www.together.ai/blog/redpajama-7b", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/togethercomputer/RedPajama-INCITE-7B-Base/blob/78f7e482443971f4873ba3239f0ac810a367833b/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/togethercomputer/RedPajama-INCITE-7B-Base/blob/78f7e482443971f4873ba3239f0ac810a367833b/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "v1 release is separate from the May 2023 partial-training checkpoint at the same repository.", "availability": "public" }, { "id": "together/redpajama-incite-7b-chat-v1", "slug": "together--redpajama-incite-7b-chat-v1", "name": "RedPajama-INCITE 7B Chat v1", "developer": "Together AI", "developer_id": "together", "family": "redpajama", "description": null, "release_date": "2023-06-06", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 6.9, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.together.ai/blog/redpajama-7b", "source_type": "primary", "sources": [ { "label": "Dated v1 release", "url": "https://www.together.ai/blog/redpajama-7b", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/togethercomputer/RedPajama-INCITE-7B-Chat/blob/47b94a739e2f3164b438501c8684acc5d5acc146/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/togethercomputer/RedPajama-INCITE-7B-Chat/blob/47b94a739e2f3164b438501c8684acc5d5acc146/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "v1 release is separate from the May 2023 partial-training checkpoint at the same repository.", "availability": "public" }, { "id": "together/redpajama-incite-7b-instruct-v1", "slug": "together--redpajama-incite-7b-instruct-v1", "name": "RedPajama-INCITE 7B Instruct v1", "developer": "Together AI", "developer_id": "together", "family": "redpajama", "description": null, "release_date": "2023-06-06", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 6.9, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.together.ai/blog/redpajama-7b", "source_type": "primary", "sources": [ { "label": "Dated v1 release", "url": "https://www.together.ai/blog/redpajama-7b", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/togethercomputer/RedPajama-INCITE-7B-Instruct/blob/7f36397b9985a3f981cdb618f8fec1c565ca5927/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/togethercomputer/RedPajama-INCITE-7B-Instruct/blob/7f36397b9985a3f981cdb618f8fec1c565ca5927/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "v1 release is separate from the May 2023 partial-training checkpoint at the same repository.", "availability": "public" }, { "id": "tiiuae/falcon-rw-1b", "slug": "tiiuae--falcon-rw-1b", "name": "Falcon-RW-1B", "developer": "Technology Innovation Institute", "developer_id": "tiiuae", "family": "falcon-rw", "description": null, "release_date": "2023-06-01", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 1.3, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://arxiv.org/abs/2306.01116v1", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/tiiuae/falcon-rw-1b/blob/e4b9872bb803165eb22f0a867d4e6a64d34fce19/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/tiiuae/falcon-rw-1b/blob/e4b9872bb803165eb22f0a867d4e6a64d34fce19/config.json", "type": "config" }, { "label": "Dated developer paper announcing public model release", "url": "https://arxiv.org/abs/2306.01116v1", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:36:54.148Z", "last_seen": "2026-10-29T10:36:54.148Z", "upstream_present": false, "availability": "public", "notes": "The original RefinedWeb paper explicitly announces public release of the 1.3B and 7.5B language models. Parameter counts use that paper rather than the rounded repository names. Trained on 350B RefinedWeb tokens." }, { "id": "tiiuae/falcon-rw-7b", "slug": "tiiuae--falcon-rw-7b", "name": "Falcon-RW-7B", "developer": "Technology Innovation Institute", "developer_id": "tiiuae", "family": "falcon-rw", "description": null, "release_date": "2023-06-01", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 7.5, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://arxiv.org/abs/2306.01116v1", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/tiiuae/falcon-rw-7b/blob/7cc3a731c0e3a7d3ece6500f3a018aa9432156ea/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/tiiuae/falcon-rw-7b/blob/7cc3a731c0e3a7d3ece6500f3a018aa9432156ea/config.json", "type": "config" }, { "label": "Dated developer paper announcing public model release", "url": "https://arxiv.org/abs/2306.01116v1", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:36:54.148Z", "last_seen": "2026-10-29T10:36:54.148Z", "upstream_present": false, "availability": "public", "notes": "The original RefinedWeb paper explicitly announces public release of the 1.3B and 7.5B language models. Parameter counts use that paper rather than the rounded repository names. Trained on 350B RefinedWeb tokens." }, { "id": "baichuan/baichuan-7b", "slug": "baichuan--baichuan-7b", "name": "Baichuan-7B", "developer": "Baichuan AI", "developer_id": "baichuan", "family": "baichuan", "description": null, "release_date": "2023-06", "date_precision": "month", "open_weights": true, "license": "Baichuan-7B Model License", "parameters_b": 7.000559616, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/baichuan-inc/Baichuan-7B/blob/baeb0617dcc0c12e17262c0c7f22182fcbda2cc1/README_EN.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/baichuan-inc/Baichuan-7B/blob/baeb0617dcc0c12e17262c0c7f22182fcbda2cc1/README_EN.md", "type": "repository" }, { "label": "Publisher release chronology", "url": "https://github.com/baichuan-inc/Baichuan-13B/blob/66b540d2c58aed798540b21386f4aedda0991ff2/README_EN.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/baichuan-inc/Baichuan-7B/blob/5d86e56a58fe4a5b3292cd9bb7468afef6f93eab/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/baichuan-inc/Baichuan-7B/blob/5d86e56a58fe4a5b3292cd9bb7468afef6f93eab/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Historical publisher release documentation verifies availability in this month; exact public launch day remains unresolved. Parameter count is from the publisher architecture table." }, { "id": "tiiuae/falcon-40b", "slug": "tiiuae--falcon-40b", "name": "Falcon 40B", "developer": "Technology Innovation Institute", "developer_id": "tiiuae", "family": "falcon", "description": null, "release_date": "2023-05-25", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 40, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.tii.ae/news/uaes-technology-innovation-institute-launches-open-source-falcon-40b-large-language-model", "source_type": "primary", "sources": [ { "label": "Dated developer release", "url": "https://www.tii.ae/news/uaes-technology-innovation-institute-launches-open-source-falcon-40b-large-language-model", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/tiiuae/falcon-40b/blob/05ab2ee8d6b593bdbab17d728de5c028a7a94d83/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/tiiuae/falcon-40b/blob/05ab2ee8d6b593bdbab17d728de5c028a7a94d83/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Nominal count from the publisher; current Apache license followed the initial launch.", "availability": "public" }, { "id": "zhipuai/visualglm-6b", "slug": "zhipuai--visualglm-6b", "name": "VisualGLM-6B", "developer": "Zhipu AI / Tsinghua University", "developer_id": "zhipuai", "family": "visualglm", "description": null, "release_date": "2023-05-17", "date_precision": "day", "open_weights": true, "license": "VisualGLM Model License", "parameters_b": 7.8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/THUDM/ChatGLM-6B/blob/401bf3a8a7dd8a26fba189551dccfc61a7079b4e/README_en.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/THUDM/ChatGLM-6B/blob/401bf3a8a7dd8a26fba189551dccfc61a7079b4e/README_en.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/THUDM/visualglm-6b/blob/a05d5f967eb67d2503fb8ccab6b5e01713305184/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/THUDM/visualglm-6b/blob/a05d5f967eb67d2503fb8ccab6b5e01713305184/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Total parameters include the vision encoder." }, { "id": "google/chat-bison-001", "slug": "google--chat-bison-001", "name": "PaLM 2 Chat Bison 001", "developer": "Google", "developer_id": "google", "family": "palm-2", "description": null, "release_date": "2023-05-10", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://docs.cloud.google.com/vertex-ai/docs/core-release-notes", "source_type": "primary", "sources": [ { "label": "Publisher historical API release notes", "url": "https://docs.cloud.google.com/vertex-ai/docs/core-release-notes", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:26:10.259Z", "last_seen": "2026-10-29T10:26:10.259Z", "upstream_present": false, "availability": "public", "notes": "Exact endpoint chat-bison@001 is named in the May 10 preview release. The model was later retired." }, { "id": "google/code-bison-001", "slug": "google--code-bison-001", "name": "Codey code-bison 001", "developer": "Google", "developer_id": "google", "family": "codey", "description": null, "release_date": "2023-05-10", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://cloud.google.com/blog/products/ai-machine-learning/google-cloud-launches-new-ai-models-opens-generative-ai-studio", "source_type": "primary", "sources": [ { "label": "Dated initial trusted-tester availability", "url": "https://cloud.google.com/blog/products/ai-machine-learning/google-cloud-launches-new-ai-models-opens-generative-ai-studio", "type": "announcement" }, { "label": "Publisher historical API release notes", "url": "https://docs.cloud.google.com/vertex-ai/docs/core-release-notes", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:26:10.259Z", "last_seen": "2026-10-29T10:26:10.259Z", "upstream_present": false, "availability": "public", "notes": "Codey code generation, chat, and completion were available to trusted testers May 10. Public preview without allowlisting began June 7; the June 29 general-availability note identifies these endpoint names. Context and original parameter count remain unpublished." }, { "id": "google/code-gecko-001", "slug": "google--code-gecko-001", "name": "Codey code-gecko 001", "developer": "Google", "developer_id": "google", "family": "codey", "description": null, "release_date": "2023-05-10", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://cloud.google.com/blog/products/ai-machine-learning/google-cloud-launches-new-ai-models-opens-generative-ai-studio", "source_type": "primary", "sources": [ { "label": "Dated initial trusted-tester availability", "url": "https://cloud.google.com/blog/products/ai-machine-learning/google-cloud-launches-new-ai-models-opens-generative-ai-studio", "type": "announcement" }, { "label": "Publisher historical API release notes", "url": "https://docs.cloud.google.com/vertex-ai/docs/core-release-notes", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:26:10.259Z", "last_seen": "2026-10-29T10:26:10.259Z", "upstream_present": false, "availability": "public", "notes": "Codey code generation, chat, and completion were available to trusted testers May 10. Public preview without allowlisting began June 7; the June 29 general-availability note identifies these endpoint names. Context and original parameter count remain unpublished." }, { "id": "google/codechat-bison-001", "slug": "google--codechat-bison-001", "name": "Codey codechat-bison 001", "developer": "Google", "developer_id": "google", "family": "codey", "description": null, "release_date": "2023-05-10", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://cloud.google.com/blog/products/ai-machine-learning/google-cloud-launches-new-ai-models-opens-generative-ai-studio", "source_type": "primary", "sources": [ { "label": "Dated initial trusted-tester availability", "url": "https://cloud.google.com/blog/products/ai-machine-learning/google-cloud-launches-new-ai-models-opens-generative-ai-studio", "type": "announcement" }, { "label": "Publisher historical API release notes", "url": "https://docs.cloud.google.com/vertex-ai/docs/core-release-notes", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:26:10.259Z", "last_seen": "2026-10-29T10:26:10.259Z", "upstream_present": false, "availability": "public", "notes": "Codey code generation, chat, and completion were available to trusted testers May 10. Public preview without allowlisting began June 7; the June 29 general-availability note identifies these endpoint names. Context and original parameter count remain unpublished." }, { "id": "google/text-bison-001", "slug": "google--text-bison-001", "name": "PaLM 2 Text Bison 001", "developer": "Google", "developer_id": "google", "family": "palm-2", "description": null, "release_date": "2023-05-10", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://docs.cloud.google.com/vertex-ai/docs/core-release-notes", "source_type": "primary", "sources": [ { "label": "Publisher historical API release notes", "url": "https://docs.cloud.google.com/vertex-ai/docs/core-release-notes", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:26:10.259Z", "last_seen": "2026-10-29T10:26:10.259Z", "upstream_present": false, "availability": "public", "notes": "Exact endpoint text-bison@001 is named in the May 10 preview release. The model was later retired." }, { "id": "mosaicml/mpt-7b", "slug": "mosaicml--mpt-7b", "name": "MPT-7B", "developer": "MosaicML", "developer_id": "mosaicml", "family": "mpt", "description": null, "release_date": "2023-05-05", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 6.7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.databricks.com/blog/mpt-7b", "source_type": "primary", "sources": [ { "label": "Dated release naming this variant", "url": "https://www.databricks.com/blog/mpt-7b", "type": "announcement" }, { "label": "Official context-length table", "url": "https://github.com/mosaicml/llm-foundry/blob/main/README.md", "type": "repository" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Context uses the published training window; ALiBi can extrapolate beyond it.", "availability": "public" }, { "id": "mosaicml/mpt-7b-chat", "slug": "mosaicml--mpt-7b-chat", "name": "MPT-7B-Chat", "developer": "MosaicML", "developer_id": "mosaicml", "family": "mpt", "description": null, "release_date": "2023-05-05", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-SA-4.0", "parameters_b": 6.7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.databricks.com/blog/mpt-7b", "source_type": "primary", "sources": [ { "label": "Dated release naming this variant", "url": "https://www.databricks.com/blog/mpt-7b", "type": "announcement" }, { "label": "Official context-length table", "url": "https://github.com/mosaicml/llm-foundry/blob/main/README.md", "type": "repository" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Context uses the published training window; ALiBi can extrapolate beyond it.", "availability": "public" }, { "id": "mosaicml/mpt-7b-instruct", "slug": "mosaicml--mpt-7b-instruct", "name": "MPT-7B-Instruct", "developer": "MosaicML", "developer_id": "mosaicml", "family": "mpt", "description": null, "release_date": "2023-05-05", "date_precision": "day", "open_weights": true, "license": "CC-BY-SA-3.0", "parameters_b": 6.7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.databricks.com/blog/mpt-7b", "source_type": "primary", "sources": [ { "label": "Dated release naming this variant", "url": "https://www.databricks.com/blog/mpt-7b", "type": "announcement" }, { "label": "Official context-length table", "url": "https://github.com/mosaicml/llm-foundry/blob/main/README.md", "type": "repository" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Context uses the published training window; ALiBi can extrapolate beyond it.", "availability": "public" }, { "id": "mosaicml/mpt-7b-storywriter", "slug": "mosaicml--mpt-7b-storywriter", "name": "MPT-7B-Storywriter", "developer": "MosaicML", "developer_id": "mosaicml", "family": "mpt", "description": null, "release_date": "2023-05-05", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 6.7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 65536, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.databricks.com/blog/mpt-7b", "source_type": "primary", "sources": [ { "label": "Dated release naming this variant", "url": "https://www.databricks.com/blog/mpt-7b", "type": "announcement" }, { "label": "Official context-length table", "url": "https://github.com/mosaicml/llm-foundry/blob/main/README.md", "type": "repository" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Context uses the published training window; ALiBi can extrapolate beyond it.", "availability": "public" }, { "id": "together/redpajama-incite-base-3b-v1", "slug": "together--redpajama-incite-base-3b-v1", "name": "RedPajama-INCITE Base 3B v1", "developer": "Together AI", "developer_id": "together", "family": "redpajama", "description": null, "release_date": "2023-05-05", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 2.8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.together.ai/blog/redpajama-models-v1", "source_type": "primary", "sources": [ { "label": "Dated release listing the three 3B variants", "url": "https://www.together.ai/blog/redpajama-models-v1", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/togethercomputer/RedPajama-INCITE-Base-3B-v1/blob/094fbdd0c911feb485ce55de1952ab2e75277e1e/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/togethercomputer/RedPajama-INCITE-Base-3B-v1/blob/094fbdd0c911feb485ce55de1952ab2e75277e1e/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Model card describes the nominal 2.8B pretrained architecture.", "availability": "public" }, { "id": "together/redpajama-incite-chat-3b-v1", "slug": "together--redpajama-incite-chat-3b-v1", "name": "RedPajama-INCITE Chat 3B v1", "developer": "Together AI", "developer_id": "together", "family": "redpajama", "description": null, "release_date": "2023-05-05", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 2.8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.together.ai/blog/redpajama-models-v1", "source_type": "primary", "sources": [ { "label": "Dated release listing the three 3B variants", "url": "https://www.together.ai/blog/redpajama-models-v1", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/togethercomputer/RedPajama-INCITE-Chat-3B-v1/blob/f0e0995eba801096ed04cb87931d96a8316871af/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/togethercomputer/RedPajama-INCITE-Chat-3B-v1/blob/f0e0995eba801096ed04cb87931d96a8316871af/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Model card describes the nominal 2.8B pretrained architecture.", "availability": "public" }, { "id": "together/redpajama-incite-instruct-3b-v1", "slug": "together--redpajama-incite-instruct-3b-v1", "name": "RedPajama-INCITE Instruct 3B v1", "developer": "Together AI", "developer_id": "together", "family": "redpajama", "description": null, "release_date": "2023-05-05", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 2.8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.together.ai/blog/redpajama-models-v1", "source_type": "primary", "sources": [ { "label": "Dated release listing the three 3B variants", "url": "https://www.together.ai/blog/redpajama-models-v1", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/togethercomputer/RedPajama-INCITE-Instruct-3B-v1/blob/0c66778ee09a036886741707733620b91057909a/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/togethercomputer/RedPajama-INCITE-Instruct-3B-v1/blob/0c66778ee09a036886741707733620b91057909a/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Model card describes the nominal 2.8B pretrained architecture.", "availability": "public" }, { "id": "bigcode/starcoder", "slug": "bigcode--starcoder", "name": "StarCoder", "developer": "BigCode", "developer_id": "bigcode", "family": "starcoder", "description": null, "release_date": "2023-05-04", "date_precision": "day", "open_weights": true, "license": "BigCode OpenRAIL-M", "parameters_b": 15.5, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/blog/starcoder", "source_type": "primary", "sources": [ { "label": "Dated release by the model team", "url": "https://huggingface.co/blog/starcoder", "type": "announcement" }, { "label": "Model-team technical report", "url": "https://arxiv.org/abs/2305.06161", "type": "paper" }, { "label": "Official model card", "url": "https://huggingface.co/bigcode/starcoder/blob/827cd7721231872c152e3faef5cae1b22964ed58/README.md", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "Publisher nominal parameter count. StarCoder is the original team’s Python continuation of StarCoderBase." }, { "id": "bigcode/starcoderbase", "slug": "bigcode--starcoderbase", "name": "StarCoderBase", "developer": "BigCode", "developer_id": "bigcode", "family": "starcoder", "description": null, "release_date": "2023-05-04", "date_precision": "day", "open_weights": true, "license": "BigCode OpenRAIL-M", "parameters_b": 15.5, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/blog/starcoder", "source_type": "primary", "sources": [ { "label": "Dated release by the model team", "url": "https://huggingface.co/blog/starcoder", "type": "announcement" }, { "label": "Model-team technical report", "url": "https://arxiv.org/abs/2305.06161", "type": "paper" }, { "label": "Official model card", "url": "https://huggingface.co/bigcode/starcoderbase/blob/88ec5781ad071a9d9e925cd28f327dea22eb5188/README.md", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "Publisher nominal parameter count. StarCoder is the original team’s Python continuation of StarCoderBase." }, { "id": "inflection/pi-initial", "slug": "inflection--pi-initial", "name": "Pi (initial model)", "developer": "Inflection AI", "developer_id": "inflection", "family": "inflection", "description": null, "release_date": "2023-05-02", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "unknown", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://inflection.ai/blog/pi", "source_type": "primary", "sources": [ { "label": "Publisher Pi launch", "url": "https://inflection.ai/blog/pi", "type": "announcement" }, { "label": "Original publisher Inflection-1 announcement", "url": "https://web.archive.org/web/20230622173045/https://inflection.ai/inflection-1", "type": "announcement" }, { "label": "Inflection-1 technical memo, June 22 2023", "url": "https://web.archive.org/web/20230622173107/https://inflection.ai/assets/Inflection-1.pdf", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:39:55.111Z", "last_seen": "2026-10-29T10:39:55.111Z", "upstream_present": false, "availability": "public", "notes": "The initial Pi product model was publicly accessible May 2, 2023. The June 22 Inflection-1 memo explicitly distinguishes the foundation model from Pi, which undergoes additional alignment and safety adaptation. Exact checkpoint parameters, architecture subtype and context window are not disclosed in the reviewed primary material. No identical checkpoint assumption is made between the base model and Pi." }, { "id": "replit/replit-code-v1-3b", "slug": "replit--replit-code-v1-3b", "name": "Replit Code v1 3B", "developer": "Replit", "developer_id": "replit", "family": "replit-code", "description": null, "release_date": "2023-05-02", "date_precision": "day", "open_weights": true, "license": "CC-BY-SA-4.0", "parameters_b": 3, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/replit/ReplitLM/blob/f95ecb74ae02536d7f638c86502f9d17d632f64f/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/replit/ReplitLM/blob/f95ecb74ae02536d7f638c86502f9d17d632f64f/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/replit/replit-code-v1-3b/blob/cc0a4f17a8d72b71d62ea53cb0e23e4dac352067/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/replit/replit-code-v1-3b/blob/cc0a4f17a8d72b71d62ea53cb0e23e4dac352067/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "The publisher release list explicitly dates this checkpoint May 2." }, { "id": "salesforce/codegen2-16b", "slug": "salesforce--codegen2-16b", "name": "CodeGen2 16B", "developer": "Salesforce", "developer_id": "salesforce", "family": "codegen", "description": null, "release_date": "2023-05", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 16, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/salesforce/CodeGen/blob/c969a96b75ad2490bb5fcff9e95662131d8d4a8a/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/salesforce/CodeGen/blob/c969a96b75ad2490bb5fcff9e95662131d8d4a8a/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Salesforce/codegen2-16B/blob/14bcc89999feb32d9ee3c51388aa4d8ac134124d/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/Salesforce/codegen2-16B/blob/14bcc89999feb32d9ee3c51388aa4d8ac134124d/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "The publisher dates CodeGen2 to May 2023; exact launch day remains unresolved. The original Hugging Face endpoint now redirects to a publisher repository with the _P suffix." }, { "id": "salesforce/codegen2-1b", "slug": "salesforce--codegen2-1b", "name": "CodeGen2 1B", "developer": "Salesforce", "developer_id": "salesforce", "family": "codegen", "description": null, "release_date": "2023-05", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 1, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/salesforce/CodeGen/blob/c969a96b75ad2490bb5fcff9e95662131d8d4a8a/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/salesforce/CodeGen/blob/c969a96b75ad2490bb5fcff9e95662131d8d4a8a/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Salesforce/codegen2-1B/blob/47f05cd3c1b357745fbead74202fc1efe87b6d25/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/Salesforce/codegen2-1B/blob/47f05cd3c1b357745fbead74202fc1efe87b6d25/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "The publisher dates CodeGen2 to May 2023; exact launch day remains unresolved. The original Hugging Face endpoint now redirects to a publisher repository with the _P suffix." }, { "id": "salesforce/codegen2-3.7b", "slug": "salesforce--codegen2-3.7b", "name": "CodeGen2 3.7B", "developer": "Salesforce", "developer_id": "salesforce", "family": "codegen", "description": null, "release_date": "2023-05", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 3.7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/salesforce/CodeGen/blob/c969a96b75ad2490bb5fcff9e95662131d8d4a8a/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/salesforce/CodeGen/blob/c969a96b75ad2490bb5fcff9e95662131d8d4a8a/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Salesforce/codegen2-3_7B/blob/2999b8fc740371db5923ef6f8ec1b74cfb9b4900/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/Salesforce/codegen2-3_7B/blob/2999b8fc740371db5923ef6f8ec1b74cfb9b4900/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "The publisher dates CodeGen2 to May 2023; exact launch day remains unresolved. The original Hugging Face endpoint now redirects to a publisher repository with the _P suffix." }, { "id": "salesforce/codegen2-7b", "slug": "salesforce--codegen2-7b", "name": "CodeGen2 7B", "developer": "Salesforce", "developer_id": "salesforce", "family": "codegen", "description": null, "release_date": "2023-05", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/salesforce/CodeGen/blob/c969a96b75ad2490bb5fcff9e95662131d8d4a8a/README.md", "source_type": "primary", "sources": [ { "label": "Publisher release chronology", "url": "https://github.com/salesforce/CodeGen/blob/c969a96b75ad2490bb5fcff9e95662131d8d4a8a/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Salesforce/codegen2-7B/blob/b3d2f281d860402d5e7acc3c2c466098ccc0b482/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/Salesforce/codegen2-7B/blob/b3d2f281d860402d5e7acc3c2c466098ccc0b482/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:17:01.563Z", "last_seen": "2026-10-29T10:17:01.563Z", "upstream_present": false, "availability": "public", "notes": "The publisher dates CodeGen2 to May 2023; exact launch day remains unresolved. The original Hugging Face endpoint now redirects to a publisher repository with the _P suffix." }, { "id": "stabilityai/stablelm-base-alpha-3b", "slug": "stabilityai--stablelm-base-alpha-3b", "name": "StableLM Base Alpha 3B", "developer": "Stability AI", "developer_id": "stabilityai", "family": "stablelm", "description": null, "release_date": "2023-04-19", "date_precision": "day", "open_weights": true, "license": "CC-BY-SA-4.0", "parameters_b": 3.638525952, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://stability.ai/news-updates/stability-ai-launches-the-first-of-its-stablelm-suite-of-language-models", "source_type": "primary", "sources": [ { "label": "Dated release of base and tuned models", "url": "https://stability.ai/news-updates/stability-ai-launches-the-first-of-its-stablelm-suite-of-language-models", "type": "announcement" }, { "label": "Official parameter table", "url": "https://github.com/Stability-AI/StableLM/blob/main/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/stabilityai/stablelm-base-alpha-3b/blob/819657db42861818efac86e8ca60f673fa912ef8/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/stabilityai/stablelm-base-alpha-3b/blob/819657db42861818efac86e8ca60f673fa912ef8/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Launch post dates availability April 19; repository changelog says April 20.", "availability": "public" }, { "id": "stabilityai/stablelm-base-alpha-7b", "slug": "stabilityai--stablelm-base-alpha-7b", "name": "StableLM Base Alpha 7B", "developer": "Stability AI", "developer_id": "stabilityai", "family": "stablelm", "description": null, "release_date": "2023-04-19", "date_precision": "day", "open_weights": true, "license": "CC-BY-SA-4.0", "parameters_b": 7.86935808, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://stability.ai/news-updates/stability-ai-launches-the-first-of-its-stablelm-suite-of-language-models", "source_type": "primary", "sources": [ { "label": "Dated release of base and tuned models", "url": "https://stability.ai/news-updates/stability-ai-launches-the-first-of-its-stablelm-suite-of-language-models", "type": "announcement" }, { "label": "Official parameter table", "url": "https://github.com/Stability-AI/StableLM/blob/main/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/stabilityai/stablelm-base-alpha-7b/blob/1d43a3b40d7d708d8084d07dfae0c4e9fd95c04f/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/stabilityai/stablelm-base-alpha-7b/blob/1d43a3b40d7d708d8084d07dfae0c4e9fd95c04f/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Launch post dates availability April 19; repository changelog says April 20.", "availability": "public" }, { "id": "stabilityai/stablelm-tuned-alpha-3b", "slug": "stabilityai--stablelm-tuned-alpha-3b", "name": "StableLM Tuned Alpha 3B", "developer": "Stability AI", "developer_id": "stabilityai", "family": "stablelm", "description": null, "release_date": "2023-04-19", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-SA-4.0", "parameters_b": 3.638525952, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://stability.ai/news-updates/stability-ai-launches-the-first-of-its-stablelm-suite-of-language-models", "source_type": "primary", "sources": [ { "label": "Dated release of base and tuned models", "url": "https://stability.ai/news-updates/stability-ai-launches-the-first-of-its-stablelm-suite-of-language-models", "type": "announcement" }, { "label": "Official parameter table", "url": "https://github.com/Stability-AI/StableLM/blob/main/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/stabilityai/stablelm-tuned-alpha-3b/blob/d1c03d2114451d562416b9efe4281d319ceff99e/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/stabilityai/stablelm-tuned-alpha-3b/blob/d1c03d2114451d562416b9efe4281d319ceff99e/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Launch post dates availability April 19; repository changelog says April 20.", "availability": "public" }, { "id": "stabilityai/stablelm-tuned-alpha-7b", "slug": "stabilityai--stablelm-tuned-alpha-7b", "name": "StableLM Tuned Alpha 7B", "developer": "Stability AI", "developer_id": "stabilityai", "family": "stablelm", "description": null, "release_date": "2023-04-19", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-SA-4.0", "parameters_b": 7.86935808, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 4096, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://stability.ai/news-updates/stability-ai-launches-the-first-of-its-stablelm-suite-of-language-models", "source_type": "primary", "sources": [ { "label": "Dated release of base and tuned models", "url": "https://stability.ai/news-updates/stability-ai-launches-the-first-of-its-stablelm-suite-of-language-models", "type": "announcement" }, { "label": "Official parameter table", "url": "https://github.com/Stability-AI/StableLM/blob/main/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/stabilityai/stablelm-tuned-alpha-7b/blob/25071b093c15c0d1cb2b2876c6deb621b764fcf5/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/stabilityai/stablelm-tuned-alpha-7b/blob/25071b093c15c0d1cb2b2876c6deb621b764fcf5/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Launch post dates availability April 19; repository changelog says April 20.", "availability": "public" }, { "id": "databricks/dolly-v2-12b", "slug": "databricks--dolly-v2-12b", "name": "Dolly v2 12B", "developer": "Databricks", "developer_id": "databricks", "family": "dolly", "description": null, "release_date": "2023-04-12", "date_precision": "day", "open_weights": true, "license": null, "parameters_b": 12, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.databricks.com/blog/2023/04/12/dolly-first-open-commercially-viable-instruction-tuned-llm", "source_type": "primary", "sources": [ { "label": "Dated exact-model release", "url": "https://www.databricks.com/blog/2023/04/12/dolly-first-open-commercially-viable-instruction-tuned-llm", "type": "announcement" }, { "label": "Release-day official model documentation", "url": "https://github.com/databrickslabs/dolly/blob/6724e9e207382d7d3ab540e3dc09cd13e40cc23a/README.md", "type": "repository" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Nominal size from the developer. Current original checkpoint endpoint returns 401; supported context and exact weight-license text remain unverified.", "availability": "public" }, { "id": "lmsys/vicuna-13b-v1.1", "slug": "lmsys--vicuna-13b-v1.1", "name": "Vicuna 13B v1.1", "developer": "LMSYS", "developer_id": "lmsys", "family": "vicuna", "description": null, "release_date": "2023-04-12", "date_precision": "day", "open_weights": true, "license": "LLaMA research license", "parameters_b": 13, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/lmsys/vicuna-13b-v1.1", "source_type": "primary", "sources": [ { "label": "LMSYS release history", "url": "https://github.com/lm-sys/FastChat/blob/587d5cfa1609a43d192cedb8441cac3c17db105d/docs/vicuna_weights_version.md", "type": "documentation" }, { "label": "Official model card", "url": "https://huggingface.co/lmsys/vicuna-13b-v1.1/blob/1acf26f93742fafe91562253ec0e5d94e40a8bea/README.md", "type": "model_card" }, { "label": "Official configuration", "url": "https://huggingface.co/lmsys/vicuna-13b-v1.1/blob/1acf26f93742fafe91562253ec0e5d94e40a8bea/config.json", "type": "config" }, { "label": "LMSYS institutional provenance", "url": "https://www.lmsys.org/about/", "type": "primary" } ], "aliases": [ "lmsys/vicuna-13b-delta-v1.1" ], "first_seen": "2026-10-30T07:09:10.442Z", "last_seen": "2026-10-30T07:09:10.442Z", "upstream_present": false, "availability": "public", "notes": "Institutional LMSYS release. Nominal publisher size; independently trained revision verified by its release notes and model card. Delta weights require the original licensed LLaMA checkpoint." }, { "id": "lmsys/vicuna-7b-v1.1", "slug": "lmsys--vicuna-7b-v1.1", "name": "Vicuna 7B v1.1", "developer": "LMSYS", "developer_id": "lmsys", "family": "vicuna", "description": null, "release_date": "2023-04-12", "date_precision": "day", "open_weights": true, "license": "LLaMA research license", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/lmsys/vicuna-7b-v1.1", "source_type": "primary", "sources": [ { "label": "LMSYS release history", "url": "https://github.com/lm-sys/FastChat/blob/587d5cfa1609a43d192cedb8441cac3c17db105d/docs/vicuna_weights_version.md", "type": "documentation" }, { "label": "Official model card", "url": "https://huggingface.co/lmsys/vicuna-7b-v1.1/blob/694d39416b1744d939760f5257c4bb08b2fae400/README.md", "type": "model_card" }, { "label": "Official configuration", "url": "https://huggingface.co/lmsys/vicuna-7b-v1.1/blob/694d39416b1744d939760f5257c4bb08b2fae400/config.json", "type": "config" }, { "label": "LMSYS institutional provenance", "url": "https://www.lmsys.org/about/", "type": "primary" } ], "aliases": [ "lmsys/vicuna-7b-delta-v1.1" ], "first_seen": "2026-10-30T07:09:10.442Z", "last_seen": "2026-10-30T07:09:10.442Z", "upstream_present": false, "availability": "public", "notes": "Institutional LMSYS release. Nominal publisher size; independently trained revision verified by its release notes and model card. Delta weights require the original licensed LLaMA checkpoint." }, { "id": "lmsys/vicuna-7b-v0", "slug": "lmsys--vicuna-7b-v0", "name": "Vicuna 7B v0", "developer": "LMSYS", "developer_id": "lmsys", "family": "vicuna", "description": null, "release_date": "2023-04-06", "date_precision": "day", "open_weights": true, "license": "LLaMA research license", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/lmsys/vicuna-7b-delta-v0", "source_type": "primary", "sources": [ { "label": "LMSYS release history", "url": "https://github.com/lm-sys/FastChat/blob/587d5cfa1609a43d192cedb8441cac3c17db105d/docs/vicuna_weights_version.md", "type": "documentation" }, { "label": "Official model card", "url": "https://huggingface.co/lmsys/vicuna-7b-delta-v0/blob/fa5967d19dcc5d89c0a39d68766281ae6021e304/README.md", "type": "model_card" }, { "label": "Official configuration", "url": "https://huggingface.co/lmsys/vicuna-7b-delta-v0/blob/fa5967d19dcc5d89c0a39d68766281ae6021e304/config.json", "type": "config" }, { "label": "LMSYS institutional provenance", "url": "https://www.lmsys.org/about/", "type": "primary" }, { "label": "Original 7B weight release", "url": "https://github.com/lm-sys/FastChat/commit/e2de15f23ea4ef159669422043516a708dad28e3", "type": "primary" } ], "aliases": [ "lmsys/vicuna-7b-delta-v0" ], "first_seen": "2026-10-30T07:09:10.442Z", "last_seen": "2026-10-30T07:09:10.442Z", "upstream_present": false, "availability": "public", "notes": "Institutional LMSYS release. Nominal publisher size; independently trained revision verified by its release notes and model card. The original April 6 release commit replaces “Coming soon” with the 7B delta download. It supersedes the later family table’s March 30 shorthand. Delta weights require the original licensed LLaMA checkpoint." }, { "id": "eleutherai/pythia-1.4b", "slug": "eleutherai--pythia-1.4b", "name": "Pythia 1.4B", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2023-04-03", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 1.414647808, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "source_type": "primary", "sources": [ { "label": "Dated retrained-suite release", "url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/EleutherAI/pythia-1.4b/blob/fedc38a16eea3bd36a96b906d78d11d2ce18ed79/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/EleutherAI/pythia-1.4b/blob/fedc38a16eea3bd36a96b906d78d11d2ce18ed79/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "April 2023 retrained suite; distinct from retained v0 checkpoints. Total includes embeddings.", "availability": "public" }, { "id": "eleutherai/pythia-1.4b-deduped", "slug": "eleutherai--pythia-1.4b-deduped", "name": "Pythia 1.4B Deduped", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2023-04-03", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 1.414647808, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "source_type": "primary", "sources": [ { "label": "Dated retrained-suite release", "url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/EleutherAI/pythia-1.4b-deduped/blob/554d9c1bae3877f740aece41feb90f912cf9fedc/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/EleutherAI/pythia-1.4b-deduped/blob/554d9c1bae3877f740aece41feb90f912cf9fedc/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "April 2023 retrained suite; distinct from retained v0 checkpoints. Total includes embeddings.", "availability": "public" }, { "id": "eleutherai/pythia-12b", "slug": "eleutherai--pythia-12b", "name": "Pythia 12B", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2023-04-03", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 11.84607232, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "source_type": "primary", "sources": [ { "label": "Dated retrained-suite release", "url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/EleutherAI/pythia-12b/blob/bb1e3e710cdf6b524461d543cfb5ba773f0a81b6/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/EleutherAI/pythia-12b/blob/bb1e3e710cdf6b524461d543cfb5ba773f0a81b6/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "April 2023 retrained suite; distinct from retained v0 checkpoints. Total includes embeddings.", "availability": "public" }, { "id": "eleutherai/pythia-12b-deduped", "slug": "eleutherai--pythia-12b-deduped", "name": "Pythia 12B Deduped", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2023-04-03", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 11.84607232, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "source_type": "primary", "sources": [ { "label": "Dated retrained-suite release", "url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/EleutherAI/pythia-12b-deduped/blob/39c1bd94f9dbe4ebd1d191f364cb33a2e5c47707/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/EleutherAI/pythia-12b-deduped/blob/39c1bd94f9dbe4ebd1d191f364cb33a2e5c47707/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "April 2023 retrained suite; distinct from retained v0 checkpoints. Total includes embeddings.", "availability": "public" }, { "id": "eleutherai/pythia-160m", "slug": "eleutherai--pythia-160m", "name": "Pythia 160M", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2023-04-03", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.162322944, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "source_type": "primary", "sources": [ { "label": "Dated retrained-suite release", "url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/EleutherAI/pythia-160m/blob/50f5173d932e8e61f858120bcb800b97af589f46/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/EleutherAI/pythia-160m/blob/50f5173d932e8e61f858120bcb800b97af589f46/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "April 2023 retrained suite; distinct from retained v0 checkpoints. Total includes embeddings.", "availability": "public" }, { "id": "eleutherai/pythia-160m-deduped", "slug": "eleutherai--pythia-160m-deduped", "name": "Pythia 160M Deduped", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2023-04-03", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.162322944, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "source_type": "primary", "sources": [ { "label": "Dated retrained-suite release", "url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/EleutherAI/pythia-160m-deduped/blob/582159a2dfe3e712a8d47ae83dec95ae3bde8e7e/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/EleutherAI/pythia-160m-deduped/blob/582159a2dfe3e712a8d47ae83dec95ae3bde8e7e/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "April 2023 retrained suite; distinct from retained v0 checkpoints. Total includes embeddings.", "availability": "public" }, { "id": "eleutherai/pythia-1b", "slug": "eleutherai--pythia-1b", "name": "Pythia 1B", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2023-04-03", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 1.011781632, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "source_type": "primary", "sources": [ { "label": "Dated retrained-suite release", "url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/EleutherAI/pythia-1b/blob/f73d7dcc545c8bd326d8559c8ef84ffe92fea6b2/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/EleutherAI/pythia-1b/blob/f73d7dcc545c8bd326d8559c8ef84ffe92fea6b2/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "April 2023 retrained suite; distinct from retained v0 checkpoints. Total includes embeddings.", "availability": "public" }, { "id": "eleutherai/pythia-1b-deduped", "slug": "eleutherai--pythia-1b-deduped", "name": "Pythia 1B Deduped", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2023-04-03", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 1.011781632, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "source_type": "primary", "sources": [ { "label": "Dated retrained-suite release", "url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/EleutherAI/pythia-1b-deduped/blob/7199d8fc61a6d565cd1f3c62bf11525b563e13b2/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/EleutherAI/pythia-1b-deduped/blob/7199d8fc61a6d565cd1f3c62bf11525b563e13b2/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "April 2023 retrained suite; distinct from retained v0 checkpoints. Total includes embeddings.", "availability": "public" }, { "id": "eleutherai/pythia-2.8b", "slug": "eleutherai--pythia-2.8b", "name": "Pythia 2.8B", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2023-04-03", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 2.77520896, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "source_type": "primary", "sources": [ { "label": "Dated retrained-suite release", "url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/EleutherAI/pythia-2.8b/blob/2a259cdd96a4beb1cdf467512e3904197345f6a9/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/EleutherAI/pythia-2.8b/blob/2a259cdd96a4beb1cdf467512e3904197345f6a9/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "April 2023 retrained suite; distinct from retained v0 checkpoints. Total includes embeddings.", "availability": "public" }, { "id": "eleutherai/pythia-2.8b-deduped", "slug": "eleutherai--pythia-2.8b-deduped", "name": "Pythia 2.8B Deduped", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2023-04-03", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 2.77520896, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "source_type": "primary", "sources": [ { "label": "Dated retrained-suite release", "url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/EleutherAI/pythia-2.8b-deduped/blob/7d977fed8c4ce9649816af8cd5fe36a639cbe5b2/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/EleutherAI/pythia-2.8b-deduped/blob/7d977fed8c4ce9649816af8cd5fe36a639cbe5b2/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "April 2023 retrained suite; distinct from retained v0 checkpoints. Total includes embeddings.", "availability": "public" }, { "id": "eleutherai/pythia-410m", "slug": "eleutherai--pythia-410m", "name": "Pythia 410M", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2023-04-03", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.405334016, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "source_type": "primary", "sources": [ { "label": "Dated retrained-suite release", "url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/EleutherAI/pythia-410m/blob/9879c9b5f8bea9051dcb0e68dff21493d67e9d4f/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/EleutherAI/pythia-410m/blob/9879c9b5f8bea9051dcb0e68dff21493d67e9d4f/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "April 2023 retrained suite; distinct from retained v0 checkpoints. Total includes embeddings.", "availability": "public" }, { "id": "eleutherai/pythia-410m-deduped", "slug": "eleutherai--pythia-410m-deduped", "name": "Pythia 410M Deduped", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2023-04-03", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.405334016, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "source_type": "primary", "sources": [ { "label": "Dated retrained-suite release", "url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/EleutherAI/pythia-410m-deduped/blob/c4fc8d586d62df497f1f9b69d66d3ca419992d3e/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/EleutherAI/pythia-410m-deduped/blob/c4fc8d586d62df497f1f9b69d66d3ca419992d3e/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "April 2023 retrained suite; distinct from retained v0 checkpoints. Total includes embeddings.", "availability": "public" }, { "id": "eleutherai/pythia-6.9b", "slug": "eleutherai--pythia-6.9b", "name": "Pythia 6.9B", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2023-04-03", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 6.857302016, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "source_type": "primary", "sources": [ { "label": "Dated retrained-suite release", "url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/EleutherAI/pythia-6.9b/blob/c0e3eee36dc47af0c49f361c74cfe459c09f7f23/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/EleutherAI/pythia-6.9b/blob/c0e3eee36dc47af0c49f361c74cfe459c09f7f23/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "April 2023 retrained suite; distinct from retained v0 checkpoints. Total includes embeddings.", "availability": "public" }, { "id": "eleutherai/pythia-6.9b-deduped", "slug": "eleutherai--pythia-6.9b-deduped", "name": "Pythia 6.9B Deduped", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2023-04-03", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 6.857302016, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "source_type": "primary", "sources": [ { "label": "Dated retrained-suite release", "url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/EleutherAI/pythia-6.9b-deduped/blob/372b1c08d9b5b0fc18ce86bbf294930e26e66ed5/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/EleutherAI/pythia-6.9b-deduped/blob/372b1c08d9b5b0fc18ce86bbf294930e26e66ed5/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "April 2023 retrained suite; distinct from retained v0 checkpoints. Total includes embeddings.", "availability": "public" }, { "id": "eleutherai/pythia-70m", "slug": "eleutherai--pythia-70m", "name": "Pythia 70M", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2023-04-03", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.070426624, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "source_type": "primary", "sources": [ { "label": "Dated retrained-suite release", "url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/EleutherAI/pythia-70m/blob/a39f36b100fe8a5377810d56c3f4789b9c53ac42/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/EleutherAI/pythia-70m/blob/a39f36b100fe8a5377810d56c3f4789b9c53ac42/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "April 2023 retrained suite; distinct from retained v0 checkpoints. Total includes embeddings.", "availability": "public" }, { "id": "eleutherai/pythia-70m-deduped", "slug": "eleutherai--pythia-70m-deduped", "name": "Pythia 70M Deduped", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2023-04-03", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.070426624, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "source_type": "primary", "sources": [ { "label": "Dated retrained-suite release", "url": "https://github.com/EleutherAI/pythia/blob/main/README.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/EleutherAI/pythia-70m-deduped/blob/e93a9faa9c77e5d09219f6c868bfc7a1bd65593c/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/EleutherAI/pythia-70m-deduped/blob/e93a9faa9c77e5d09219f6c868bfc7a1bd65593c/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "April 2023 retrained suite; distinct from retained v0 checkpoints. Total includes embeddings.", "availability": "public" }, { "id": "databricks/dolly-v2-3b", "slug": "databricks--dolly-v2-3b", "name": "Dolly v2 3B", "developer": "Databricks", "developer_id": "databricks", "family": "dolly", "description": null, "release_date": "2023-04", "date_precision": "month", "open_weights": true, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/databrickslabs/dolly/blob/905e58a1ef9f5f22ba39285855f39b20b41003fe/examples/pipeline.py", "source_type": "primary", "sources": [ { "label": "April 15 publisher notebook lists this exact pretrained checkpoint", "url": "https://github.com/databrickslabs/dolly/blob/905e58a1ef9f5f22ba39285855f39b20b41003fe/examples/pipeline.py", "type": "repository" }, { "label": "Publisher family release", "url": "https://www.databricks.com/blog/2023/04/12/dolly-first-open-commercially-viable-instruction-tuned-llm", "type": "announcement" }, { "label": "April 18 Databricks usage example loads this checkpoint", "url": "https://www.databricks.com/blog/2023/04/18/introducing-mlflow-23-enhanced-native-llm-support-and-new-features.html", "type": "blog" } ], "aliases": [], "first_seen": "2026-10-29T11:26:55.844Z", "last_seen": "2026-10-29T11:26:55.844Z", "upstream_present": false, "availability": "public", "notes": "Separate Databricks-trained Dolly v2 size. The April 15 publisher notebook lists the exact pretrained model, while the April 12 announcement establishes the 12B launch; exact first availability of this sibling is not established, so month precision is retained. Current original card/config endpoints return 401. Parameter count, weight license and context remain unverified rather than being inferred from the size suffix or another model’s configuration." }, { "id": "databricks/dolly-v2-7b", "slug": "databricks--dolly-v2-7b", "name": "Dolly v2 7B", "developer": "Databricks", "developer_id": "databricks", "family": "dolly", "description": null, "release_date": "2023-04", "date_precision": "month", "open_weights": true, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/databrickslabs/dolly/blob/905e58a1ef9f5f22ba39285855f39b20b41003fe/examples/pipeline.py", "source_type": "primary", "sources": [ { "label": "April 15 publisher notebook lists this exact pretrained checkpoint", "url": "https://github.com/databrickslabs/dolly/blob/905e58a1ef9f5f22ba39285855f39b20b41003fe/examples/pipeline.py", "type": "repository" }, { "label": "Publisher family release", "url": "https://www.databricks.com/blog/2023/04/12/dolly-first-open-commercially-viable-instruction-tuned-llm", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T11:26:55.844Z", "last_seen": "2026-10-29T11:26:55.844Z", "upstream_present": false, "availability": "public", "notes": "Separate Databricks-trained Dolly v2 size. The April 15 publisher notebook lists the exact pretrained model, while the April 12 announcement establishes the 12B launch; exact first availability of this sibling is not established, so month precision is retained. Current original card/config endpoints return 401. Parameter count, weight license and context remain unverified rather than being inferred from the size suffix or another model’s configuration." }, { "id": "lmsys/vicuna-13b-v0", "slug": "lmsys--vicuna-13b-v0", "name": "Vicuna 13B v0", "developer": "LMSYS", "developer_id": "lmsys", "family": "vicuna", "description": null, "release_date": "2023-03-30", "date_precision": "day", "open_weights": true, "license": "LLaMA research license", "parameters_b": 13, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/lmsys/vicuna-13b-delta-v0", "source_type": "primary", "sources": [ { "label": "LMSYS release history", "url": "https://github.com/lm-sys/FastChat/blob/587d5cfa1609a43d192cedb8441cac3c17db105d/docs/vicuna_weights_version.md", "type": "documentation" }, { "label": "Official model card", "url": "https://huggingface.co/lmsys/vicuna-13b-delta-v0/blob/3082b9f4b63712f003c3d751a4a737606c2e68f7/README.md", "type": "model_card" }, { "label": "Official configuration", "url": "https://huggingface.co/lmsys/vicuna-13b-delta-v0/blob/3082b9f4b63712f003c3d751a4a737606c2e68f7/config.json", "type": "config" }, { "label": "LMSYS institutional provenance", "url": "https://www.lmsys.org/about/", "type": "primary" }, { "label": "Original demo announcement", "url": "https://www.lmsys.org/blog/2023-03-30-vicuna/", "type": "primary" }, { "label": "Original 13B delta release", "url": "https://github.com/lm-sys/FastChat/commit/1edb5c0823f5b4027d1cdeac8dc9f67a6b53ba47", "type": "primary" } ], "aliases": [ "lmsys/vicuna-13b-delta-v0" ], "first_seen": "2026-10-30T07:09:10.442Z", "last_seen": "2026-10-30T07:09:10.442Z", "upstream_present": false, "availability": "public", "notes": "Institutional LMSYS release. Nominal publisher size; independently trained revision verified by its release notes and model card. First external access was the March 30 demo; the original April 3 commit releases the reconstructable weight delta. Delta weights require the original licensed LLaMA checkpoint." }, { "id": "cerebras/cerebras-gpt-1.3b", "slug": "cerebras--cerebras-gpt-1.3b", "name": "Cerebras-GPT 1.3B", "developer": "Cerebras", "developer_id": "cerebras", "family": "cerebras-gpt", "description": null, "release_date": "2023-03-28", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 1.3, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.cerebras.ai/press-release/cerebras-systems-releases-seven-new-gpt-models-trained-on-cs-2-wafer-scale-systems", "source_type": "primary", "sources": [ { "label": "Dated seven-model release", "url": "https://www.cerebras.ai/press-release/cerebras-systems-releases-seven-new-gpt-models-trained-on-cs-2-wafer-scale-systems", "type": "announcement" }, { "label": "Official model architecture and size table", "url": "https://huggingface.co/cerebras/Cerebras-GPT-111M/blob/f91e002fe344cd90163ec56422de495ffd6576ba/README.md", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Publisher reports this nominal parameter count.", "availability": "public" }, { "id": "cerebras/cerebras-gpt-111m", "slug": "cerebras--cerebras-gpt-111m", "name": "Cerebras-GPT 111M", "developer": "Cerebras", "developer_id": "cerebras", "family": "cerebras-gpt", "description": null, "release_date": "2023-03-28", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.111, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.cerebras.ai/press-release/cerebras-systems-releases-seven-new-gpt-models-trained-on-cs-2-wafer-scale-systems", "source_type": "primary", "sources": [ { "label": "Dated seven-model release", "url": "https://www.cerebras.ai/press-release/cerebras-systems-releases-seven-new-gpt-models-trained-on-cs-2-wafer-scale-systems", "type": "announcement" }, { "label": "Official model architecture and size table", "url": "https://huggingface.co/cerebras/Cerebras-GPT-111M/blob/f91e002fe344cd90163ec56422de495ffd6576ba/README.md", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Publisher reports this nominal parameter count.", "availability": "public" }, { "id": "cerebras/cerebras-gpt-13b", "slug": "cerebras--cerebras-gpt-13b", "name": "Cerebras-GPT 13B", "developer": "Cerebras", "developer_id": "cerebras", "family": "cerebras-gpt", "description": null, "release_date": "2023-03-28", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 13, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.cerebras.ai/press-release/cerebras-systems-releases-seven-new-gpt-models-trained-on-cs-2-wafer-scale-systems", "source_type": "primary", "sources": [ { "label": "Dated seven-model release", "url": "https://www.cerebras.ai/press-release/cerebras-systems-releases-seven-new-gpt-models-trained-on-cs-2-wafer-scale-systems", "type": "announcement" }, { "label": "Official model architecture and size table", "url": "https://huggingface.co/cerebras/Cerebras-GPT-111M/blob/f91e002fe344cd90163ec56422de495ffd6576ba/README.md", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Publisher reports this nominal parameter count.", "availability": "public" }, { "id": "cerebras/cerebras-gpt-2.7b", "slug": "cerebras--cerebras-gpt-2.7b", "name": "Cerebras-GPT 2.7B", "developer": "Cerebras", "developer_id": "cerebras", "family": "cerebras-gpt", "description": null, "release_date": "2023-03-28", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 2.7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.cerebras.ai/press-release/cerebras-systems-releases-seven-new-gpt-models-trained-on-cs-2-wafer-scale-systems", "source_type": "primary", "sources": [ { "label": "Dated seven-model release", "url": "https://www.cerebras.ai/press-release/cerebras-systems-releases-seven-new-gpt-models-trained-on-cs-2-wafer-scale-systems", "type": "announcement" }, { "label": "Official model architecture and size table", "url": "https://huggingface.co/cerebras/Cerebras-GPT-111M/blob/f91e002fe344cd90163ec56422de495ffd6576ba/README.md", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Publisher reports this nominal parameter count.", "availability": "public" }, { "id": "cerebras/cerebras-gpt-256m", "slug": "cerebras--cerebras-gpt-256m", "name": "Cerebras-GPT 256M", "developer": "Cerebras", "developer_id": "cerebras", "family": "cerebras-gpt", "description": null, "release_date": "2023-03-28", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.256, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.cerebras.ai/press-release/cerebras-systems-releases-seven-new-gpt-models-trained-on-cs-2-wafer-scale-systems", "source_type": "primary", "sources": [ { "label": "Dated seven-model release", "url": "https://www.cerebras.ai/press-release/cerebras-systems-releases-seven-new-gpt-models-trained-on-cs-2-wafer-scale-systems", "type": "announcement" }, { "label": "Official model architecture and size table", "url": "https://huggingface.co/cerebras/Cerebras-GPT-111M/blob/f91e002fe344cd90163ec56422de495ffd6576ba/README.md", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Publisher reports this nominal parameter count.", "availability": "public" }, { "id": "cerebras/cerebras-gpt-590m", "slug": "cerebras--cerebras-gpt-590m", "name": "Cerebras-GPT 590M", "developer": "Cerebras", "developer_id": "cerebras", "family": "cerebras-gpt", "description": null, "release_date": "2023-03-28", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.59, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.cerebras.ai/press-release/cerebras-systems-releases-seven-new-gpt-models-trained-on-cs-2-wafer-scale-systems", "source_type": "primary", "sources": [ { "label": "Dated seven-model release", "url": "https://www.cerebras.ai/press-release/cerebras-systems-releases-seven-new-gpt-models-trained-on-cs-2-wafer-scale-systems", "type": "announcement" }, { "label": "Official model architecture and size table", "url": "https://huggingface.co/cerebras/Cerebras-GPT-111M/blob/f91e002fe344cd90163ec56422de495ffd6576ba/README.md", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Publisher reports this nominal parameter count.", "availability": "public" }, { "id": "cerebras/cerebras-gpt-6.7b", "slug": "cerebras--cerebras-gpt-6.7b", "name": "Cerebras-GPT 6.7B", "developer": "Cerebras", "developer_id": "cerebras", "family": "cerebras-gpt", "description": null, "release_date": "2023-03-28", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 6.7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.cerebras.ai/press-release/cerebras-systems-releases-seven-new-gpt-models-trained-on-cs-2-wafer-scale-systems", "source_type": "primary", "sources": [ { "label": "Dated seven-model release", "url": "https://www.cerebras.ai/press-release/cerebras-systems-releases-seven-new-gpt-models-trained-on-cs-2-wafer-scale-systems", "type": "announcement" }, { "label": "Official model architecture and size table", "url": "https://huggingface.co/cerebras/Cerebras-GPT-111M/blob/f91e002fe344cd90163ec56422de495ffd6576ba/README.md", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T09:56:46.659Z", "last_seen": "2026-10-29T09:56:46.659Z", "upstream_present": false, "notes": "Publisher reports this nominal parameter count.", "availability": "public" }, { "id": "anthropic/claude-1", "slug": "anthropic--claude-1", "name": "Claude 1", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude", "description": "Original Claude text assistant released by Anthropic in March 2023, before the later 100K-context upgrade.", "release_date": "2023-03-14", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 9000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.anthropic.com/news/introducing-claude", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://www.anthropic.com/news/introducing-claude", "type": "announcement" }, { "label": "Original 9K context and later expansion", "url": "https://www.anthropic.com/news/100k-context-windows", "type": "announcement" }, { "label": "Later model card covering previous Claude models", "url": "https://www-cdn.anthropic.com/files/4zrzovbb/website/5c49cc247484cecf107c699baf29250302e5da70.pdf?dl=ModelCardClaudev2_with_appendix_v1.pdf", "type": "model_card" }, { "label": "Claude 1 model retirement history", "url": "https://docs.anthropic.com/en/docs/about-claude/model-deprecations", "type": "documentation" }, { "label": "Anthropic statement on Claude weight availability", "url": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities", "type": "research" }, { "label": "Constitutional AI research model sizes and identity boundary", "url": "https://arxiv.org/pdf/2212.08073", "type": "paper" }, { "label": "Earlier RLHF research model sizes and identity boundary", "url": "https://arxiv.org/pdf/2204.05862", "type": "paper" }, { "label": "Later Claude comparison card", "url": "https://www-cdn.anthropic.com/c6a80a657af445f40e31afac050f3bf76d3b1404.pdf", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "This record denotes the original March 14, 2023 Claude release, not Claude 1.3 or its 100K-context upgrade. The earlier context is documented as a nominal 9K tokens. A later Anthropic model card identifies previous Claude models as transformers; original-Claude total/active parameters and dense versus MoE structure remain undisclosed. Published 52B research models are not explicitly identified as this commercial model. Anthropic states that Claude weights are not publicly released; proprietary is a descriptive classification. Claude 1.x API models were retired November 6, 2024.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T10:12:59Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/086.json" }, { "id": "anthropic/claude-instant-1", "slug": "anthropic--claude-instant-1", "name": "Claude Instant 1", "developer": "Anthropic", "developer_id": "anthropic", "family": "claude", "description": "Anthropic's original Claude Instant text model, introduced as a faster, lower-cost alternative to Claude.", "release_date": "2023-03-14", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 9000, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.anthropic.com/news/introducing-claude", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://www.anthropic.com/news/introducing-claude", "type": "announcement" }, { "label": "Claude model architecture card", "url": "https://www-cdn.anthropic.com/5c49cc247484cecf107c699baf29250302e5da70/claude-2-model-card.pdf", "type": "model_card" }, { "label": "Original Claude Instant context pricing sheet", "url": "https://www-cdn.anthropic.com/b0d46f8c8e2f4cca2833c17cb68435035c0bb5a9/apr-pricing-tokens_2023-05-10-213025_tnty.pdf", "type": "documentation" }, { "label": "Later 100K context upgrade", "url": "https://www.anthropic.com/news/100k-context-windows", "type": "announcement" }, { "label": "Later Claude Instant 1.2 release", "url": "https://www.anthropic.com/news/releasing-claude-instant-1-2", "type": "announcement" }, { "label": "Official retirement history", "url": "https://docs.anthropic.com/en/docs/about-claude/model-deprecations", "type": "documentation" }, { "label": "Anthropic statement on Claude weight availability", "url": "https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities", "type": "research" }, { "label": "Later Claude 2 comparison and appendix", "url": "https://www-cdn.anthropic.com/5c49cc247484cecf107c699baf29250302e5da70/ModelCardClaudev2_with_appendix_v1.pdf", "type": "model_card" }, { "label": "Later Claude 3 and 3.5 comparison cards", "url": "https://www-cdn.anthropic.com/c6a80a657af445f40e31afac050f3bf76d3b1404.pdf", "type": "model_card" }, { "label": "Earlier RLHF research model sizes", "url": "https://arxiv.org/html/2204.05862v1", "type": "research" }, { "label": "Earlier Constitutional AI research model sizes", "url": "https://arxiv.org/html/2212.08073v1", "type": "research" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "This record describes the original March 2023 Claude Instant offering, documented with a 9,000-token context. Anthropic announced a 100K context upgrade on May 11, 2023 and Instant 1.2 on August 9, 2023. Nominal total and active parameter counts remain undisclosed; published 52B research models are not identified as this release. Dense-versus-MoE structure remains undisclosed. Claude Instant 1.0, 1.1 and 1.2 API identifiers were retired November 6, 2024.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T10:28:58.398242Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/100.json" }, { "id": "openai/gpt-4", "slug": "openai--gpt-4", "name": "GPT-4", "developer": "OpenAI", "developer_id": "openai", "family": "gpt", "description": "Text-only GPT-4 API model with an 8,192-token context, later updated to the gpt-4-0613 snapshot.", "release_date": "2023-03-14", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "transformer", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://openai.com/index/gpt-4-research/", "source_type": "primary", "sources": [ { "label": "GPT-4 launch", "url": "https://openai.com/index/gpt-4-research/", "type": "announcement" }, { "label": "GPT-4 Technical Report", "url": "https://cdn.openai.com/papers/gpt-4.pdf", "type": "paper" }, { "label": "GPT-4 model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-4", "type": "documentation" }, { "label": "Function calling and other API updates", "url": "https://openai.com/index/function-calling-and-other-api-updates/", "type": "announcement" }, { "label": "Introducing Structured Outputs in the API", "url": "https://openai.com/index/introducing-structured-outputs-in-the-api/", "type": "announcement" }, { "label": "OpenAI model deprecations", "url": "https://developers.openai.com/api/docs/deprecations", "type": "documentation" }, { "label": "Extracting concepts from GPT-4", "url": "https://openai.com/index/extracting-concepts-from-gpt-4/", "type": "announcement" }, { "label": "Scaling and evaluating sparse autoencoders", "url": "https://cdn.openai.com/papers/sparse-autoencoders.pdf", "type": "paper" }, { "label": "Introducing gpt-oss", "url": "https://openai.com/index/introducing-gpt-oss/", "type": "announcement" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "documentation" }, { "label": "InfiBench: Evaluating the Question-Answering Capabilities of Code Large Language Models", "url": "https://papers.nips.cc/paper_files/paper/2024/file/e888eb9400fe14bb70e057aa1d719188-Paper-Datasets_and_Benchmarks_Track.pdf", "type": "paper" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-4.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "Historical gpt-4 text API alias, last pointing to gpt-4-0613. Initial public text access was March 14, 2023; image access described in the research launch was separate. Function calling was added with the June 2023 alias upgrade and is retained as historical capability despite today’s contradictory unsupported feature table. API access ended October 23, 2026. Transformer structure is documented; dense versus expert structure remains undisclosed. OpenAI withholds model size, and later exact-model comparison tables do not establish total or active parameter counts. Published estimates are unverified. The dedicated reasoning flag is unverified.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "reasoning": "unverified" }, "audited_at": "2026-10-27T16:21:03.928624Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/403.json" }, { "id": "zhipuai/chatglm-130b", "slug": "zhipuai--chatglm-130b", "name": "ChatGLM-130B", "developer": "Zhipu AI / Tsinghua University", "developer_id": "zhipuai", "family": "chatglm", "description": null, "release_date": "2023-03-14", "date_precision": "day", "open_weights": false, "license": null, "parameters_b": 130, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://arxiv.org/html/2406.12793v1", "source_type": "primary", "sources": [ { "label": "Model-team technical history", "url": "https://arxiv.org/html/2406.12793v1", "type": "paper" }, { "label": "Publisher release chronology", "url": "https://github.com/THUDM/GLM-130B/blob/212215c54f8a9da2deed51455868305235664370/README.md", "type": "repository" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "Aligned conversational model served at chatglm.cn; separate from the openly released GLM-130B pretrained checkpoint." }, { "id": "zhipuai/chatglm-6b", "slug": "zhipuai--chatglm-6b", "name": "ChatGLM-6B", "developer": "Zhipu AI / Tsinghua University", "developer_id": "zhipuai", "family": "chatglm", "description": null, "release_date": "2023-03-14", "date_precision": "day", "open_weights": true, "license": "ChatGLM Model License", "parameters_b": 6.2, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://arxiv.org/html/2406.12793v1", "source_type": "primary", "sources": [ { "label": "Model-team technical history", "url": "https://arxiv.org/html/2406.12793v1", "type": "paper" }, { "label": "Publisher release chronology", "url": "https://github.com/THUDM/ChatGLM-6B/blob/401bf3a8a7dd8a26fba189551dccfc61a7079b4e/README_en.md", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/THUDM/chatglm-6b/blob/bf0f5cfb575eebebf9b655c5861177acfee03f16/README.md", "type": "model_card" }, { "label": "Published configuration", "url": "https://huggingface.co/THUDM/chatglm-6b/blob/bf0f5cfb575eebebf9b655c5861177acfee03f16/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:05:52.591Z", "last_seen": "2026-10-29T10:05:52.591Z", "upstream_present": false, "availability": "public", "notes": "" }, { "id": "stanford/alpaca-7b", "slug": "stanford--alpaca-7b", "name": "Alpaca 7B", "developer": "Stanford University", "developer_id": "stanford", "family": "alpaca", "description": null, "release_date": "2023-03-13", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-4.0; LLaMA research license", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/tatsu-lab/alpaca-7b-wdiff", "source_type": "primary", "sources": [ { "label": "Stanford release announcement", "url": "https://crfm.stanford.edu/2023/03/13/alpaca.html", "type": "primary" }, { "label": "Official weight reconstruction instructions", "url": "https://github.com/tatsu-lab/stanford_alpaca/blob/761dc5bfbdeeffa89b8bff5d038781a4055f796a/README.md", "type": "documentation" }, { "label": "Weight-diff license", "url": "https://github.com/tatsu-lab/stanford_alpaca/blob/761dc5bfbdeeffa89b8bff5d038781a4055f796a/WEIGHT_DIFF_LICENSE", "type": "documentation" }, { "label": "Official configuration", "url": "https://huggingface.co/tatsu-lab/alpaca-7b-wdiff/blob/8648f73a244a70f4e39884e0f7670fe362aa3cf3/config.json", "type": "config" } ], "aliases": [ "tatsu-lab/alpaca-7b-wdiff" ], "first_seen": "2026-10-30T07:09:10.442Z", "last_seen": "2026-10-30T07:09:10.442Z", "upstream_present": false, "availability": "public", "notes": "First external access was the March 13 public demo. The later publisher weight diff reconstructs the original Alpaca checkpoint from LLaMA; it is an equivalent downloadable release, not a separate model. The diff is CC-BY-NC-4.0 and the base retains its research restrictions. Context comes from the published configuration; fine-tuning sequences were 512 tokens. Size is the publisher’s nominal count." }, { "id": "ai21/j2-grande", "slug": "ai21--j2-grande", "name": "Jurassic-2 Grande", "developer": "AI21 Labs", "developer_id": "ai21", "family": "jurassic-2", "description": null, "release_date": "2023-03-09", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.ai21.com/blog/introducing-j2/", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://www.ai21.com/blog/introducing-j2/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:17:34.586Z", "last_seen": "2026-10-29T10:17:34.586Z", "upstream_present": false, "availability": "public", "notes": "The publisher release explicitly lists three base models and two instruction-tuned variants as available through its API and playground. Parameter counts and supported context for this original revision remain unverified." }, { "id": "ai21/j2-grande-instruct", "slug": "ai21--j2-grande-instruct", "name": "Jurassic-2 Grande Instruct", "developer": "AI21 Labs", "developer_id": "ai21", "family": "jurassic-2", "description": null, "release_date": "2023-03-09", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.ai21.com/blog/introducing-j2/", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://www.ai21.com/blog/introducing-j2/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:17:34.586Z", "last_seen": "2026-10-29T10:17:34.586Z", "upstream_present": false, "availability": "public", "notes": "The publisher release explicitly lists three base models and two instruction-tuned variants as available through its API and playground. Parameter counts and supported context for this original revision remain unverified." }, { "id": "ai21/j2-jumbo", "slug": "ai21--j2-jumbo", "name": "Jurassic-2 Jumbo", "developer": "AI21 Labs", "developer_id": "ai21", "family": "jurassic-2", "description": null, "release_date": "2023-03-09", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.ai21.com/blog/introducing-j2/", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://www.ai21.com/blog/introducing-j2/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:17:34.586Z", "last_seen": "2026-10-29T10:17:34.586Z", "upstream_present": false, "availability": "public", "notes": "The publisher release explicitly lists three base models and two instruction-tuned variants as available through its API and playground. Parameter counts and supported context for this original revision remain unverified." }, { "id": "ai21/j2-jumbo-instruct", "slug": "ai21--j2-jumbo-instruct", "name": "Jurassic-2 Jumbo Instruct", "developer": "AI21 Labs", "developer_id": "ai21", "family": "jurassic-2", "description": null, "release_date": "2023-03-09", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.ai21.com/blog/introducing-j2/", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://www.ai21.com/blog/introducing-j2/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:17:34.586Z", "last_seen": "2026-10-29T10:17:34.586Z", "upstream_present": false, "availability": "public", "notes": "The publisher release explicitly lists three base models and two instruction-tuned variants as available through its API and playground. Parameter counts and supported context for this original revision remain unverified." }, { "id": "ai21/j2-large", "slug": "ai21--j2-large", "name": "Jurassic-2 Large", "developer": "AI21 Labs", "developer_id": "ai21", "family": "jurassic-2", "description": null, "release_date": "2023-03-09", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.ai21.com/blog/introducing-j2/", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://www.ai21.com/blog/introducing-j2/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:17:34.586Z", "last_seen": "2026-10-29T10:17:34.586Z", "upstream_present": false, "availability": "public", "notes": "The publisher release explicitly lists three base models and two instruction-tuned variants as available through its API and playground. Parameter counts and supported context for this original revision remain unverified." }, { "id": "openai/gpt-3.5-turbo", "slug": "openai--gpt-3.5-turbo", "name": "GPT-3.5 Turbo", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-3.5", "description": "Chat model for text and code tasks, with an unpinned alias that last pointed to gpt-3.5-turbo-0125.", "release_date": "2023-03-01", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 16385, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": true, "source_url": "https://openai.com/index/introducing-chatgpt-and-whisper-apis/", "source_type": "primary", "sources": [ { "label": "GPT-3.5 Turbo model documentation", "url": "https://developers.openai.com/api/docs/models/gpt-3.5-turbo.md", "type": "primary" }, { "label": "Introducing ChatGPT and Whisper APIs", "url": "https://openai.com/index/introducing-chatgpt-and-whisper-apis/", "type": "primary" }, { "label": "OpenAI API launch announcement, March 1, 2023", "url": "https://community.openai.com/t/introducing-chatgpt-and-whisper-apis/80485", "type": "primary" }, { "label": "Function calling and other API updates", "url": "https://openai.com/index/function-calling-and-other-api-updates/", "type": "primary" }, { "label": "New embedding models and API updates", "url": "https://openai.com/index/new-embedding-models-and-api-updates/", "type": "primary" }, { "label": "OpenAI model deprecations", "url": "https://developers.openai.com/api/docs/deprecations", "type": "primary" }, { "label": "OpenAI Services Agreement", "url": "https://openai.com/policies/services-agreement/", "type": "primary" }, { "label": "CODEFUSION author withdrawal notice", "url": "https://arxiv.org/abs/2310.17680", "type": "primary" }, { "label": "CODEFUSION revised camera-ready paper", "url": "https://www.microsoft.com/en-us/research/wp-content/uploads/2023/11/CodeFusion-Revised-CameraReady.pdf", "type": "primary" }, { "label": "CODEFUSION final EMNLP 2023 proceedings paper", "url": "https://aclanthology.org/2023.emnlp-main.716.pdf", "type": "primary" }, { "label": "RAD-Bench: Evaluating Large Language Models’ Capabilities in Retrieval Augmented Dialogues", "url": "https://aclanthology.org/2025.naacl-industry.66.pdf", "type": "primary" }, { "label": "DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model", "url": "https://arxiv.org/html/2405.04434v1", "type": "primary" }, { "label": "Models.dev record", "url": "https://github.com/anomalyco/models.dev/blob/dev/models/openai/gpt-3.5-turbo.toml", "type": "catalog" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": true, "notes": "The alias launched March 1, 2023 and last pointed to gpt-3.5-turbo-0125; its 16,385-token context describes that later target, not the original 0301 snapshot. Function calling is documented by the June 2023 upgrade and January 2024 release, despite a conflicting current model-page feature table. OpenAI shut down API access on October 23, 2026. Total and active parameter counts remain undisclosed. The CODEFUSION 20B claim was withdrawn; its revised tables say Unknown, and RAD-Bench leaves the active count blank. Exact architecture remains unverified.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "unverified", "reasoning": "unverified" }, "audited_at": "2026-10-27T16:21:14.118878Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/402.json" }, { "id": "databricks/dolly-v1-6b", "slug": "databricks--dolly-v1-6b", "name": "Dolly v1 6B", "developer": "Databricks", "developer_id": "databricks", "family": "dolly", "description": null, "release_date": "2023-03", "date_precision": "month", "open_weights": true, "license": null, "parameters_b": 6, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/databrickslabs/dolly/blob/e876704bf4086a63af34f7c6f56cbd0911fd3025/README.md", "source_type": "primary", "sources": [ { "label": "March 30 publisher documentation adds explicit pretrained download and model specifications", "url": "https://github.com/databrickslabs/dolly/blob/e876704bf4086a63af34f7c6f56cbd0911fd3025/README.md", "type": "repository" }, { "label": "Original model announcement", "url": "https://www.databricks.com/blog/2023/03/24/hello-dolly-democratizing-magic-chatgpt-open-models.html", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T11:26:55.844Z", "last_seen": "2026-10-29T11:26:55.844Z", "upstream_present": false, "availability": "public", "announcement_date": "2023-03-24", "notes": "Databricks instruction-tuned GPT-J using Stanford Alpaca data. The March 24 announcement initially released training code; March 30 publisher documentation explicitly links pretrained Dolly v1 weights. Month precision avoids assigning the code-announcement day to weight availability. Nominal 6B size is explicit in the developer description. The current original weight endpoint returns 401; exact model-weight license and supported context remain unverified." }, { "id": "meta/llama-1-13b", "slug": "meta--llama-1-13b", "name": "LLaMA 13B", "developer": "Meta", "developer_id": "meta", "family": "llama", "description": "First-generation LLaMA research checkpoint.", "release_date": "2023-02-24", "date_precision": "day", "open_weights": true, "license": "LLaMA non-commercial research license", "parameters_b": 13, "active_parameters_b": 13, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/large-language-model-Llama-meta-ai/", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://ai.meta.com/blog/large-language-model-Llama-meta-ai/", "type": "announcement" }, { "label": "Original LLaMA paper, Table 2", "url": "https://arxiv.org/html/2302.13971v1", "type": "primary" }, { "label": "Original LLaMA v1 model card", "url": "https://raw.githubusercontent.com/meta-llama/llama/57b0eb62de0636e75af471e49e2f1862d908d9d8/MODEL_CARD.md", "type": "primary" }, { "label": "Original LLaMA architecture code", "url": "https://raw.githubusercontent.com/meta-llama/llama/57b0eb62de0636e75af471e49e2f1862d908d9d8/llama/model.py", "type": "primary" }, { "label": "Original gated-checkpoint release procedure", "url": "https://raw.githubusercontent.com/meta-llama/llama/57b0eb62de0636e75af471e49e2f1862d908d9d8/README.md", "type": "primary" }, { "label": "Meta first-generation context comparison", "url": "https://arxiv.org/html/2307.09288v2", "type": "primary" }, { "label": "Original LLaMA generation interface", "url": "https://raw.githubusercontent.com/meta-llama/llama/57b0eb62de0636e75af471e49e2f1862d908d9d8/llama/generation.py", "type": "primary" }, { "label": "JetMoE team comparison: original LLaMA-13B active parameters", "url": "https://research.myshell.ai/jetmoe", "type": "primary" }, { "label": "JetMoE pinned model card: LLaMA-13B nominal active count", "url": "https://huggingface.co/jetmoe/jetmoe-8b/blob/0ca477416c6fd12ced34fba84630462b241a6439/README.md", "type": "primary" }, { "label": "LLaMA paper", "url": "https://arxiv.org/abs/2302.13971", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "Original first-generation LLaMA 13B. Total parameters follow Meta's rounded 13.0B specification; nominal active 13B follows the JetMoE research team's comparison table, which does not specify exact per-token participation or embedding inclusion. Release date is the February 24, 2023 announcement and access-request opening; weights were delivered after case-by-case research approval. open_weights denotes released gated weights. Weights use a bespoke noncommercial research license; reference code uses GPLv3. Dedicated thinking mode and native tool calling are not documented.", "field_status": { "reasoning": "undisclosed", "tool_call": "undisclosed" }, "audited_at": "2026-10-27T13:46:35.564119+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/254.json" }, { "id": "meta/llama-1-33b", "slug": "meta--llama-1-33b", "name": "LLaMA 33B", "developer": "Meta", "developer_id": "meta", "family": "llama", "description": "First-generation LLaMA research checkpoint.", "release_date": "2023-02-24", "date_precision": "day", "open_weights": true, "license": "LLaMA non-commercial research license", "parameters_b": 32.5, "active_parameters_b": 33, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/large-language-model-Llama-meta-ai/", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://ai.meta.com/blog/large-language-model-Llama-meta-ai/", "type": "announcement" }, { "label": "Original LLaMA paper: Table 2 parameter count", "url": "https://arxiv.org/pdf/2302.13971", "type": "primary" }, { "label": "Original LLaMA v1 model card", "url": "https://github.com/meta-llama/llama/blob/53011c3d7946dadb8274a4c5c7586ab54edf792d/MODEL_CARD.md", "type": "primary" }, { "label": "Original LLaMA v1 architecture", "url": "https://github.com/meta-llama/llama/blob/57b0eb62de0636e75af471e49e2f1862d908d9d8/llama/model.py", "type": "primary" }, { "label": "Original gated-checkpoint download documentation", "url": "https://github.com/meta-llama/llama/blob/57b0eb62de0636e75af471e49e2f1862d908d9d8/README.md", "type": "primary" }, { "label": "Original LLaMA text-generation interface", "url": "https://github.com/meta-llama/llama/blob/57b0eb62de0636e75af471e49e2f1862d908d9d8/llama/generation.py", "type": "primary" }, { "label": "Meta Llama 2 paper: explicit Llama 1 context comparison", "url": "https://arxiv.org/html/2307.09288v2", "type": "primary" }, { "label": "Mistral Mixtral paper: LLaMA 1 33B nominal active count", "url": "https://arxiv.org/html/2401.04088v1", "type": "primary" }, { "label": "DeepSeek-V2 Figure 1: LLaMA 1 33B activated-count corroboration", "url": "https://arxiv.org/pdf/2405.04434v1", "type": "primary" }, { "label": "LLaMA paper", "url": "https://arxiv.org/abs/2302.13971", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "Original LLaMA-33B: Meta reports 32.5B total parameters. Mistral reports nominal 33B active in its Mixtral benchmark tables; the coarser count is a comparison convention, not Meta's exact per-token disclosure. Native thinking/tool-call flags are undocumented. February 24, 2023 dates the gated research announcement; access required approval. Weights use the noncommercial research license, separate from the code's GPLv3 and later Llama 2 license.", "field_status": { "reasoning": "undisclosed", "tool_call": "undisclosed" }, "audited_at": "2026-10-27T13:49:53.859766Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/255.json" }, { "id": "meta/llama-1-65b", "slug": "meta--llama-1-65b", "name": "LLaMA 65B", "developer": "Meta", "developer_id": "meta", "family": "llama", "description": "First-generation LLaMA research checkpoint.", "release_date": "2023-02-24", "date_precision": "day", "open_weights": true, "license": "LLaMA non-commercial research license", "parameters_b": 65.2, "active_parameters_b": 65, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/large-language-model-Llama-meta-ai/", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://ai.meta.com/blog/large-language-model-Llama-meta-ai/", "type": "announcement" }, { "label": "Original LLaMA model card, llama_v1", "url": "https://raw.githubusercontent.com/meta-llama/llama/57b0eb62de0636e75af471e49e2f1862d908d9d8/MODEL_CARD.md", "type": "primary" }, { "label": "LLaMA: Open and Efficient Foundation Language Models", "url": "https://arxiv.org/html/2302.13971v1", "type": "primary" }, { "label": "Original LLaMA Transformer implementation", "url": "https://raw.githubusercontent.com/meta-llama/llama/57b0eb62de0636e75af471e49e2f1862d908d9d8/llama/model.py", "type": "primary" }, { "label": "Original LLaMA generation interface", "url": "https://raw.githubusercontent.com/meta-llama/llama/57b0eb62de0636e75af471e49e2f1862d908d9d8/llama/generation.py", "type": "primary" }, { "label": "Original LLaMA release README", "url": "https://raw.githubusercontent.com/meta-llama/llama/57b0eb62de0636e75af471e49e2f1862d908d9d8/README.md", "type": "primary" }, { "label": "Llama 2: Open Foundation and Fine-Tuned Chat Models, Table 1 comparison", "url": "https://arxiv.org/html/2307.09288v2", "type": "primary" }, { "label": "LLaMA: Open and Efficient Foundation Language Models, original PDF Table 2", "url": "https://arxiv.org/pdf/2302.13971v1", "type": "primary" }, { "label": "DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model, Figure 1(a)", "url": "https://arxiv.org/html/2405.04434v1", "type": "primary" }, { "label": "LLaMA paper", "url": "https://arxiv.org/abs/2302.13971", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "Gated research weights under a noncommercial model license; February 24, 2023 is the release announcement date. Meta reports 65.2B total parameters. Approximately 65B active follows DeepSeek-V2 Figure 1(a)'s benchmark convention for LLaMA 1 65B, not a Meta disclosure or exact per-token census. Dedicated thinking/tool-call capabilities remain unspecified; the model license differs from GPL for inference code.", "field_status": { "reasoning": "undisclosed", "tool_call": "undisclosed" }, "audited_at": "2026-10-27T13:50:38.276596Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/256.json" }, { "id": "meta/llama-1-7b", "slug": "meta--llama-1-7b", "name": "LLaMA 7B", "developer": "Meta", "developer_id": "meta", "family": "llama", "description": "First-generation LLaMA research checkpoint.", "release_date": "2023-02-24", "date_precision": "day", "open_weights": true, "license": "LLaMA non-commercial research license", "parameters_b": 6.7, "active_parameters_b": 6.7, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/large-language-model-Llama-meta-ai/", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://ai.meta.com/blog/large-language-model-Llama-meta-ai/", "type": "announcement" }, { "label": "LLaMA: Open and Efficient Foundation Language Models", "url": "https://arxiv.org/html/2302.13971v1", "type": "primary" }, { "label": "Original release LLaMA version 1 model card", "url": "https://raw.githubusercontent.com/meta-llama/llama/1076b9c51c77ad06e9d7ba8a4c6df775741732bd/MODEL_CARD.md", "type": "primary" }, { "label": "Original release LLaMA Transformer implementation", "url": "https://raw.githubusercontent.com/meta-llama/llama/1076b9c51c77ad06e9d7ba8a4c6df775741732bd/llama/model.py", "type": "primary" }, { "label": "Original release LLaMA README", "url": "https://raw.githubusercontent.com/meta-llama/llama/1076b9c51c77ad06e9d7ba8a4c6df775741732bd/README.md", "type": "primary" }, { "label": "Original release LLaMA generation interface", "url": "https://raw.githubusercontent.com/meta-llama/llama/1076b9c51c77ad06e9d7ba8a4c6df775741732bd/llama/generation.py", "type": "primary" }, { "label": "Original LLaMA context-default correction", "url": "https://github.com/meta-llama/llama/commit/a81fb4e211d6656854a5dd24cf8631dc319234d1", "type": "primary" }, { "label": "Llama 2 technical report: explicit Llama 1 comparison", "url": "https://arxiv.org/html/2307.09288v2", "type": "primary" }, { "label": "Original release repository software license", "url": "https://raw.githubusercontent.com/meta-llama/llama/1076b9c51c77ad06e9d7ba8a4c6df775741732bd/LICENSE", "type": "primary" }, { "label": "DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture-of-Experts Language Models", "url": "https://aclanthology.org/2024.acl-long.70.pdf", "type": "primary" }, { "label": "LLaMA paper", "url": "https://arxiv.org/abs/2302.13971", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "Original LLaMA 7B: Meta reports a rounded 6.7B total; DeepSeekMoE's comparison plots approximately 6.7B activated parameters, a benchmark convention rather than a Meta tensor-exact specification. February 24, 2023 is the public announcement and access-request date; checkpoint delivery required research approval. open_weights means released gated weights, not unrestricted access. Weights use a bespoke noncommercial research license; reference software uses GPLv3. The original context is 2048 tokens. Dedicated thinking mode and native tool calling are not documented.", "field_status": { "reasoning": "undisclosed", "tool_call": "undisclosed" }, "audited_at": "2026-10-27T13:51:47.714116Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/257.json" }, { "id": "bigcode/santacoder", "slug": "bigcode--santacoder", "name": "SantaCoder", "developer": "BigCode", "developer_id": "bigcode", "family": "santacoder", "description": null, "release_date": "2022-12-22", "date_precision": "day", "open_weights": true, "license": "BigCode OpenRAIL-M v1", "parameters_b": 1.1, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.bigcode-project.org/docs/about/timeline/", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://www.bigcode-project.org/docs/about/timeline/", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/bigcode/santacoder", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:10:00.648Z", "last_seen": "2026-10-29T10:10:00.648Z", "upstream_present": false, "availability": "public", "notes": "The project timeline explicitly dates the public release. Parameters are the developer nominal count." }, { "id": "meta/opt-iml-30b", "slug": "meta--opt-iml-30b", "name": "OPT-IML 30B", "developer": "Meta", "developer_id": "meta", "family": "opt", "description": null, "release_date": "2022-12-22", "date_precision": "day", "open_weights": true, "license": "OPT-IML 175B License Agreement", "parameters_b": 30, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/facebookresearch/metaseq/blob/e6f2ad17f5756ebce769e2a53ade0e5d93b94008/projects/OPT-IML/README.md", "source_type": "primary", "sources": [ { "label": "Original December 22 publisher release page", "url": "https://github.com/facebookresearch/metaseq/blob/e6f2ad17f5756ebce769e2a53ade0e5d93b94008/projects/OPT-IML/README.md", "type": "documentation" }, { "label": "Original model paper, variants and training sequence length", "url": "https://arxiv.org/html/2212.12017v1", "type": "paper" }, { "label": "Publisher current model card", "url": "https://huggingface.co/facebook/opt-iml-30b", "type": "model_card" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/facebook/opt-iml-30b/blob/main/config.json", "type": "documentation" }, { "label": "Model weights license", "url": "https://huggingface.co/facebook/opt-iml-30b/blob/main/LICENSE.md", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:57:14.174Z", "last_seen": "2026-10-29T10:57:14.174Z", "upstream_present": false, "availability": "public", "notes": "Original publisher instruction-tuned OPT checkpoint. Direct complete model links are present in the dated December 22 GitHub release; January 2023 Hugging Face uploads are later hosting. Ordinary OPT-IML uses approximately 1,500 tasks, holding out others for evaluation. Parameters are nominal; context is verified by original paper and checkpoint config.", "announcement_date": "2022-12-22" }, { "id": "meta/opt-iml-max-30b", "slug": "meta--opt-iml-max-30b", "name": "OPT-IML-Max 30B", "developer": "Meta", "developer_id": "meta", "family": "opt", "description": null, "release_date": "2022-12-22", "date_precision": "day", "open_weights": true, "license": "OPT-IML 175B License Agreement", "parameters_b": 30, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/facebookresearch/metaseq/blob/e6f2ad17f5756ebce769e2a53ade0e5d93b94008/projects/OPT-IML/README.md", "source_type": "primary", "sources": [ { "label": "Original December 22 publisher release page", "url": "https://github.com/facebookresearch/metaseq/blob/e6f2ad17f5756ebce769e2a53ade0e5d93b94008/projects/OPT-IML/README.md", "type": "documentation" }, { "label": "Original model paper, variants and training sequence length", "url": "https://arxiv.org/html/2212.12017v1", "type": "paper" }, { "label": "Publisher current model card", "url": "https://huggingface.co/facebook/opt-iml-max-30b", "type": "model_card" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/facebook/opt-iml-max-30b/blob/main/config.json", "type": "documentation" }, { "label": "Model weights license", "url": "https://huggingface.co/facebook/opt-iml-max-30b/blob/main/LICENSE.md", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:57:14.174Z", "last_seen": "2026-10-29T10:57:14.174Z", "upstream_present": false, "availability": "public", "notes": "Original publisher instruction-tuned OPT checkpoint. Direct complete model links are present in the dated December 22 GitHub release; January 2023 Hugging Face uploads are later hosting. Max uses approximately 2,000 tasks. Parameters are nominal; context is verified by original paper and checkpoint config.", "announcement_date": "2022-12-22" }, { "id": "eleutherai/pythia-1.4b-deduped-v0", "slug": "eleutherai--pythia-1.4b-deduped-v0", "name": "Pythia 1.4B Deduped v0", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2022-12", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 1.414647808, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "source_type": "primary", "sources": [ { "label": "Original release documentation", "url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "type": "documentation" }, { "label": "Publisher v0 model card and size mapping", "url": "https://huggingface.co/EleutherAI/pythia-1.4b-deduped-v0", "type": "model_card" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/EleutherAI/pythia-1.4b-deduped-v0/blob/main/config.json", "type": "documentation" }, { "label": "Version history", "url": "https://github.com/EleutherAI/pythia#changelog", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:36:19.054Z", "last_seen": "2026-10-29T10:36:19.054Z", "upstream_present": false, "availability": "public", "notes": "Original completed training run, distinct from the corrected suite released April 3, 2023. The December 2022 publisher documentation links all 16 checkpoints and says each main branch is the completed 143000-step model. Originally named Pythia-1.3B-Deduped; later renamed to include embedding parameters. Total parameters use the publisher table, not tensor-storage totals that also include attention-mask buffers." }, { "id": "eleutherai/pythia-1.4b-v0", "slug": "eleutherai--pythia-1.4b-v0", "name": "Pythia 1.4B v0", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2022-12", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 1.414647808, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "source_type": "primary", "sources": [ { "label": "Original release documentation", "url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "type": "documentation" }, { "label": "Publisher v0 model card and size mapping", "url": "https://huggingface.co/EleutherAI/pythia-1.4b-v0", "type": "model_card" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/EleutherAI/pythia-1.4b-v0/blob/main/config.json", "type": "documentation" }, { "label": "Version history", "url": "https://github.com/EleutherAI/pythia#changelog", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:36:19.054Z", "last_seen": "2026-10-29T10:36:19.054Z", "upstream_present": false, "availability": "public", "notes": "Original completed training run, distinct from the corrected suite released April 3, 2023. The December 2022 publisher documentation links all 16 checkpoints and says each main branch is the completed 143000-step model. Originally named Pythia-1.3B; later renamed to include embedding parameters. Total parameters use the publisher table, not tensor-storage totals that also include attention-mask buffers." }, { "id": "eleutherai/pythia-12b-deduped-v0", "slug": "eleutherai--pythia-12b-deduped-v0", "name": "Pythia 12B Deduped v0", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2022-12", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 11.84607232, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "source_type": "primary", "sources": [ { "label": "Original release documentation", "url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "type": "documentation" }, { "label": "Publisher v0 model card and size mapping", "url": "https://huggingface.co/EleutherAI/pythia-12b-deduped-v0", "type": "model_card" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/EleutherAI/pythia-12b-deduped-v0/blob/main/config.json", "type": "documentation" }, { "label": "Version history", "url": "https://github.com/EleutherAI/pythia#changelog", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:36:19.054Z", "last_seen": "2026-10-29T10:36:19.054Z", "upstream_present": false, "availability": "public", "notes": "Original completed training run, distinct from the corrected suite released April 3, 2023. The December 2022 publisher documentation links all 16 checkpoints and says each main branch is the completed 143000-step model. Originally named Pythia-13B-Deduped; later renamed to include embedding parameters. Total parameters use the publisher table, not tensor-storage totals that also include attention-mask buffers." }, { "id": "eleutherai/pythia-12b-v0", "slug": "eleutherai--pythia-12b-v0", "name": "Pythia 12B v0", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2022-12", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 11.84607232, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "source_type": "primary", "sources": [ { "label": "Original release documentation", "url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "type": "documentation" }, { "label": "Publisher v0 model card and size mapping", "url": "https://huggingface.co/EleutherAI/pythia-12b-v0", "type": "model_card" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/EleutherAI/pythia-12b-v0/blob/main/config.json", "type": "documentation" }, { "label": "Version history", "url": "https://github.com/EleutherAI/pythia#changelog", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:36:19.054Z", "last_seen": "2026-10-29T10:36:19.054Z", "upstream_present": false, "availability": "public", "notes": "Original completed training run, distinct from the corrected suite released April 3, 2023. The December 2022 publisher documentation links all 16 checkpoints and says each main branch is the completed 143000-step model. Originally named Pythia-13B; later renamed to include embedding parameters. Total parameters use the publisher table, not tensor-storage totals that also include attention-mask buffers." }, { "id": "eleutherai/pythia-160m-deduped-v0", "slug": "eleutherai--pythia-160m-deduped-v0", "name": "Pythia 160M Deduped v0", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2022-12", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.162322944, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "source_type": "primary", "sources": [ { "label": "Original release documentation", "url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "type": "documentation" }, { "label": "Publisher v0 model card and size mapping", "url": "https://huggingface.co/EleutherAI/pythia-160m-deduped-v0", "type": "model_card" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/EleutherAI/pythia-160m-deduped-v0/blob/main/config.json", "type": "documentation" }, { "label": "Version history", "url": "https://github.com/EleutherAI/pythia#changelog", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:36:19.054Z", "last_seen": "2026-10-29T10:36:19.054Z", "upstream_present": false, "availability": "public", "notes": "Original completed training run, distinct from the corrected suite released April 3, 2023. The December 2022 publisher documentation links all 16 checkpoints and says each main branch is the completed 143000-step model. Originally named Pythia-125M-Deduped; later renamed to include embedding parameters. Total parameters use the publisher table, not tensor-storage totals that also include attention-mask buffers." }, { "id": "eleutherai/pythia-160m-v0", "slug": "eleutherai--pythia-160m-v0", "name": "Pythia 160M v0", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2022-12", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.162322944, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "source_type": "primary", "sources": [ { "label": "Original release documentation", "url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "type": "documentation" }, { "label": "Publisher v0 model card and size mapping", "url": "https://huggingface.co/EleutherAI/pythia-160m-v0", "type": "model_card" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/EleutherAI/pythia-160m-v0/blob/main/config.json", "type": "documentation" }, { "label": "Version history", "url": "https://github.com/EleutherAI/pythia#changelog", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:36:19.054Z", "last_seen": "2026-10-29T10:36:19.054Z", "upstream_present": false, "availability": "public", "notes": "Original completed training run, distinct from the corrected suite released April 3, 2023. The December 2022 publisher documentation links all 16 checkpoints and says each main branch is the completed 143000-step model. Originally named Pythia-125M; later renamed to include embedding parameters. Total parameters use the publisher table, not tensor-storage totals that also include attention-mask buffers." }, { "id": "eleutherai/pythia-1b-deduped-v0", "slug": "eleutherai--pythia-1b-deduped-v0", "name": "Pythia 1B Deduped v0", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2022-12", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 1.011781632, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "source_type": "primary", "sources": [ { "label": "Original release documentation", "url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "type": "documentation" }, { "label": "Publisher v0 model card and size mapping", "url": "https://huggingface.co/EleutherAI/pythia-1b-deduped-v0", "type": "model_card" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/EleutherAI/pythia-1b-deduped-v0/blob/main/config.json", "type": "documentation" }, { "label": "Version history", "url": "https://github.com/EleutherAI/pythia#changelog", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:36:19.054Z", "last_seen": "2026-10-29T10:36:19.054Z", "upstream_present": false, "availability": "public", "notes": "Original completed training run, distinct from the corrected suite released April 3, 2023. The December 2022 publisher documentation links all 16 checkpoints and says each main branch is the completed 143000-step model. Originally named Pythia-800M-Deduped; later renamed to include embedding parameters. Total parameters use the publisher table, not tensor-storage totals that also include attention-mask buffers." }, { "id": "eleutherai/pythia-1b-v0", "slug": "eleutherai--pythia-1b-v0", "name": "Pythia 1B v0", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2022-12", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 1.011781632, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "source_type": "primary", "sources": [ { "label": "Original release documentation", "url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "type": "documentation" }, { "label": "Publisher v0 model card and size mapping", "url": "https://huggingface.co/EleutherAI/pythia-1b-v0", "type": "model_card" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/EleutherAI/pythia-1b-v0/blob/main/config.json", "type": "documentation" }, { "label": "Version history", "url": "https://github.com/EleutherAI/pythia#changelog", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:36:19.054Z", "last_seen": "2026-10-29T10:36:19.054Z", "upstream_present": false, "availability": "public", "notes": "Original completed training run, distinct from the corrected suite released April 3, 2023. The December 2022 publisher documentation links all 16 checkpoints and says each main branch is the completed 143000-step model. Originally named Pythia-800M; later renamed to include embedding parameters. Total parameters use the publisher table, not tensor-storage totals that also include attention-mask buffers." }, { "id": "eleutherai/pythia-2.8b-deduped-v0", "slug": "eleutherai--pythia-2.8b-deduped-v0", "name": "Pythia 2.8B Deduped v0", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2022-12", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 2.77520896, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "source_type": "primary", "sources": [ { "label": "Original release documentation", "url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "type": "documentation" }, { "label": "Publisher v0 model card and size mapping", "url": "https://huggingface.co/EleutherAI/pythia-2.8b-deduped-v0", "type": "model_card" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/EleutherAI/pythia-2.8b-deduped-v0/blob/main/config.json", "type": "documentation" }, { "label": "Version history", "url": "https://github.com/EleutherAI/pythia#changelog", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:36:19.054Z", "last_seen": "2026-10-29T10:36:19.054Z", "upstream_present": false, "availability": "public", "notes": "Original completed training run, distinct from the corrected suite released April 3, 2023. The December 2022 publisher documentation links all 16 checkpoints and says each main branch is the completed 143000-step model. Originally named Pythia-2.7B-Deduped; later renamed to include embedding parameters. Total parameters use the publisher table, not tensor-storage totals that also include attention-mask buffers." }, { "id": "eleutherai/pythia-2.8b-v0", "slug": "eleutherai--pythia-2.8b-v0", "name": "Pythia 2.8B v0", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2022-12", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 2.77520896, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "source_type": "primary", "sources": [ { "label": "Original release documentation", "url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "type": "documentation" }, { "label": "Publisher v0 model card and size mapping", "url": "https://huggingface.co/EleutherAI/pythia-2.8b-v0", "type": "model_card" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/EleutherAI/pythia-2.8b-v0/blob/main/config.json", "type": "documentation" }, { "label": "Version history", "url": "https://github.com/EleutherAI/pythia#changelog", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:36:19.054Z", "last_seen": "2026-10-29T10:36:19.054Z", "upstream_present": false, "availability": "public", "notes": "Original completed training run, distinct from the corrected suite released April 3, 2023. The December 2022 publisher documentation links all 16 checkpoints and says each main branch is the completed 143000-step model. Originally named Pythia-2.7B; later renamed to include embedding parameters. Total parameters use the publisher table, not tensor-storage totals that also include attention-mask buffers." }, { "id": "eleutherai/pythia-410m-deduped-v0", "slug": "eleutherai--pythia-410m-deduped-v0", "name": "Pythia 410M Deduped v0", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2022-12", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.405334016, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "source_type": "primary", "sources": [ { "label": "Original release documentation", "url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "type": "documentation" }, { "label": "Publisher v0 model card and size mapping", "url": "https://huggingface.co/EleutherAI/pythia-410m-deduped-v0", "type": "model_card" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/EleutherAI/pythia-410m-deduped-v0/blob/main/config.json", "type": "documentation" }, { "label": "Version history", "url": "https://github.com/EleutherAI/pythia#changelog", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:36:19.054Z", "last_seen": "2026-10-29T10:36:19.054Z", "upstream_present": false, "availability": "public", "notes": "Original completed training run, distinct from the corrected suite released April 3, 2023. The December 2022 publisher documentation links all 16 checkpoints and says each main branch is the completed 143000-step model. Originally named Pythia-350M-Deduped; later renamed to include embedding parameters. Total parameters use the publisher table, not tensor-storage totals that also include attention-mask buffers." }, { "id": "eleutherai/pythia-410m-v0", "slug": "eleutherai--pythia-410m-v0", "name": "Pythia 410M v0", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2022-12", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.405334016, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "source_type": "primary", "sources": [ { "label": "Original release documentation", "url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "type": "documentation" }, { "label": "Publisher v0 model card and size mapping", "url": "https://huggingface.co/EleutherAI/pythia-410m-v0", "type": "model_card" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/EleutherAI/pythia-410m-v0/blob/main/config.json", "type": "documentation" }, { "label": "Version history", "url": "https://github.com/EleutherAI/pythia#changelog", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:36:19.054Z", "last_seen": "2026-10-29T10:36:19.054Z", "upstream_present": false, "availability": "public", "notes": "Original completed training run, distinct from the corrected suite released April 3, 2023. The December 2022 publisher documentation links all 16 checkpoints and says each main branch is the completed 143000-step model. Originally named Pythia-350M; later renamed to include embedding parameters. Total parameters use the publisher table, not tensor-storage totals that also include attention-mask buffers." }, { "id": "eleutherai/pythia-6.9b-deduped-v0", "slug": "eleutherai--pythia-6.9b-deduped-v0", "name": "Pythia 6.9B Deduped v0", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2022-12", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 6.857302016, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "source_type": "primary", "sources": [ { "label": "Original release documentation", "url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "type": "documentation" }, { "label": "Publisher v0 model card and size mapping", "url": "https://huggingface.co/EleutherAI/pythia-6.9b-deduped-v0", "type": "model_card" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/EleutherAI/pythia-6.9b-deduped-v0/blob/main/config.json", "type": "documentation" }, { "label": "Version history", "url": "https://github.com/EleutherAI/pythia#changelog", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:36:19.054Z", "last_seen": "2026-10-29T10:36:19.054Z", "upstream_present": false, "availability": "public", "notes": "Original completed training run, distinct from the corrected suite released April 3, 2023. The December 2022 publisher documentation links all 16 checkpoints and says each main branch is the completed 143000-step model. Originally named Pythia-6.7B-Deduped; later renamed to include embedding parameters. Total parameters use the publisher table, not tensor-storage totals that also include attention-mask buffers." }, { "id": "eleutherai/pythia-6.9b-v0", "slug": "eleutherai--pythia-6.9b-v0", "name": "Pythia 6.9B v0", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2022-12", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 6.857302016, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "source_type": "primary", "sources": [ { "label": "Original release documentation", "url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "type": "documentation" }, { "label": "Publisher v0 model card and size mapping", "url": "https://huggingface.co/EleutherAI/pythia-6.9b-v0", "type": "model_card" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/EleutherAI/pythia-6.9b-v0/blob/main/config.json", "type": "documentation" }, { "label": "Version history", "url": "https://github.com/EleutherAI/pythia#changelog", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:36:19.054Z", "last_seen": "2026-10-29T10:36:19.054Z", "upstream_present": false, "availability": "public", "notes": "Original completed training run, distinct from the corrected suite released April 3, 2023. The December 2022 publisher documentation links all 16 checkpoints and says each main branch is the completed 143000-step model. Originally named Pythia-6.7B; later renamed to include embedding parameters. Total parameters use the publisher table, not tensor-storage totals that also include attention-mask buffers." }, { "id": "eleutherai/pythia-70m-deduped-v0", "slug": "eleutherai--pythia-70m-deduped-v0", "name": "Pythia 70M Deduped v0", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2022-12", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.070426624, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "source_type": "primary", "sources": [ { "label": "Original release documentation", "url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "type": "documentation" }, { "label": "Publisher v0 model card and size mapping", "url": "https://huggingface.co/EleutherAI/pythia-70m-deduped-v0", "type": "model_card" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/EleutherAI/pythia-70m-deduped-v0/blob/main/config.json", "type": "documentation" }, { "label": "Version history", "url": "https://github.com/EleutherAI/pythia#changelog", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:36:19.054Z", "last_seen": "2026-10-29T10:36:19.054Z", "upstream_present": false, "availability": "public", "notes": "Original completed training run, distinct from the corrected suite released April 3, 2023. The December 2022 publisher documentation links all 16 checkpoints and says each main branch is the completed 143000-step model. Originally named Pythia-19M-Deduped; later renamed to include embedding parameters. Total parameters use the publisher table, not tensor-storage totals that also include attention-mask buffers." }, { "id": "eleutherai/pythia-70m-v0", "slug": "eleutherai--pythia-70m-v0", "name": "Pythia 70M v0", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "pythia", "description": null, "release_date": "2022-12", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.070426624, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "source_type": "primary", "sources": [ { "label": "Original release documentation", "url": "https://github.com/EleutherAI/pythia/blob/c6ec6964f87dec895dea8f0ab491243f8c6ca972/README.md", "type": "documentation" }, { "label": "Publisher v0 model card and size mapping", "url": "https://huggingface.co/EleutherAI/pythia-70m-v0", "type": "model_card" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/EleutherAI/pythia-70m-v0/blob/main/config.json", "type": "documentation" }, { "label": "Version history", "url": "https://github.com/EleutherAI/pythia#changelog", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:36:19.054Z", "last_seen": "2026-10-29T10:36:19.054Z", "upstream_present": false, "availability": "public", "notes": "Original completed training run, distinct from the corrected suite released April 3, 2023. The December 2022 publisher documentation links all 16 checkpoints and says each main branch is the completed 143000-step model. Originally named Pythia-19M; later renamed to include embedding parameters. Total parameters use the publisher table, not tensor-storage totals that also include attention-mask buffers." }, { "id": "openai/chatgpt-2022", "slug": "openai--chatgpt-2022", "name": "ChatGPT (GPT-3.5 research preview)", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-3.5", "description": "Original ChatGPT research preview, fine-tuned from GPT-3.5 using supervised dialogue training and reinforcement learning from human feedback.", "release_date": "2022-11-30", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://openai.com/index/chatgpt/", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://openai.com/index/chatgpt/", "type": "announcement" }, { "label": "Original conversation retention FAQ, December 2022 archive", "url": "https://web.archive.org/web/20221206132723id_/https://help.openai.com/en/articles/6787051-does-chatgpt-remember-what-happened-earlier-in-the-conversation", "type": "documentation" }, { "label": "GPT-4 report comparison of ChatGPT revisions", "url": "https://cdn.openai.com/papers/gpt-4.pdf", "type": "technical_report" }, { "label": "GPT-3.5 Turbo API launch and versioning", "url": "https://openai.com/index/introducing-chatgpt-and-whisper-apis/", "type": "announcement" }, { "label": "ChatGPT model-version history", "url": "https://help.openai.com/en/articles/6825453-chatgpt-release-notes", "type": "documentation" }, { "label": "OpenAI historical account of the original ChatGPT launch", "url": "https://cdn.openai.com/pdf/a253471f-8260-40c6-a2cc-aa93fe9f142e/economic-research-chatgpt-usage-paper.pdf", "type": "technical_report" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "Historical November 2022 research preview, not a callable API model ID. A December 2022 OpenAI FAQ described approximately 4000 tokens of retained conversation history; an exact context budget was not verified. GPT-3.5-series lineage does not identify a sized base checkpoint; total and active parameter counts remain undisclosed. Architecture is not inferred from InstructGPT or later GPT-3.5 Turbo versions.", "field_status": { "parameters_b": "undisclosed", "active_parameters_b": "undisclosed", "architecture": "unverified", "context_window": "unverified", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T16:11:08.404825Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/395.json" }, { "id": "together/gpt-jt-6b-v1", "slug": "together--gpt-jt-6b-v1", "name": "GPT-JT 6B v1", "developer": "Together AI", "developer_id": "together", "family": "gpt-jt", "description": null, "release_date": "2022-11-29", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 6, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/togethercomputer/GPT-JT-6B-v1", "source_type": "primary", "sources": [ { "label": "Together release announcement", "url": "https://www.together.ai/blog/releasing-v1-of-gpt-jt-powered-by-open-source-ai", "type": "primary" }, { "label": "Official model card", "url": "https://huggingface.co/togethercomputer/GPT-JT-6B-v1/blob/f34aa35f906895602c1f86f5685e598afdea8051/README.md", "type": "model_card" }, { "label": "Official configuration", "url": "https://huggingface.co/togethercomputer/GPT-JT-6B-v1/blob/f34aa35f906895602c1f86f5685e598afdea8051/config.json", "type": "config" } ], "aliases": [ "togethercomputer/GPT-JT-6B-v1" ], "first_seen": "2026-10-30T07:09:10.442Z", "last_seen": "2026-10-30T07:09:10.442Z", "upstream_present": false, "availability": "public", "notes": "Together’s November 29 announcement explicitly releases downloadable weights and a public demo. This company-trained GPT-J adaptation generates text as well as handling classification prompts. Parameter count is the publisher’s nominal total; context is verified in its configuration." }, { "id": "openai/text-davinci-003", "slug": "openai--text-davinci-003", "name": "text-davinci-003", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-3.5", "description": null, "release_date": "2022-11-28", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "unknown", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://x.com/janleike/status/1597355354433916928", "source_type": "primary", "sources": [ { "label": "Publisher release announcement", "url": "https://x.com/janleike/status/1597355354433916928", "type": "announcement" }, { "label": "Publisher historical model retirement list", "url": "https://openai.com/index/gpt-4-api-general-availability/", "type": "documentation" }, { "label": "Publisher team lead announcement through X syndication", "url": "https://cdn.syndication.twimg.com/tweet-result?id=1597355354433916928&lang=en&token=1", "type": "announcement" }, { "label": "Publisher deprecation history", "url": "https://developers.openai.com/api/docs/deprecations", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:38:48.065Z", "last_seen": "2026-10-29T10:38:48.065Z", "upstream_present": false, "availability": "public", "notes": "Historical OpenAI API model. Release is verified by the dated publisher announcement. Retired January 4, 2024. Exact model parameters and context window have not been verified from the reviewed primary sources; GPT-3 paper sizes are not assigned to API model aliases." }, { "id": "meta/galactica-1.3b", "slug": "meta--galactica-1.3b", "name": "Galactica 1.3B", "developer": "Meta", "developer_id": "meta", "family": "galactica", "description": null, "release_date": "2022-11-15", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-4.0", "parameters_b": 1.3, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/paperswithcode/galai/commit/3f6ef0e668dd9140b90ee8f7d8490c61c0c54b63", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://github.com/paperswithcode/galai/commit/3f6ef0e668dd9140b90ee8f7d8490c61c0c54b63", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/facebook/galactica-1.3b", "type": "model_card" }, { "label": "Model license", "url": "https://github.com/paperswithcode/galai/blob/main/LICENSE-MODEL.md", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T09:44:55.983Z", "last_seen": "2026-10-29T09:44:55.983Z", "upstream_present": false, "availability": "public", "notes": "The original release repository lists all five downloadable sizes. Parameters are the nominal counts published by the developer." }, { "id": "meta/galactica-120b", "slug": "meta--galactica-120b", "name": "Galactica 120B", "developer": "Meta", "developer_id": "meta", "family": "galactica", "description": null, "release_date": "2022-11-15", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-4.0", "parameters_b": 120, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/paperswithcode/galai/commit/3f6ef0e668dd9140b90ee8f7d8490c61c0c54b63", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://github.com/paperswithcode/galai/commit/3f6ef0e668dd9140b90ee8f7d8490c61c0c54b63", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/facebook/galactica-120b", "type": "model_card" }, { "label": "Model license", "url": "https://github.com/paperswithcode/galai/blob/main/LICENSE-MODEL.md", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T09:44:55.983Z", "last_seen": "2026-10-29T09:44:55.983Z", "upstream_present": false, "availability": "public", "notes": "The original release repository lists all five downloadable sizes. Parameters are the nominal counts published by the developer." }, { "id": "meta/galactica-125m", "slug": "meta--galactica-125m", "name": "Galactica 125M", "developer": "Meta", "developer_id": "meta", "family": "galactica", "description": null, "release_date": "2022-11-15", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-4.0", "parameters_b": 0.125, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/paperswithcode/galai/commit/3f6ef0e668dd9140b90ee8f7d8490c61c0c54b63", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://github.com/paperswithcode/galai/commit/3f6ef0e668dd9140b90ee8f7d8490c61c0c54b63", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/facebook/galactica-125m", "type": "model_card" }, { "label": "Model license", "url": "https://github.com/paperswithcode/galai/blob/main/LICENSE-MODEL.md", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T09:44:55.983Z", "last_seen": "2026-10-29T09:44:55.983Z", "upstream_present": false, "availability": "public", "notes": "The original release repository lists all five downloadable sizes. Parameters are the nominal counts published by the developer." }, { "id": "meta/galactica-30b", "slug": "meta--galactica-30b", "name": "Galactica 30B", "developer": "Meta", "developer_id": "meta", "family": "galactica", "description": null, "release_date": "2022-11-15", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-4.0", "parameters_b": 30, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/paperswithcode/galai/commit/3f6ef0e668dd9140b90ee8f7d8490c61c0c54b63", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://github.com/paperswithcode/galai/commit/3f6ef0e668dd9140b90ee8f7d8490c61c0c54b63", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/facebook/galactica-30b", "type": "model_card" }, { "label": "Model license", "url": "https://github.com/paperswithcode/galai/blob/main/LICENSE-MODEL.md", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T09:44:55.983Z", "last_seen": "2026-10-29T09:44:55.983Z", "upstream_present": false, "availability": "public", "notes": "The original release repository lists all five downloadable sizes. Parameters are the nominal counts published by the developer." }, { "id": "meta/galactica-6.7b", "slug": "meta--galactica-6.7b", "name": "Galactica 6.7B", "developer": "Meta", "developer_id": "meta", "family": "galactica", "description": null, "release_date": "2022-11-15", "date_precision": "day", "open_weights": true, "license": "CC-BY-NC-4.0", "parameters_b": 6.7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/paperswithcode/galai/commit/3f6ef0e668dd9140b90ee8f7d8490c61c0c54b63", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://github.com/paperswithcode/galai/commit/3f6ef0e668dd9140b90ee8f7d8490c61c0c54b63", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/facebook/galactica-6.7b", "type": "model_card" }, { "label": "Model license", "url": "https://github.com/paperswithcode/galai/blob/main/LICENSE-MODEL.md", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T09:44:55.983Z", "last_seen": "2026-10-29T09:44:55.983Z", "upstream_present": false, "availability": "public", "notes": "The original release repository lists all five downloadable sizes. Parameters are the nominal counts published by the developer." }, { "id": "cohere/command-xlarge-20221108", "slug": "cohere--command-xlarge-20221108", "name": "Command Xlarge 20221108", "developer": "Cohere", "developer_id": "cohere", "family": "command", "description": null, "release_date": "2022-11-08", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://docs.cohere.com/v1/changelog/improvements-to-current-models-new-beta-model-command", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://docs.cohere.com/v1/changelog/improvements-to-current-models-new-beta-model-command", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:10:00.648Z", "last_seen": "2026-10-29T10:10:00.648Z", "upstream_present": false, "availability": "public", "notes": "The dated API release notes explicitly introduce this version. The retired checkpoint’s parameters and context remain unverified." }, { "id": "cohere/medium-20221108", "slug": "cohere--medium-20221108", "name": "Medium 20221108", "developer": "Cohere", "developer_id": "cohere", "family": "cohere-generate", "description": null, "release_date": "2022-11-08", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://docs.cohere.com/v1/changelog/improvements-to-current-models-new-beta-model-command", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://docs.cohere.com/v1/changelog/improvements-to-current-models-new-beta-model-command", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:10:00.648Z", "last_seen": "2026-10-29T10:10:00.648Z", "upstream_present": false, "availability": "public", "notes": "The dated API release notes explicitly introduce this version. The retired checkpoint’s parameters and context remain unverified." }, { "id": "cohere/xlarge-20221108", "slug": "cohere--xlarge-20221108", "name": "Xlarge 20221108", "developer": "Cohere", "developer_id": "cohere", "family": "cohere-generate", "description": null, "release_date": "2022-11-08", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://docs.cohere.com/v1/changelog/improvements-to-current-models-new-beta-model-command", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://docs.cohere.com/v1/changelog/improvements-to-current-models-new-beta-model-command", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:10:00.648Z", "last_seen": "2026-10-29T10:10:00.648Z", "upstream_present": false, "availability": "public", "notes": "The dated API release notes explicitly introduce this version. The retired checkpoint’s parameters and context remain unverified." }, { "id": "amazon/alexatm-20b", "slug": "amazon--alexatm-20b", "name": "AlexaTM 20B", "developer": "Amazon", "developer_id": "amazon", "family": "alexatm", "description": null, "release_date": "2022-11", "date_precision": "month", "open_weights": true, "license": "Alexa Teacher Model License Agreement", "parameters_b": 20, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "encoder-decoder", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://aws.amazon.com/about-aws/whats-new/2022/11/alexatm-20b-model-available-sagemaker-jumpstart/", "source_type": "primary", "sources": [ { "label": "Publisher public-release announcement", "url": "https://aws.amazon.com/about-aws/whats-new/2022/11/alexatm-20b-model-available-sagemaker-jumpstart/", "type": "announcement" }, { "label": "Publisher model deployment and weight retrieval guide", "url": "https://aws.amazon.com/blogs/machine-learning/alexatm-20b-is-now-available-in-amazon-sagemaker-jumpstart/", "type": "documentation" }, { "label": "Publisher model code and description", "url": "https://github.com/amazon-science/alexa-teacher-models", "type": "documentation" }, { "label": "Model weights license", "url": "https://jumpstart-cache-prod-us-east-2.s3.us-east-2.amazonaws.com/licenses/Alexa20b-License/MODEL_LICENSE.md", "type": "license" }, { "label": "Original model research paper", "url": "https://arxiv.org/abs/2208.01448", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:45:04.786Z", "last_seen": "2026-10-29T10:45:04.786Z", "upstream_present": false, "availability": "public", "notes": "Amazon announces public SageMaker JumpStart access on November 18, while its deployment blog is dated November 17. Month precision avoids assigning an unsupported exact first day. The deployment guide retrieves model parameters for user-controlled hosting; the Alexa Teacher Model License Agreement permits copying/distributing model weights for non-commercial use. The August research paper is not treated as an external-release date." }, { "id": "bigscience/bloomz-1b1", "slug": "bigscience--bloomz-1b1", "name": "BLOOMZ-1b1", "developer": "BigScience", "developer_id": "bigscience", "family": "bloomz", "description": null, "release_date": "2022-11", "date_precision": "month", "open_weights": true, "license": "BigScience RAIL License v1.0", "parameters_b": 1.065314304, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigscience/bloomz-1b1", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/bigscience/bloomz-1b1", "type": "model_card" }, { "label": "Model configuration", "url": "https://huggingface.co/bigscience/bloomz-1b1/blob/main/config.json", "type": "documentation" }, { "label": "Model license", "url": "https://huggingface.co/spaces/bigscience/license", "type": "documentation" }, { "label": "Original project release", "url": "https://github.com/bigscience-workshop/xmtf/blob/1110fb0c39f2a3e34503fdc0134a139faa5e03af/README.md", "type": "documentation" }, { "label": "Original paper release", "url": "https://arxiv.org/abs/2211.01786v1", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:26:25.648Z", "last_seen": "2026-10-29T10:26:25.648Z", "upstream_present": false, "availability": "public", "notes": "Publisher-trained instruction variant of BigScience BLOOM. The November 3 primary paper explicitly releases the model family; accompanying individual cards appeared November 3–4. Month precision avoids inferring that earlier repository uploads were already public. The October 18 project README links only the 176B checkpoint. Trained on xP3 with English prompts." }, { "id": "bigscience/bloomz-1b7", "slug": "bigscience--bloomz-1b7", "name": "BLOOMZ-1b7", "developer": "BigScience", "developer_id": "bigscience", "family": "bloomz", "description": null, "release_date": "2022-11", "date_precision": "month", "open_weights": true, "license": "BigScience RAIL License v1.0", "parameters_b": 1.72240896, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigscience/bloomz-1b7", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/bigscience/bloomz-1b7", "type": "model_card" }, { "label": "Model configuration", "url": "https://huggingface.co/bigscience/bloomz-1b7/blob/main/config.json", "type": "documentation" }, { "label": "Model license", "url": "https://huggingface.co/spaces/bigscience/license", "type": "documentation" }, { "label": "Original project release", "url": "https://github.com/bigscience-workshop/xmtf/blob/1110fb0c39f2a3e34503fdc0134a139faa5e03af/README.md", "type": "documentation" }, { "label": "Original paper release", "url": "https://arxiv.org/abs/2211.01786v1", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:26:25.648Z", "last_seen": "2026-10-29T10:26:25.648Z", "upstream_present": false, "availability": "public", "notes": "Publisher-trained instruction variant of BigScience BLOOM. The November 3 primary paper explicitly releases the model family; accompanying individual cards appeared November 3–4. Month precision avoids inferring that earlier repository uploads were already public. The October 18 project README links only the 176B checkpoint. Trained on xP3 with English prompts." }, { "id": "bigscience/bloomz-3b", "slug": "bigscience--bloomz-3b", "name": "BLOOMZ-3b", "developer": "BigScience", "developer_id": "bigscience", "family": "bloomz", "description": null, "release_date": "2022-11", "date_precision": "month", "open_weights": true, "license": "BigScience RAIL License v1.0", "parameters_b": 3.00255744, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigscience/bloomz-3b", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/bigscience/bloomz-3b", "type": "model_card" }, { "label": "Model configuration", "url": "https://huggingface.co/bigscience/bloomz-3b/blob/main/config.json", "type": "documentation" }, { "label": "Model license", "url": "https://huggingface.co/spaces/bigscience/license", "type": "documentation" }, { "label": "Original project release", "url": "https://github.com/bigscience-workshop/xmtf/blob/1110fb0c39f2a3e34503fdc0134a139faa5e03af/README.md", "type": "documentation" }, { "label": "Original paper release", "url": "https://arxiv.org/abs/2211.01786v1", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:26:25.648Z", "last_seen": "2026-10-29T10:26:25.648Z", "upstream_present": false, "availability": "public", "notes": "Publisher-trained instruction variant of BigScience BLOOM. The November 3 primary paper explicitly releases the model family; accompanying individual cards appeared November 3–4. Month precision avoids inferring that earlier repository uploads were already public. The October 18 project README links only the 176B checkpoint. Trained on xP3 with English prompts." }, { "id": "bigscience/bloomz-560m", "slug": "bigscience--bloomz-560m", "name": "BLOOMZ-560m", "developer": "BigScience", "developer_id": "bigscience", "family": "bloomz", "description": null, "release_date": "2022-11", "date_precision": "month", "open_weights": true, "license": "BigScience RAIL License v1.0", "parameters_b": 0.559214592, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigscience/bloomz-560m", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/bigscience/bloomz-560m", "type": "model_card" }, { "label": "Model configuration", "url": "https://huggingface.co/bigscience/bloomz-560m/blob/main/config.json", "type": "documentation" }, { "label": "Model license", "url": "https://huggingface.co/spaces/bigscience/license", "type": "documentation" }, { "label": "Original project release", "url": "https://github.com/bigscience-workshop/xmtf/blob/1110fb0c39f2a3e34503fdc0134a139faa5e03af/README.md", "type": "documentation" }, { "label": "Original paper release", "url": "https://arxiv.org/abs/2211.01786v1", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:26:25.648Z", "last_seen": "2026-10-29T10:26:25.648Z", "upstream_present": false, "availability": "public", "notes": "Publisher-trained instruction variant of BigScience BLOOM. The November 3 primary paper explicitly releases the model family; accompanying individual cards appeared November 3–4. Month precision avoids inferring that earlier repository uploads were already public. The October 18 project README links only the 176B checkpoint. Trained on xP3 with English prompts." }, { "id": "bigscience/bloomz-7b1", "slug": "bigscience--bloomz-7b1", "name": "BLOOMZ-7b1", "developer": "BigScience", "developer_id": "bigscience", "family": "bloomz", "description": null, "release_date": "2022-11", "date_precision": "month", "open_weights": true, "license": "BigScience RAIL License v1.0", "parameters_b": 7.069016064, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigscience/bloomz-7b1", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/bigscience/bloomz-7b1", "type": "model_card" }, { "label": "Model configuration", "url": "https://huggingface.co/bigscience/bloomz-7b1/blob/main/config.json", "type": "documentation" }, { "label": "Model license", "url": "https://huggingface.co/spaces/bigscience/license", "type": "documentation" }, { "label": "Original project release", "url": "https://github.com/bigscience-workshop/xmtf/blob/1110fb0c39f2a3e34503fdc0134a139faa5e03af/README.md", "type": "documentation" }, { "label": "Original paper release", "url": "https://arxiv.org/abs/2211.01786v1", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:26:25.648Z", "last_seen": "2026-10-29T10:26:25.648Z", "upstream_present": false, "availability": "public", "notes": "Publisher-trained instruction variant of BigScience BLOOM. The November 3 primary paper explicitly releases the model family; accompanying individual cards appeared November 3–4. Month precision avoids inferring that earlier repository uploads were already public. The October 18 project README links only the 176B checkpoint. Trained on xP3 with English prompts." }, { "id": "bigscience/bloomz-7b1-mt", "slug": "bigscience--bloomz-7b1-mt", "name": "BLOOMZ-7b1-mt", "developer": "BigScience", "developer_id": "bigscience", "family": "bloomz", "description": null, "release_date": "2022-11", "date_precision": "month", "open_weights": true, "license": "BigScience RAIL License v1.0", "parameters_b": 7.069016064, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigscience/bloomz-7b1-mt", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/bigscience/bloomz-7b1-mt", "type": "model_card" }, { "label": "Model configuration", "url": "https://huggingface.co/bigscience/bloomz-7b1-mt/blob/main/config.json", "type": "documentation" }, { "label": "Model license", "url": "https://huggingface.co/spaces/bigscience/license", "type": "documentation" }, { "label": "Original project release", "url": "https://github.com/bigscience-workshop/xmtf/blob/1110fb0c39f2a3e34503fdc0134a139faa5e03af/README.md", "type": "documentation" }, { "label": "Original paper release", "url": "https://arxiv.org/abs/2211.01786v1", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:26:25.648Z", "last_seen": "2026-10-29T10:26:25.648Z", "upstream_present": false, "availability": "public", "notes": "Publisher-trained instruction variant of BigScience BLOOM. The November 3 primary paper explicitly releases the model family; accompanying individual cards appeared November 3–4. Month precision avoids inferring that earlier repository uploads were already public. The October 18 project README links only the 176B checkpoint. Trained on xP3mt with translated prompts." }, { "id": "bigscience/bloomz-7b1-p3", "slug": "bigscience--bloomz-7b1-p3", "name": "BLOOMZ-7b1-p3", "developer": "BigScience", "developer_id": "bigscience", "family": "bloomz", "description": null, "release_date": "2022-11", "date_precision": "month", "open_weights": true, "license": "BigScience RAIL License v1.0", "parameters_b": 7.069016064, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigscience/bloomz-7b1-p3", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/bigscience/bloomz-7b1-p3", "type": "model_card" }, { "label": "Model configuration", "url": "https://huggingface.co/bigscience/bloomz-7b1-p3/blob/main/config.json", "type": "documentation" }, { "label": "Model license", "url": "https://huggingface.co/spaces/bigscience/license", "type": "documentation" }, { "label": "Original project release", "url": "https://github.com/bigscience-workshop/xmtf/blob/1110fb0c39f2a3e34503fdc0134a139faa5e03af/README.md", "type": "documentation" }, { "label": "Original paper release", "url": "https://arxiv.org/abs/2211.01786v1", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:26:25.648Z", "last_seen": "2026-10-29T10:26:25.648Z", "upstream_present": false, "availability": "public", "notes": "Publisher-trained instruction variant of BigScience BLOOM. The November 3 primary paper explicitly releases the model family; accompanying individual cards appeared November 3–4. Month precision avoids inferring that earlier repository uploads were already public. The October 18 project README links only the 176B checkpoint. Trained on English-only P3; a distinct research checkpoint." }, { "id": "bigscience/bloomz-mt", "slug": "bigscience--bloomz-mt", "name": "BLOOMZ-mt", "developer": "BigScience", "developer_id": "bigscience", "family": "bloomz", "description": null, "release_date": "2022-11", "date_precision": "month", "open_weights": true, "license": "BigScience RAIL License v1.0", "parameters_b": 176.247271424, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigscience/bloomz-mt", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/bigscience/bloomz-mt", "type": "model_card" }, { "label": "Model configuration", "url": "https://huggingface.co/bigscience/bloomz-mt/blob/main/config.json", "type": "documentation" }, { "label": "Model license", "url": "https://huggingface.co/spaces/bigscience/license", "type": "documentation" }, { "label": "Original project release", "url": "https://github.com/bigscience-workshop/xmtf/blob/1110fb0c39f2a3e34503fdc0134a139faa5e03af/README.md", "type": "documentation" }, { "label": "Original paper release", "url": "https://arxiv.org/abs/2211.01786v1", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:26:25.648Z", "last_seen": "2026-10-29T10:26:25.648Z", "upstream_present": false, "availability": "public", "notes": "Publisher-trained instruction variant of BigScience BLOOM. The November 3 primary paper explicitly releases the model family; accompanying individual cards appeared November 3–4. Month precision avoids inferring that earlier repository uploads were already public. The October 18 project README links only the 176B checkpoint. Trained on xP3mt with translated prompts." }, { "id": "bigscience/bloomz-p3", "slug": "bigscience--bloomz-p3", "name": "BLOOMZ-p3", "developer": "BigScience", "developer_id": "bigscience", "family": "bloomz", "description": null, "release_date": "2022-11", "date_precision": "month", "open_weights": true, "license": "BigScience RAIL License v1.0", "parameters_b": 176.247271424, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigscience/bloomz-p3", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/bigscience/bloomz-p3", "type": "model_card" }, { "label": "Model configuration", "url": "https://huggingface.co/bigscience/bloomz-p3/blob/main/config.json", "type": "documentation" }, { "label": "Model license", "url": "https://huggingface.co/spaces/bigscience/license", "type": "documentation" }, { "label": "Original project release", "url": "https://github.com/bigscience-workshop/xmtf/blob/1110fb0c39f2a3e34503fdc0134a139faa5e03af/README.md", "type": "documentation" }, { "label": "Original paper release", "url": "https://arxiv.org/abs/2211.01786v1", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:26:25.648Z", "last_seen": "2026-10-29T10:26:25.648Z", "upstream_present": false, "availability": "public", "notes": "Publisher-trained instruction variant of BigScience BLOOM. The November 3 primary paper explicitly releases the model family; accompanying individual cards appeared November 3–4. Month precision avoids inferring that earlier repository uploads were already public. The October 18 project README links only the 176B checkpoint. Trained on English-only P3; a distinct research checkpoint." }, { "id": "bigscience/mt0-base", "slug": "bigscience--mt0-base", "name": "mT0-base", "developer": "BigScience", "developer_id": "bigscience", "family": "mt0", "description": null, "release_date": "2022-11", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.58, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "encoder-decoder", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigscience/mt0-base", "source_type": "primary", "sources": [ { "label": "Publisher model card and variant table", "url": "https://huggingface.co/bigscience/mt0-base/blob/0c500042c1f80b6b6ec208cf73d9ee43f1dac245/README.md", "type": "model_card" }, { "label": "Original November 2022 paper with public download links for all seven variants", "url": "https://arxiv.org/html/2211.01786v1", "type": "paper" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/bigscience/mt0-base/blob/0c500042c1f80b6b6ec208cf73d9ee43f1dac245/config.json", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T11:22:26.610Z", "last_seen": "2026-10-29T11:22:26.610Z", "upstream_present": false, "availability": "public", "notes": "BigScience institutional research release that instruction-tunes Google mT5. Trained on multilingual xP3 with English prompts. The November 3 paper states public availability and links this exact checkpoint. Month precision avoids treating earlier private uploads as release dates. Parameter count is the publisher’s nominal family-table value; supported context remains unverified." }, { "id": "bigscience/mt0-large", "slug": "bigscience--mt0-large", "name": "mT0-large", "developer": "BigScience", "developer_id": "bigscience", "family": "mt0", "description": null, "release_date": "2022-11", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 1.2, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "encoder-decoder", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigscience/mt0-large", "source_type": "primary", "sources": [ { "label": "Publisher model card and variant table", "url": "https://huggingface.co/bigscience/mt0-large/blob/c2cdb7dbabf83140e288e60f73a936ed9c5b9ed6/README.md", "type": "model_card" }, { "label": "Original November 2022 paper with public download links for all seven variants", "url": "https://arxiv.org/html/2211.01786v1", "type": "paper" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/bigscience/mt0-large/blob/c2cdb7dbabf83140e288e60f73a936ed9c5b9ed6/config.json", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T11:22:26.610Z", "last_seen": "2026-10-29T11:22:26.610Z", "upstream_present": false, "availability": "public", "notes": "BigScience institutional research release that instruction-tunes Google mT5. Trained on multilingual xP3 with English prompts. The November 3 paper states public availability and links this exact checkpoint. Month precision avoids treating earlier private uploads as release dates. Parameter count is the publisher’s nominal family-table value; supported context remains unverified." }, { "id": "bigscience/mt0-small", "slug": "bigscience--mt0-small", "name": "mT0-small", "developer": "BigScience", "developer_id": "bigscience", "family": "mt0", "description": null, "release_date": "2022-11", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.3, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "encoder-decoder", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigscience/mt0-small", "source_type": "primary", "sources": [ { "label": "Publisher model card and variant table", "url": "https://huggingface.co/bigscience/mt0-small/blob/8116a34237e19160ec003147e758f065876d95f0/README.md", "type": "model_card" }, { "label": "Original November 2022 paper with public download links for all seven variants", "url": "https://arxiv.org/html/2211.01786v1", "type": "paper" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/bigscience/mt0-small/blob/8116a34237e19160ec003147e758f065876d95f0/config.json", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T11:22:26.610Z", "last_seen": "2026-10-29T11:22:26.610Z", "upstream_present": false, "availability": "public", "notes": "BigScience institutional research release that instruction-tunes Google mT5. Trained on multilingual xP3 with English prompts. The November 3 paper states public availability and links this exact checkpoint. Month precision avoids treating earlier private uploads as release dates. Parameter count is the publisher’s nominal family-table value; supported context remains unverified." }, { "id": "bigscience/mt0-xl", "slug": "bigscience--mt0-xl", "name": "mT0-xl", "developer": "BigScience", "developer_id": "bigscience", "family": "mt0", "description": null, "release_date": "2022-11", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 3.7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "encoder-decoder", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigscience/mt0-xl", "source_type": "primary", "sources": [ { "label": "Publisher model card and variant table", "url": "https://huggingface.co/bigscience/mt0-xl/blob/7332741e6029c63e0226c0b5e81820941414c7fd/README.md", "type": "model_card" }, { "label": "Original November 2022 paper with public download links for all seven variants", "url": "https://arxiv.org/html/2211.01786v1", "type": "paper" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/bigscience/mt0-xl/blob/7332741e6029c63e0226c0b5e81820941414c7fd/config.json", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T11:22:26.610Z", "last_seen": "2026-10-29T11:22:26.610Z", "upstream_present": false, "availability": "public", "notes": "BigScience institutional research release that instruction-tunes Google mT5. Trained on multilingual xP3 with English prompts. The November 3 paper states public availability and links this exact checkpoint. Month precision avoids treating earlier private uploads as release dates. Parameter count is the publisher’s nominal family-table value; supported context remains unverified." }, { "id": "bigscience/mt0-xxl", "slug": "bigscience--mt0-xxl", "name": "mT0-xxl", "developer": "BigScience", "developer_id": "bigscience", "family": "mt0", "description": null, "release_date": "2022-11", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 13, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "encoder-decoder", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigscience/mt0-xxl", "source_type": "primary", "sources": [ { "label": "Publisher model card and variant table", "url": "https://huggingface.co/bigscience/mt0-xxl/blob/bedb470a4e99b00ff962660d49e5c037131ebcac/README.md", "type": "model_card" }, { "label": "Original November 2022 paper with public download links for all seven variants", "url": "https://arxiv.org/html/2211.01786v1", "type": "paper" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/bigscience/mt0-xxl/blob/bedb470a4e99b00ff962660d49e5c037131ebcac/config.json", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T11:22:26.610Z", "last_seen": "2026-10-29T11:22:26.610Z", "upstream_present": false, "availability": "public", "notes": "BigScience institutional research release that instruction-tunes Google mT5. Trained on multilingual xP3 with English prompts. The November 3 paper states public availability and links this exact checkpoint. Month precision avoids treating earlier private uploads as release dates. Parameter count is the publisher’s nominal family-table value; supported context remains unverified." }, { "id": "bigscience/mt0-xxl-mt", "slug": "bigscience--mt0-xxl-mt", "name": "mT0-xxl-mt", "developer": "BigScience", "developer_id": "bigscience", "family": "mt0", "description": null, "release_date": "2022-11", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 13, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "encoder-decoder", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigscience/mt0-xxl-mt", "source_type": "primary", "sources": [ { "label": "Publisher model card and variant table", "url": "https://huggingface.co/bigscience/mt0-xxl-mt/blob/db491aacb6a2601e3e8aac52214dbf6af7cd3f49/README.md", "type": "model_card" }, { "label": "Original November 2022 paper with public download links for all seven variants", "url": "https://arxiv.org/html/2211.01786v1", "type": "paper" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/bigscience/mt0-xxl-mt/blob/db491aacb6a2601e3e8aac52214dbf6af7cd3f49/config.json", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T11:22:26.610Z", "last_seen": "2026-10-29T11:22:26.610Z", "upstream_present": false, "availability": "public", "notes": "BigScience institutional research release that instruction-tunes Google mT5. Trained on xP3mt with translated prompts. The November 3 paper states public availability and links this exact checkpoint. Month precision avoids treating earlier private uploads as release dates. Parameter count is the publisher’s nominal family-table value; supported context remains unverified." }, { "id": "bigscience/mt0-xxl-p3", "slug": "bigscience--mt0-xxl-p3", "name": "mT0-xxl-p3", "developer": "BigScience", "developer_id": "bigscience", "family": "mt0", "description": null, "release_date": "2022-11", "date_precision": "month", "open_weights": true, "license": "Apache-2.0", "parameters_b": 13, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "encoder-decoder", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigscience/mt0-xxl-p3", "source_type": "primary", "sources": [ { "label": "Publisher model card and variant table", "url": "https://huggingface.co/bigscience/mt0-xxl-p3/blob/f3deb74cd2dd01ba8b14cb2e4328e705b24e500c/README.md", "type": "model_card" }, { "label": "Original November 2022 paper with public download links for all seven variants", "url": "https://arxiv.org/html/2211.01786v1", "type": "paper" }, { "label": "Publisher model configuration", "url": "https://huggingface.co/bigscience/mt0-xxl-p3/blob/f3deb74cd2dd01ba8b14cb2e4328e705b24e500c/config.json", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T11:22:26.610Z", "last_seen": "2026-10-29T11:22:26.610Z", "upstream_present": false, "availability": "public", "notes": "BigScience institutional research release that instruction-tunes Google mT5. Trained on English-only P3. The November 3 paper states public availability and links this exact checkpoint. Month precision avoids treating earlier private uploads as release dates. Parameter count is the publisher’s nominal family-table value; supported context remains unverified." }, { "id": "google/flan-t5-base", "slug": "google--flan-t5-base", "name": "Flan-T5 BASE", "developer": "Google", "developer_id": "google", "family": "flan-t5", "description": null, "release_date": "2022-10-20", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.247577856, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/google-research/t5x/commit/a998a4814f9f5413630cdcb5067483b9c12252c6", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://github.com/google-research/t5x/commit/a998a4814f9f5413630cdcb5067483b9c12252c6", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/google/flan-t5-base", "type": "model_card" }, { "label": "Checkpoint parameter metadata", "url": "https://huggingface.co/api/models/google/flan-t5-base", "type": "parameters" } ], "aliases": [], "first_seen": "2026-10-29T09:44:55.983Z", "last_seen": "2026-10-29T09:44:55.983Z", "upstream_present": false, "availability": "public", "notes": "Date follows the original T5X commit publishing the checkpoint links, before the Hugging Face conversion. Parameter count is the unique safetensors total reported for the official checkpoint." }, { "id": "google/flan-t5-large", "slug": "google--flan-t5-large", "name": "Flan-T5 LARGE", "developer": "Google", "developer_id": "google", "family": "flan-t5", "description": null, "release_date": "2022-10-20", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.78315008, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/google-research/t5x/commit/a998a4814f9f5413630cdcb5067483b9c12252c6", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://github.com/google-research/t5x/commit/a998a4814f9f5413630cdcb5067483b9c12252c6", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/google/flan-t5-large", "type": "model_card" }, { "label": "Checkpoint parameter metadata", "url": "https://huggingface.co/api/models/google/flan-t5-large", "type": "parameters" } ], "aliases": [], "first_seen": "2026-10-29T09:44:55.983Z", "last_seen": "2026-10-29T09:44:55.983Z", "upstream_present": false, "availability": "public", "notes": "Date follows the original T5X commit publishing the checkpoint links, before the Hugging Face conversion. Parameter count is the unique safetensors total reported for the official checkpoint." }, { "id": "google/flan-t5-small", "slug": "google--flan-t5-small", "name": "Flan-T5 SMALL", "developer": "Google", "developer_id": "google", "family": "flan-t5", "description": null, "release_date": "2022-10-20", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.076961152, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/google-research/t5x/commit/a998a4814f9f5413630cdcb5067483b9c12252c6", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://github.com/google-research/t5x/commit/a998a4814f9f5413630cdcb5067483b9c12252c6", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/google/flan-t5-small", "type": "model_card" }, { "label": "Checkpoint parameter metadata", "url": "https://huggingface.co/api/models/google/flan-t5-small", "type": "parameters" } ], "aliases": [], "first_seen": "2026-10-29T09:44:55.983Z", "last_seen": "2026-10-29T09:44:55.983Z", "upstream_present": false, "availability": "public", "notes": "Date follows the original T5X commit publishing the checkpoint links, before the Hugging Face conversion. Parameter count is the unique safetensors total reported for the official checkpoint." }, { "id": "google/flan-t5-xl", "slug": "google--flan-t5-xl", "name": "Flan-T5 XL", "developer": "Google", "developer_id": "google", "family": "flan-t5", "description": null, "release_date": "2022-10-20", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 2.849757184, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/google-research/t5x/commit/a998a4814f9f5413630cdcb5067483b9c12252c6", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://github.com/google-research/t5x/commit/a998a4814f9f5413630cdcb5067483b9c12252c6", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/google/flan-t5-xl", "type": "model_card" }, { "label": "Checkpoint parameter metadata", "url": "https://huggingface.co/api/models/google/flan-t5-xl", "type": "parameters" } ], "aliases": [], "first_seen": "2026-10-29T09:44:55.983Z", "last_seen": "2026-10-29T09:44:55.983Z", "upstream_present": false, "availability": "public", "notes": "Date follows the original T5X commit publishing the checkpoint links, before the Hugging Face conversion. Parameter count is the unique safetensors total reported for the official checkpoint." }, { "id": "google/flan-t5-xxl", "slug": "google--flan-t5-xxl", "name": "Flan-T5 XXL", "developer": "Google", "developer_id": "google", "family": "flan-t5", "description": null, "release_date": "2022-10-20", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 11.26692864, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/google-research/t5x/commit/a998a4814f9f5413630cdcb5067483b9c12252c6", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://github.com/google-research/t5x/commit/a998a4814f9f5413630cdcb5067483b9c12252c6", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/google/flan-t5-xxl", "type": "model_card" }, { "label": "Checkpoint parameter metadata", "url": "https://huggingface.co/api/models/google/flan-t5-xxl", "type": "parameters" } ], "aliases": [], "first_seen": "2026-10-29T09:44:55.983Z", "last_seen": "2026-10-29T09:44:55.983Z", "upstream_present": false, "availability": "public", "notes": "Date follows the original T5X commit publishing the checkpoint links, before the Hugging Face conversion. Parameter count is the unique safetensors total reported for the official checkpoint." }, { "id": "bigscience/bloomz", "slug": "bigscience--bloomz", "name": "BLOOMZ", "developer": "BigScience", "developer_id": "bigscience", "family": "bloomz", "description": null, "release_date": "2022-10-18", "date_precision": "day", "open_weights": true, "license": "BigScience RAIL License v1.0", "parameters_b": 176.247271424, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigscience/bloomz", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/bigscience/bloomz", "type": "model_card" }, { "label": "Model configuration", "url": "https://huggingface.co/bigscience/bloomz/blob/main/config.json", "type": "documentation" }, { "label": "Model license", "url": "https://huggingface.co/spaces/bigscience/license", "type": "documentation" }, { "label": "Original project release", "url": "https://github.com/bigscience-workshop/xmtf/blob/1110fb0c39f2a3e34503fdc0134a139faa5e03af/README.md", "type": "documentation" }, { "label": "Original paper release", "url": "https://arxiv.org/abs/2211.01786v1", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:26:25.648Z", "last_seen": "2026-10-29T10:26:25.648Z", "upstream_present": false, "availability": "public", "notes": "First publisher documentation linking the 176B checkpoint is the October 18 original xmtf README, before the November 3 paper. Earlier repository upload timestamps are not used as release evidence. Publisher-trained xP3 instruction variant of BigScience BLOOM." }, { "id": "openbmb/cpm-ant-plus-10b", "slug": "openbmb--cpm-ant-plus-10b", "name": "CPM-Ant+ 10B", "developer": "OpenBMB", "developer_id": "openbmb", "family": "cpm", "description": null, "release_date": "2022-10-12", "date_precision": "day", "open_weights": true, "license": null, "parameters_b": 10, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/OpenBMB/CPM-Live", "source_type": "primary", "sources": [ { "label": "Publisher dated release log", "url": "https://github.com/OpenBMB/CPM-Live", "type": "documentation" }, { "label": "Publisher model description and checkpoint link", "url": "https://github.com/OpenBMB/CPM-Live/blob/cpm-ant-plus/cpm-live/README.MD", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T10:45:04.786Z", "last_seen": "2026-10-29T10:45:04.786Z", "upstream_present": false, "availability": "public", "notes": "Original publisher 10B bilingual checkpoint, continued from its own CPM-Ant. The developer states an open model license, but the exact license text/title was not recovered. Context window remains unverified." }, { "id": "zhipuai/codegeex-13b", "slug": "zhipuai--codegeex-13b", "name": "CodeGeeX 13B", "developer": "Zhipu AI", "developer_id": "zhipuai", "family": "codegeex", "description": null, "release_date": "2022-09-30", "date_precision": "day", "open_weights": true, "license": "CodeGeeX License", "parameters_b": 13, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/zai-org/CodeGeeX", "source_type": "primary", "sources": [ { "label": "Publisher release log and model description", "url": "https://github.com/zai-org/CodeGeeX", "type": "documentation" }, { "label": "Publisher configuration", "url": "https://github.com/zai-org/CodeGeeX/blob/main/configs/codegeex_13b.sh", "type": "documentation" }, { "label": "Model weights license", "url": "https://github.com/zai-org/CodeGeeX/blob/main/MODEL_LICENSE", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:38:50.966Z", "last_seen": "2026-10-29T10:38:50.966Z", "upstream_present": false, "availability": "public", "notes": "Publisher release log dates public source code and model weights to September 30, 2022. The model license permits non-commercial research; the repository Apache-2.0 code license does not license model weights. Parameters are the publisher nominal count." }, { "id": "openbmb/cpm-ant-10b", "slug": "openbmb--cpm-ant-10b", "name": "CPM-Ant 10B", "developer": "OpenBMB", "developer_id": "openbmb", "family": "cpm", "description": null, "release_date": "2022-09", "date_precision": "month", "open_weights": true, "license": "General Model License — Attribution and Publicity Restriction", "parameters_b": 10, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/OpenBMB/CPM-Live", "source_type": "primary", "sources": [ { "label": "Publisher dated release log", "url": "https://github.com/OpenBMB/CPM-Live", "type": "documentation" }, { "label": "Publisher model description and checkpoint link", "url": "https://github.com/OpenBMB/CPM-Live/blob/cpm-ant/cpm-live/README.MD", "type": "model_card" }, { "label": "Publisher explicit model-license link", "url": "https://medium.com/@openbmb/cpm-ant-the-first-big-model-in-cpm-live-project-369dacd2236b", "type": "blog" }, { "label": "Publisher-linked model license", "url": "https://github.com/OpenBMB/General-Model-License/blob/main/%E9%80%9A%E7%94%A8%E6%A8%A1%E5%9E%8B%E8%AE%B8%E5%8F%AF%E5%8D%8F%E8%AE%AE-%E6%9D%A5%E6%BA%90%E8%AF%B4%E6%98%8E-%E5%AE%A3%E4%BC%A0%E9%99%90%E5%88%B6.md", "type": "license" }, { "label": "Earlier September 15 checkpoint documentation", "url": "https://github.com/OpenBMB/CPM-Live/blob/03425aa319f63b36bf4e10622f988636710105d1/cpm-live/README.MD", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:45:04.786Z", "last_seen": "2026-10-29T10:45:04.786Z", "upstream_present": false, "availability": "public", "notes": "September 15 checkpoint documentation and September 16 formal release log establish month precision. Exact model license recovered through publisher OpenBMB article; context remains unverified." }, { "id": "openbmb/cpm-ant-1b", "slug": "openbmb--cpm-ant-1b", "name": "CPM-Ant 1B", "developer": "OpenBMB", "developer_id": "openbmb", "family": "cpm", "description": null, "release_date": "2022-09", "date_precision": "month", "open_weights": true, "license": "General Model License — Attribution and Publicity Restriction", "parameters_b": 1, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/OpenBMB/CPM-Live/blob/f98c0f7368a9bf299be9f5c633861a3ecefd9787/cpm-live/README.MD", "source_type": "primary", "sources": [ { "label": "Dated publisher final-checkpoint table", "url": "https://github.com/OpenBMB/CPM-Live/blob/f98c0f7368a9bf299be9f5c633861a3ecefd9787/cpm-live/README.MD", "type": "documentation" }, { "label": "Earlier September 15 checkpoint documentation", "url": "https://github.com/OpenBMB/CPM-Live/blob/03425aa319f63b36bf4e10622f988636710105d1/cpm-live/README.MD", "type": "documentation" }, { "label": "Publisher release log", "url": "https://github.com/OpenBMB/CPM-Live", "type": "documentation" }, { "label": "Publisher learned-pruning and further-pretraining explanation", "url": "https://medium.com/@openbmb/cpm-ant-the-first-big-model-in-cpm-live-project-369dacd2236b", "type": "blog" }, { "label": "Publisher-linked model license", "url": "https://github.com/OpenBMB/General-Model-License/blob/main/%E9%80%9A%E7%94%A8%E6%A8%A1%E5%9E%8B%E8%AE%B8%E5%8F%AF%E5%8D%8F%E8%AE%AE-%E6%9D%A5%E6%BA%90%E8%AF%B4%E6%98%8E-%E5%AE%A3%E4%BC%A0%E9%99%90%E5%88%B6.md", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:57:14.174Z", "last_seen": "2026-10-29T10:57:14.174Z", "upstream_present": false, "availability": "public", "notes": "Publisher-trained structural variant of its own CPM-Ant 10B. OpenBMB learns pruning masks, removes layers/heads/feedforward dimensions and further pretrains the smaller model; this is not a numerical-precision conversion. September 15 README already links these final checkpoints while the release log uses September 16, so month precision is retained. Nominal size is from the publisher checkpoint table; context remains unverified." }, { "id": "openbmb/cpm-ant-300m", "slug": "openbmb--cpm-ant-300m", "name": "CPM-Ant 300M", "developer": "OpenBMB", "developer_id": "openbmb", "family": "cpm", "description": null, "release_date": "2022-09", "date_precision": "month", "open_weights": true, "license": "General Model License — Attribution and Publicity Restriction", "parameters_b": 0.3, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/OpenBMB/CPM-Live/blob/f98c0f7368a9bf299be9f5c633861a3ecefd9787/cpm-live/README.MD", "source_type": "primary", "sources": [ { "label": "Dated publisher final-checkpoint table", "url": "https://github.com/OpenBMB/CPM-Live/blob/f98c0f7368a9bf299be9f5c633861a3ecefd9787/cpm-live/README.MD", "type": "documentation" }, { "label": "Earlier September 15 checkpoint documentation", "url": "https://github.com/OpenBMB/CPM-Live/blob/03425aa319f63b36bf4e10622f988636710105d1/cpm-live/README.MD", "type": "documentation" }, { "label": "Publisher release log", "url": "https://github.com/OpenBMB/CPM-Live", "type": "documentation" }, { "label": "Publisher learned-pruning and further-pretraining explanation", "url": "https://medium.com/@openbmb/cpm-ant-the-first-big-model-in-cpm-live-project-369dacd2236b", "type": "blog" }, { "label": "Publisher-linked model license", "url": "https://github.com/OpenBMB/General-Model-License/blob/main/%E9%80%9A%E7%94%A8%E6%A8%A1%E5%9E%8B%E8%AE%B8%E5%8F%AF%E5%8D%8F%E8%AE%AE-%E6%9D%A5%E6%BA%90%E8%AF%B4%E6%98%8E-%E5%AE%A3%E4%BC%A0%E9%99%90%E5%88%B6.md", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:57:14.174Z", "last_seen": "2026-10-29T10:57:14.174Z", "upstream_present": false, "availability": "public", "notes": "Publisher-trained structural variant of its own CPM-Ant 10B. OpenBMB learns pruning masks, removes layers/heads/feedforward dimensions and further pretrains the smaller model; this is not a numerical-precision conversion. September 15 README already links these final checkpoints while the release log uses September 16, so month precision is retained. Nominal size is from the publisher checkpoint table; context remains unverified." }, { "id": "openbmb/cpm-ant-3b", "slug": "openbmb--cpm-ant-3b", "name": "CPM-Ant 3B", "developer": "OpenBMB", "developer_id": "openbmb", "family": "cpm", "description": null, "release_date": "2022-09", "date_precision": "month", "open_weights": true, "license": "General Model License — Attribution and Publicity Restriction", "parameters_b": 3, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/OpenBMB/CPM-Live/blob/f98c0f7368a9bf299be9f5c633861a3ecefd9787/cpm-live/README.MD", "source_type": "primary", "sources": [ { "label": "Dated publisher final-checkpoint table", "url": "https://github.com/OpenBMB/CPM-Live/blob/f98c0f7368a9bf299be9f5c633861a3ecefd9787/cpm-live/README.MD", "type": "documentation" }, { "label": "Earlier September 15 checkpoint documentation", "url": "https://github.com/OpenBMB/CPM-Live/blob/03425aa319f63b36bf4e10622f988636710105d1/cpm-live/README.MD", "type": "documentation" }, { "label": "Publisher release log", "url": "https://github.com/OpenBMB/CPM-Live", "type": "documentation" }, { "label": "Publisher learned-pruning and further-pretraining explanation", "url": "https://medium.com/@openbmb/cpm-ant-the-first-big-model-in-cpm-live-project-369dacd2236b", "type": "blog" }, { "label": "Publisher-linked model license", "url": "https://github.com/OpenBMB/General-Model-License/blob/main/%E9%80%9A%E7%94%A8%E6%A8%A1%E5%9E%8B%E8%AE%B8%E5%8F%AF%E5%8D%8F%E8%AE%AE-%E6%9D%A5%E6%BA%90%E8%AF%B4%E6%98%8E-%E5%AE%A3%E4%BC%A0%E9%99%90%E5%88%B6.md", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:57:14.174Z", "last_seen": "2026-10-29T10:57:14.174Z", "upstream_present": false, "availability": "public", "notes": "Publisher-trained structural variant of its own CPM-Ant 10B. OpenBMB learns pruning masks, removes layers/heads/feedforward dimensions and further pretrains the smaller model; this is not a numerical-precision conversion. September 15 README already links these final checkpoints while the release log uses September 16, so month precision is retained. Nominal size is from the publisher checkpoint table; context remains unverified." }, { "id": "openbmb/cpm-ant-7b", "slug": "openbmb--cpm-ant-7b", "name": "CPM-Ant 7B", "developer": "OpenBMB", "developer_id": "openbmb", "family": "cpm", "description": null, "release_date": "2022-09", "date_precision": "month", "open_weights": true, "license": "General Model License — Attribution and Publicity Restriction", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/OpenBMB/CPM-Live/blob/f98c0f7368a9bf299be9f5c633861a3ecefd9787/cpm-live/README.MD", "source_type": "primary", "sources": [ { "label": "Dated publisher final-checkpoint table", "url": "https://github.com/OpenBMB/CPM-Live/blob/f98c0f7368a9bf299be9f5c633861a3ecefd9787/cpm-live/README.MD", "type": "documentation" }, { "label": "Earlier September 15 checkpoint documentation", "url": "https://github.com/OpenBMB/CPM-Live/blob/03425aa319f63b36bf4e10622f988636710105d1/cpm-live/README.MD", "type": "documentation" }, { "label": "Publisher release log", "url": "https://github.com/OpenBMB/CPM-Live", "type": "documentation" }, { "label": "Publisher learned-pruning and further-pretraining explanation", "url": "https://medium.com/@openbmb/cpm-ant-the-first-big-model-in-cpm-live-project-369dacd2236b", "type": "blog" }, { "label": "Publisher-linked model license", "url": "https://github.com/OpenBMB/General-Model-License/blob/main/%E9%80%9A%E7%94%A8%E6%A8%A1%E5%9E%8B%E8%AE%B8%E5%8F%AF%E5%8D%8F%E8%AE%AE-%E6%9D%A5%E6%BA%90%E8%AF%B4%E6%98%8E-%E5%AE%A3%E4%BC%A0%E9%99%90%E5%88%B6.md", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:57:14.174Z", "last_seen": "2026-10-29T10:57:14.174Z", "upstream_present": false, "availability": "public", "notes": "Publisher-trained structural variant of its own CPM-Ant 10B. OpenBMB learns pruning masks, removes layers/heads/feedforward dimensions and further pretrains the smaller model; this is not a numerical-precision conversion. September 15 README already links these final checkpoints while the release log uses September 16, so month precision is retained. Nominal size is from the publisher checkpoint table; context remains unverified." }, { "id": "microsoft/biogpt", "slug": "microsoft--biogpt", "name": "BioGPT", "developer": "Microsoft", "developer_id": "microsoft", "family": "biogpt", "description": null, "release_date": "2022-08-29", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 0.347, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 1024, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/microsoft/BioGPT/commit/00d0421fbae0379a741df2e00c97b56b1b0b1eaf", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://github.com/microsoft/BioGPT/commit/00d0421fbae0379a741df2e00c97b56b1b0b1eaf", "type": "announcement" }, { "label": "Developer technical report", "url": "https://arxiv.org/abs/2210.10341", "type": "paper" }, { "label": "Official model card", "url": "https://huggingface.co/microsoft/biogpt", "type": "model_card" }, { "label": "Model configuration", "url": "https://huggingface.co/microsoft/biogpt/blob/main/config.json", "type": "config" } ], "aliases": [], "first_seen": "2026-10-29T10:10:00.648Z", "last_seen": "2026-10-29T10:10:00.648Z", "upstream_present": false, "availability": "public", "notes": "The August 29 repository revision provides the pretrained checkpoint download; the later paper and Transformers conversion are not used as its first release. Nominal parameter count; trained from scratch on biomedical literature." }, { "id": "meta/blenderbot-3-175b", "slug": "meta--blenderbot-3-175b", "name": "BlenderBot 3 175B", "developer": "Meta", "developer_id": "meta", "family": "blenderbot", "description": null, "release_date": "2022-08-05", "date_precision": "day", "open_weights": false, "license": "BB3 License Agreement", "parameters_b": 175, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://about.fb.com/news/2022/08/blenderbot-ai-chatbot-improves-through-conversation/", "source_type": "primary", "sources": [ { "label": "Publisher public chatbot launch", "url": "https://about.fb.com/news/2022/08/blenderbot-ai-chatbot-improves-through-conversation/", "type": "announcement" }, { "label": "Publisher checkpoint access instructions", "url": "https://github.com/facebookresearch/ParlAI/tree/main/projects/bb3", "type": "documentation" }, { "label": "Publisher 175B model card and training configuration", "url": "https://github.com/facebookresearch/ParlAI/blob/main/parlai/zoo/bb3/model_card.md", "type": "model_card" }, { "label": "Model weights license", "url": "https://github.com/facebookresearch/ParlAI/blob/main/parlai/zoo/bb3/MODEL_LICENSE.md", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:45:04.786Z", "last_seen": "2026-10-29T10:45:04.786Z", "upstream_present": false, "availability": "public", "notes": "Meta-trained OPT-175B variant underlying the public BlenderBot 3 research demo. The 175B weights are shared only after application approval, so this record is classified closed weights. 2048 is the model card training/inference truncation, not a claim about long-term external memory." }, { "id": "meta/blenderbot-3-30b", "slug": "meta--blenderbot-3-30b", "name": "BlenderBot 3 30B", "developer": "Meta", "developer_id": "meta", "family": "blenderbot", "description": null, "release_date": "2022-08-05", "date_precision": "day", "open_weights": true, "license": "BB3 License Agreement", "parameters_b": 30, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/facebookresearch/ParlAI/blob/b1acb681207559da56a787ba96e16f0e23697d92/projects/bb3/README.md", "source_type": "primary", "sources": [ { "label": "Original dated checkpoint release", "url": "https://github.com/facebookresearch/ParlAI/blob/b1acb681207559da56a787ba96e16f0e23697d92/projects/bb3/README.md", "type": "documentation" }, { "label": "Original model paper and size-specific specifications", "url": "https://arxiv.org/html/2208.03188v1", "type": "paper" }, { "label": "Model weights license", "url": "https://github.com/facebookresearch/ParlAI/blob/main/parlai/zoo/bb3/MODEL_LICENSE.md", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:57:14.174Z", "last_seen": "2026-10-29T10:57:14.174Z", "upstream_present": false, "availability": "public", "notes": "Original publisher variant of Meta OPT-30B; the paper identifies a 2048-token context. The original August 5 release commit provides direct 30B checkpoint download instructions. BB3 performs its different modules using one trained transformer with control codes, so it is not an ensemble of different language models. Parameters are nominal." }, { "id": "meta/blenderbot-3-3b", "slug": "meta--blenderbot-3-3b", "name": "BlenderBot 3 3B", "developer": "Meta", "developer_id": "meta", "family": "blenderbot", "description": null, "release_date": "2022-08-05", "date_precision": "day", "open_weights": true, "license": "BB3 License Agreement", "parameters_b": 3, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "encoder-decoder", "context_window": 1024, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/facebookresearch/ParlAI/blob/b1acb681207559da56a787ba96e16f0e23697d92/projects/bb3/README.md", "source_type": "primary", "sources": [ { "label": "Original dated checkpoint release", "url": "https://github.com/facebookresearch/ParlAI/blob/b1acb681207559da56a787ba96e16f0e23697d92/projects/bb3/README.md", "type": "documentation" }, { "label": "Original model paper and size-specific specifications", "url": "https://arxiv.org/html/2208.03188v1", "type": "paper" }, { "label": "Model weights license", "url": "https://github.com/facebookresearch/ParlAI/blob/main/parlai/zoo/bb3/MODEL_LICENSE.md", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:57:14.174Z", "last_seen": "2026-10-29T10:57:14.174Z", "upstream_present": false, "availability": "public", "notes": "Original publisher variant of Meta R2C2; the paper explicitly identifies the 3B encoder-decoder model and 1024-token training context. This is BlenderBot version 3 at 3B size, not the older Hugging Face facebook/blenderbot-3B checkpoint. BB3 performs its different modules using one trained transformer with control codes, so it is not an ensemble of different language models. Parameters are nominal." }, { "id": "zhipuai/glm-130b", "slug": "zhipuai--glm-130b", "name": "GLM-130B", "developer": "Z.ai", "developer_id": "zhipuai", "family": "glm", "description": null, "release_date": "2022-08-04", "date_precision": "day", "open_weights": true, "license": "GLM-130B License", "parameters_b": 130, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://keg.cs.tsinghua.edu.cn/glm-130b/posts/glm-130b/", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://keg.cs.tsinghua.edu.cn/glm-130b/posts/glm-130b/", "type": "announcement" }, { "label": "Original project and weights", "url": "https://github.com/THUDM/GLM-130B", "type": "weights" }, { "label": "Model license", "url": "https://github.com/THUDM/GLM-130B/blob/main/MODEL_LICENSE", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T09:45:45.650Z", "last_seen": "2026-10-29T09:45:45.650Z", "upstream_present": false, "availability": "public", "notes": "Developed by Tsinghua KEG and Zhipu AI. Nominal parameter count. The model license restricts use to non-commercial research." }, { "id": "bigscience/bloom-7b1", "slug": "bigscience--bloom-7b1", "name": "BLOOM-7b1", "developer": "BigScience", "developer_id": "bigscience", "family": "bloom", "description": null, "release_date": "2022-08", "date_precision": "month", "open_weights": true, "license": "BigScience RAIL License v1.0", "parameters_b": 7.069016064, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigscience/bloom-7b1", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/bigscience/bloom-7b1", "type": "model_card" }, { "label": "Model configuration", "url": "https://huggingface.co/bigscience/bloom-7b1/blob/main/config.json", "type": "documentation" }, { "label": "Model license", "url": "https://huggingface.co/spaces/bigscience/license", "type": "documentation" }, { "label": "Primary repository history", "url": "https://huggingface.co/bigscience/bloom-7b1/commits/main", "type": "documentation" }, { "label": "Publisher completed-checkpoint release", "url": "https://huggingface.co/bigscience/bloom-7b1/discussions/12", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:26:25.648Z", "last_seen": "2026-10-29T10:26:25.648Z", "upstream_present": false, "availability": "public", "notes": "The completed 341B-token checkpoint was proposed August 5 and merged August 6. Earlier May weights were explicitly intermediary and are excluded. Month precision reflects the public branch versus main-branch publication." }, { "id": "bigscience/bloom-3b", "slug": "bigscience--bloom-3b", "name": "BLOOM-3b", "developer": "BigScience", "developer_id": "bigscience", "family": "bloom", "description": null, "release_date": "2022-07-13", "date_precision": "day", "open_weights": true, "license": "BigScience RAIL License v1.0", "parameters_b": 3.00255744, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigscience/bloom-3b", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/bigscience/bloom-3b", "type": "model_card" }, { "label": "Model configuration", "url": "https://huggingface.co/bigscience/bloom-3b/blob/main/config.json", "type": "documentation" }, { "label": "Model license", "url": "https://huggingface.co/spaces/bigscience/license", "type": "documentation" }, { "label": "Primary repository history", "url": "https://huggingface.co/bigscience/bloom-3b/commits/main", "type": "documentation" }, { "label": "Publisher completed-checkpoint release", "url": "https://huggingface.co/bigscience/bloom-3b/discussions/7", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:26:25.648Z", "last_seen": "2026-10-29T10:26:25.648Z", "upstream_present": false, "availability": "public", "notes": "The completed 341B-token checkpoint was published and merged July 13, and the intermediary warning removed. Earlier May training checkpoints are excluded." }, { "id": "bigscience/bloom", "slug": "bigscience--bloom", "name": "BLOOM 176B", "developer": "BigScience", "developer_id": "bigscience", "family": "bloom", "description": "Multilingual pretrained model from the BigScience collaboration.", "release_date": "2022-07-12", "date_precision": "day", "open_weights": true, "license": "BigScience BLOOM RAIL 1.0", "parameters_b": 176.247271424, "active_parameters_b": 176.247271424, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": false, "tool_call": false, "source_url": "https://huggingface.co/blog/bloom", "source_type": "primary", "sources": [ { "label": "BLOOM model card and license", "url": "https://huggingface.co/bigscience/bloom", "type": "weights" }, { "label": "BLOOM technical report", "url": "https://arxiv.org/html/2211.05100v4", "type": "paper" }, { "label": "BLOOM checkpoint configuration", "url": "https://huggingface.co/bigscience/bloom/raw/main/config.json", "type": "config" }, { "label": "BigScience clarification of active parameters", "url": "https://huggingface.co/bigscience/bloom/discussions/284", "type": "documentation" }, { "label": "BLOOM tensor index and shard metadata", "url": "https://huggingface.co/bigscience/bloom/raw/main/model.safetensors.index.json", "type": "weights" }, { "label": "BLOOM tokenizer configuration", "url": "https://huggingface.co/bigscience/bloom/raw/main/tokenizer_config.json", "type": "config" }, { "label": "Original release source", "url": "https://huggingface.co/blog/bloom", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "Public launch July 12, 2022; card’s July 11 date is an estimate. Exact total is 176,247,271,424 learned parameters with tied embeddings counted once. BigScience explicitly states all parameters are active. Context is the 2,048-token training sequence. Reasoning/tool flags describe dedicated native modes in the base release, not prompted behavior or downstream wrappers.", "field_status": {}, "audited_at": "2026-10-26T17:40:28.264301+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/127.json" }, { "id": "bigscience/bloom-1b1", "slug": "bigscience--bloom-1b1", "name": "BLOOM-1b1", "developer": "BigScience", "developer_id": "bigscience", "family": "bloom", "description": null, "release_date": "2022-07", "date_precision": "month", "open_weights": true, "license": "BigScience RAIL License v1.0", "parameters_b": 1.065314304, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigscience/bloom-1b1", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/bigscience/bloom-1b1", "type": "model_card" }, { "label": "Model configuration", "url": "https://huggingface.co/bigscience/bloom-1b1/blob/main/config.json", "type": "documentation" }, { "label": "Model license", "url": "https://huggingface.co/spaces/bigscience/license", "type": "documentation" }, { "label": "Primary repository history", "url": "https://huggingface.co/bigscience/bloom-1b1/commits/main", "type": "documentation" }, { "label": "Publisher completed-checkpoint release", "url": "https://huggingface.co/bigscience/bloom-1b1/discussions/8", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:26:25.648Z", "last_seen": "2026-10-29T10:26:25.648Z", "upstream_present": false, "availability": "public", "notes": "The publisher says the model just finished training in the July 22 completed 341B-token checkpoint proposal, merged July 25. May intermediary weights are excluded. Month precision reflects the public branch versus main-branch publication." }, { "id": "bigscience/bloom-560m", "slug": "bigscience--bloom-560m", "name": "BLOOM-560m", "developer": "BigScience", "developer_id": "bigscience", "family": "bloom", "description": null, "release_date": "2022-07", "date_precision": "month", "open_weights": true, "license": "BigScience RAIL License v1.0", "parameters_b": 0.559214592, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigscience/bloom-560m", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/bigscience/bloom-560m", "type": "model_card" }, { "label": "Model configuration", "url": "https://huggingface.co/bigscience/bloom-560m/blob/main/config.json", "type": "documentation" }, { "label": "Model license", "url": "https://huggingface.co/spaces/bigscience/license", "type": "documentation" }, { "label": "Primary repository history", "url": "https://huggingface.co/bigscience/bloom-560m/commits/main", "type": "documentation" }, { "label": "Publisher completed-checkpoint release", "url": "https://huggingface.co/bigscience/bloom-560m/discussions/5", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:26:25.648Z", "last_seen": "2026-10-29T10:26:25.648Z", "upstream_present": false, "availability": "public", "notes": "The completed 341B-token checkpoint was proposed July 2 and merged July 21. The earlier May checkpoint was explicitly intermediary and is excluded from the release date. Month precision reflects the public branch versus main-branch publication." }, { "id": "yandex/yalm-100b", "slug": "yandex--yalm-100b", "name": "YaLM 100B", "developer": "Yandex", "developer_id": "yandex", "family": "yalm", "description": null, "release_date": "2022-06-23", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 100, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ir.yandex/press-releases?id=2022-23-06&year=2022", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://ir.yandex/press-releases?id=2022-23-06&year=2022", "type": "announcement" }, { "label": "Original model repository", "url": "https://github.com/yandex/YaLM-100B", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-29T10:10:00.648Z", "last_seen": "2026-10-29T10:10:00.648Z", "upstream_present": false, "availability": "public", "notes": "Nominal parameter count in the publisher announcement. The repository identifies an independently trained Russian and English language model." }, { "id": "meta/opt-66b", "slug": "meta--opt-66b", "name": "OPT 66B", "developer": "Meta", "developer_id": "meta", "family": "opt", "description": null, "release_date": "2022-06-22", "date_precision": "day", "open_weights": true, "license": "OPT Model License", "parameters_b": 66, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/facebookresearch/metaseq/commit/84c412fbe99da7518b0c1ed7a3379106cd7459b3", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://github.com/facebookresearch/metaseq/commit/84c412fbe99da7518b0c1ed7a3379106cd7459b3", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/facebook/opt-66b", "type": "model_card" }, { "label": "Model license", "url": "https://github.com/facebookresearch/metaseq/blob/main/projects/OPT/MODEL_LICENSE.md", "type": "license" }, { "label": "Original OPT announcement", "url": "https://ai.meta.com/blog/democratizing-access-to-large-scale-language-models-with-opt-175b/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:10:00.648Z", "last_seen": "2026-10-29T10:10:00.648Z", "upstream_present": false, "availability": "public", "notes": "The 66B checkpoint was explicitly pending in the original May release; the June 22 commit publishes its downloads. Nominal parameter count." }, { "id": "google/ul2", "slug": "google--ul2", "name": "UL2 20B", "developer": "Google", "developer_id": "google", "family": "ul2", "description": null, "release_date": "2022-05-03", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 20, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 512, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/google-research/google-research/commit/f967fdda3c73b522d1a3965b106af43449b2bba8", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://github.com/google-research/google-research/commit/f967fdda3c73b522d1a3965b106af43449b2bba8", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/google/ul2", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-29T09:45:45.650Z", "last_seen": "2026-10-29T09:45:45.650Z", "upstream_present": false, "availability": "public", "notes": "The initial repository commit publishes configuration and pretrained checkpoint locations. The paper and later October blog are not the first checkpoint publication. Size is nominal; 512 is the model-card input sequence length." }, { "id": "meta/opt-1.3b", "slug": "meta--opt-1.3b", "name": "OPT 1.3B", "developer": "Meta", "developer_id": "meta", "family": "opt", "description": "Open Pretrained Transformer research checkpoint.", "release_date": "2022-05-03", "date_precision": "day", "open_weights": true, "license": "OPT non-commercial research license", "parameters_b": 1.31575808, "active_parameters_b": 1.3, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/democratizing-access-to-large-scale-language-models-with-opt-175b/", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://ai.meta.com/blog/democratizing-access-to-large-scale-language-models-with-opt-175b/", "type": "announcement" }, { "label": "OPT technical report", "url": "https://arxiv.org/html/2205.01068v4", "type": "primary" }, { "label": "Official OPT model downloads", "url": "https://raw.githubusercontent.com/facebookresearch/metaseq/main/projects/OPT/README.md", "type": "weights" }, { "label": "Exact OPT-1.3B configuration", "url": "https://huggingface.co/facebook/opt-1.3b/raw/3f5c25d0bc631cb57ac65913f76e22c2dfb61d62/config.json", "type": "primary" }, { "label": "Exact OPT-1.3B checkpoint metadata", "url": "https://huggingface.co/facebook/opt-1.3b/resolve/3f5c25d0bc631cb57ac65913f76e22c2dfb61d62/pytorch_model.bin", "type": "weights" }, { "label": "OPT parameter architecture implementation", "url": "https://raw.githubusercontent.com/huggingface/transformers/v4.21.0/src/transformers/models/opt/modeling_opt.py", "type": "primary" }, { "label": "OPT model license", "url": "https://huggingface.co/facebook/opt-1.3b/raw/main/LICENSE.md", "type": "primary" }, { "label": "MoE++ original benchmark active-parameter convention (Table 4)", "url": "https://proceedings.iclr.cc/paper_files/paper/2025/file/7efe88bb4138d602e56637cfcf713654-Paper-Conference.pdf", "type": "primary" }, { "label": "OPT paper", "url": "https://arxiv.org/abs/2205.01068", "type": "paper" }, { "label": "Official OPT repository", "url": "https://github.com/facebookresearch/metaseq", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "OPT-1.3B is the nominal size name; the exact checkpoint has 1,315,758,080 unique parameters, counting tied token/output embeddings once and including positional embeddings, biases and layer norms. The nominal 1.3B active count follows Jin et al. / Skywork AI's MoE++ Table 4 benchmark convention; it is not a Meta specification or exact per-token tensor count. This is a pretrained text-completion checkpoint with no documented native reasoning mode or tool-call protocol. The May 3, 2022 public release predates its May 11 Hugging Face repository creation. Weights use the custom OPT noncommercial research license.", "field_status": { "reasoning": "undisclosed", "tool_call": "undisclosed" }, "audited_at": "2026-10-27T14:30:49.607Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/288.json" }, { "id": "meta/opt-125m", "slug": "meta--opt-125m", "name": "OPT 125M", "developer": "Meta", "developer_id": "meta", "family": "opt", "description": "Meta's OPT-125M pretrained dense decoder-only Transformer for text generation.", "release_date": "2022-05-03", "date_precision": "day", "open_weights": true, "license": "OPT non-commercial research license", "parameters_b": 0.125239296, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/democratizing-access-to-large-scale-language-models-with-opt-175b/", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://ai.meta.com/blog/democratizing-access-to-large-scale-language-models-with-opt-175b/", "type": "announcement" }, { "label": "OPT: Open Pre-trained Transformer Language Models, Table 1 and Sections 2 and D", "url": "https://arxiv.org/html/2205.01068v4", "type": "primary" }, { "label": "facebook/opt-125m configuration", "url": "https://huggingface.co/facebook/opt-125m/blob/27dcfa74d334bc871f3234de431e71c6eeba5dd6/config.json", "type": "weights" }, { "label": "facebook/opt-125m model card", "url": "https://huggingface.co/facebook/opt-125m", "type": "weights" }, { "label": "OPT-125M distributed weight license", "url": "https://huggingface.co/facebook/opt-125m/raw/main/LICENSE.md", "type": "weights" }, { "label": "Transformers v4.21.0 OPT implementation", "url": "https://raw.githubusercontent.com/huggingface/transformers/v4.21.0/src/transformers/models/opt/modeling_opt.py", "type": "primary" }, { "label": "Transformers v4.21.0 OPT configuration defaults", "url": "https://raw.githubusercontent.com/huggingface/transformers/v4.21.0/src/transformers/models/opt/configuration_opt.py", "type": "primary" }, { "label": "Official OPT pretrained checkpoint listing", "url": "https://raw.githubusercontent.com/facebookresearch/metaseq/main/projects/OPT/README.md", "type": "weights" }, { "label": "Hugging Face metadata for facebook/opt-125m", "url": "https://huggingface.co/api/models/facebook/opt-125m", "type": "catalog" }, { "label": "Meta metaseq Transformer decoder reference implementation", "url": "https://raw.githubusercontent.com/facebookresearch/metaseq/main/metaseq/models/transformer_decoder.py", "type": "primary" }, { "label": "Transformers v4.21.0 shared configuration defaults", "url": "https://raw.githubusercontent.com/huggingface/transformers/v4.21.0/src/transformers/configuration_utils.py", "type": "primary" }, { "label": "Transformers v4.21.0 weight tying implementation", "url": "https://raw.githubusercontent.com/huggingface/transformers/v4.21.0/src/transformers/modeling_utils.py", "type": "primary" }, { "label": "µ-MoE: Test-Time Pruning as Micro-Grained Mixture-of-Experts, Table 1 and Section 3", "url": "https://www.merl.com/publications/docs/TR2025-112.pdf", "type": "primary" }, { "label": "Neurons in Large Language Models: Dead, N-gram, Positional", "url": "https://arxiv.org/pdf/2309.04827", "type": "primary" }, { "label": "MoE++: Accelerating Mixture-of-Experts Methods with Zero-Computation Experts, Table 4", "url": "https://proceedings.iclr.cc/paper_files/paper/2025/file/7efe88bb4138d602e56637cfcf713654-Paper-Conference.pdf", "type": "primary" }, { "label": "OPT paper", "url": "https://arxiv.org/abs/2205.01068", "type": "paper" }, { "label": "Official OPT repository", "url": "https://github.com/facebookresearch/metaseq", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "Meta publishes a nominal 125M size. The retained 125,239,296 total is derived from the official configuration and reference implementation, with tied embeddings counted once; it is not checkpoint-header verified. No separate complete-model active count was established; published FFN activation statistics and pruning percentages do not provide one. This is a pretrained text model with no documented native reasoning mode or tool-call protocol. Weights use the custom non-commercial research license. Original release: May 3, 2022; HF repository creation: May 11, 2022.", "field_status": { "active_parameters_b": "undisclosed", "reasoning": "undisclosed", "tool_call": "undisclosed" }, "audited_at": "2026-10-27T14:32:36.898281Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/289.json" }, { "id": "meta/opt-13b", "slug": "meta--opt-13b", "name": "OPT 13B", "developer": "Meta", "developer_id": "meta", "family": "opt", "description": "Open Pretrained Transformer research checkpoint.", "release_date": "2022-05-03", "date_precision": "day", "open_weights": true, "license": "OPT non-commercial research license", "parameters_b": 12.85347328, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/democratizing-access-to-large-scale-language-models-with-opt-175b/", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://ai.meta.com/blog/democratizing-access-to-large-scale-language-models-with-opt-175b/", "type": "announcement" }, { "label": "OPT technical report", "url": "https://arxiv.org/html/2205.01068v4", "type": "primary" }, { "label": "Official OPT pretrained weights", "url": "https://raw.githubusercontent.com/facebookresearch/metaseq/main/projects/OPT/README.md", "type": "weights" }, { "label": "Exact OPT-13B configuration", "url": "https://huggingface.co/facebook/opt-13b/blob/e515202d1e7750da62d245fbccb2723b9c1790f5/config.json", "type": "weights" }, { "label": "Exact OPT-13B tensor index", "url": "https://huggingface.co/facebook/opt-13b/blob/e515202d1e7750da62d245fbccb2723b9c1790f5/pytorch_model.bin.index.json", "type": "weights" }, { "label": "OPT parameter implementation", "url": "https://github.com/huggingface/transformers/blob/v4.21.0/src/transformers/models/opt/modeling_opt.py", "type": "primary" }, { "label": "Meta OPT shared-embedding training configuration", "url": "https://github.com/facebookresearch/metaseq/blob/main/metaseq/launcher/opt_baselines.py", "type": "primary" }, { "label": "Meta OPT decoder implementation", "url": "https://github.com/facebookresearch/metaseq/blob/main/metaseq/models/transformer_decoder.py", "type": "primary" }, { "label": "OPT-13B weight license", "url": "https://huggingface.co/facebook/opt-13b/blob/e515202d1e7750da62d245fbccb2723b9c1790f5/LICENSE.md", "type": "primary" }, { "label": "μ-MoE OPT pruning comparison", "url": "https://www.merl.com/publications/docs/TR2025-112.pdf", "type": "primary" }, { "label": "MoE++ activated-parameter comparison", "url": "https://proceedings.iclr.cc/paper_files/paper/2025/file/7efe88bb4138d602e56637cfcf713654-Paper-Conference.pdf", "type": "primary" }, { "label": "GRIFFIN OPT pruning comparison", "url": "https://arxiv.org/html/2404.01365v3", "type": "primary" }, { "label": "OPT paper", "url": "https://arxiv.org/abs/2205.01068", "type": "paper" }, { "label": "Official OPT repository", "url": "https://github.com/facebookresearch/metaseq", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "Meta publishes OPT-13B as nominally 13B parameters; the audited configuration and implementation reconcile 12.85347328B unique parameters, counting tied embeddings once. A separate active-parameter count remains unestablished; published pruning ratios concern modified linear layers. Dedicated reasoning and tool-call capabilities are not separately documented. Weights use the OPT non-commercial research license.", "field_status": { "active_parameters_b": "undisclosed", "reasoning": "undisclosed", "tool_call": "undisclosed" }, "audited_at": "2026-10-27T14:34:09.143Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/290.json" }, { "id": "meta/opt-175b", "slug": "meta--opt-175b", "name": "OPT 175B", "developer": "Meta", "developer_id": "meta", "family": "opt", "description": "Open Pretrained Transformer research checkpoint.", "release_date": "2022-05-03", "date_precision": "day", "open_weights": true, "license": "OPT non-commercial research license", "parameters_b": 175, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/democratizing-access-to-large-scale-language-models-with-opt-175b/", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://ai.meta.com/blog/democratizing-access-to-large-scale-language-models-with-opt-175b/", "type": "announcement" }, { "label": "OPT: Open Pre-trained Transformer Language Models", "url": "https://arxiv.org/html/2205.01068v4", "type": "primary" }, { "label": "Applications of Meta’s OPT-175B: Protein design, quantum physics, translation, and more", "url": "https://ai.meta.com/blog/opt-175b-large-language-model-applications/", "type": "primary" }, { "label": "OPT-175B License Agreement", "url": "https://raw.githubusercontent.com/facebookresearch/metaseq/main/projects/OPT/MODEL_LICENSE.md", "type": "primary" }, { "label": "OPT pretrained model weights", "url": "https://raw.githubusercontent.com/facebookresearch/metaseq/main/projects/OPT/README.md", "type": "primary" }, { "label": "Release-day OPT model-size configuration", "url": "https://raw.githubusercontent.com/facebookresearch/metaseq/3591201ebc1c668d34aef6679a1792a07b5edf35/metaseq/launcher/opt_job_constants.py", "type": "primary" }, { "label": "Release-day OPT launcher", "url": "https://raw.githubusercontent.com/facebookresearch/metaseq/3591201ebc1c668d34aef6679a1792a07b5edf35/metaseq/launcher/opt_baselines.py", "type": "primary" }, { "label": "Release-day tensor-parallel language model", "url": "https://raw.githubusercontent.com/facebookresearch/metaseq/3591201ebc1c668d34aef6679a1792a07b5edf35/metaseq/model_parallel/models/transformer_lm.py", "type": "primary" }, { "label": "Release-day Transformer decoder implementation", "url": "https://raw.githubusercontent.com/facebookresearch/metaseq/3591201ebc1c668d34aef6679a1792a07b5edf35/metaseq/models/transformer.py", "type": "primary" }, { "label": "Release-day Transformer layer implementation", "url": "https://raw.githubusercontent.com/facebookresearch/metaseq/3591201ebc1c668d34aef6679a1792a07b5edf35/metaseq/modules/transformer_layer.py", "type": "primary" }, { "label": "Release-day model-parallel attention implementation", "url": "https://raw.githubusercontent.com/facebookresearch/metaseq/3591201ebc1c668d34aef6679a1792a07b5edf35/metaseq/model_parallel/modules/multihead_attention.py", "type": "primary" }, { "label": "Release-day positional embedding implementation", "url": "https://raw.githubusercontent.com/facebookresearch/metaseq/3591201ebc1c668d34aef6679a1792a07b5edf35/metaseq/modules/positional_embedding.py", "type": "primary" }, { "label": "Release-day OPT tokenizer vocabulary", "url": "https://raw.githubusercontent.com/facebookresearch/metaseq/3591201ebc1c668d34aef6679a1792a07b5edf35/projects/OPT/assets/gpt2-vocab.json", "type": "primary" }, { "label": "Release-day language-model dictionary loading", "url": "https://raw.githubusercontent.com/facebookresearch/metaseq/3591201ebc1c668d34aef6679a1792a07b5edf35/metaseq/tasks/language_modeling.py", "type": "primary" }, { "label": "175B Training Run Completed", "url": "https://raw.githubusercontent.com/facebookresearch/metaseq/main/projects/OPT/chronicles/final_update.md", "type": "primary" }, { "label": "Accessing OPT-175B", "url": "https://raw.githubusercontent.com/facebookresearch/metaseq/main/projects/OPT/download_opt175b.md", "type": "primary" }, { "label": "Release-day model-parallel FFN implementation", "url": "https://raw.githubusercontent.com/facebookresearch/metaseq/3591201ebc1c668d34aef6679a1792a07b5edf35/metaseq/model_parallel/modules/transformer_layer.py", "type": "primary" }, { "label": "OPT model card, version 1.0.0", "url": "https://raw.githubusercontent.com/facebookresearch/metaseq/main/projects/OPT/model_card.md", "type": "primary" }, { "label": "Deja Vu: Contextual Sparsity for Efficient LLMs at Inference Time", "url": "https://proceedings.mlr.press/v202/liu23am/liu23am.pdf", "type": "primary" }, { "label": "MoE++: Accelerating Mixture-of-Experts Methods with Zero-Computation Experts", "url": "https://proceedings.iclr.cc/paper_files/paper/2025/file/7efe88bb4138d602e56637cfcf713654-Paper-Conference.pdf", "type": "primary" }, { "label": "OPT paper", "url": "https://arxiv.org/abs/2205.01068", "type": "paper" }, { "label": "Official OPT repository", "url": "https://github.com/facebookresearch/metaseq", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "Meta publishes a nominal total of 175B parameters; a separate active-parameter count for OPT-175B was not established. Meta documents May 3, 2022 as the request-based research release, with the first individual access-grant date unresolved and actual grants confirmed by July 26. Weights were distributed to approved researchers under non-commercial terms. Native reasoning mode and tool calling are not separately disclosed.", "field_status": { "active_parameters_b": "undisclosed", "reasoning": "undisclosed", "tool_call": "undisclosed" }, "audited_at": "2026-10-27T14:34:39.709654Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/291.json" }, { "id": "meta/opt-2.7b", "slug": "meta--opt-2.7b", "name": "OPT 2.7B", "developer": "Meta", "developer_id": "meta", "family": "opt", "description": null, "release_date": "2022-05-03", "date_precision": "day", "open_weights": true, "license": "OPT Model License", "parameters_b": 2.7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/facebookresearch/metaseq/commit/f2cd36798793604cf51ab8b8a2cb167c964f9667", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://github.com/facebookresearch/metaseq/commit/f2cd36798793604cf51ab8b8a2cb167c964f9667", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/facebook/opt-2.7b", "type": "model_card" }, { "label": "Model license", "url": "https://github.com/facebookresearch/metaseq/blob/main/projects/OPT/MODEL_LICENSE.md", "type": "license" }, { "label": "Original OPT announcement", "url": "https://ai.meta.com/blog/democratizing-access-to-large-scale-language-models-with-opt-175b/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:10:00.648Z", "last_seen": "2026-10-29T10:10:00.648Z", "upstream_present": false, "availability": "public", "notes": "The initial repository lists this downloadable checkpoint; date follows the May 3 public family announcement. Nominal parameter count." }, { "id": "meta/opt-30b", "slug": "meta--opt-30b", "name": "OPT 30B", "developer": "Meta", "developer_id": "meta", "family": "opt", "description": null, "release_date": "2022-05-03", "date_precision": "day", "open_weights": true, "license": "OPT Model License", "parameters_b": 30, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/facebookresearch/metaseq/commit/f2cd36798793604cf51ab8b8a2cb167c964f9667", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://github.com/facebookresearch/metaseq/commit/f2cd36798793604cf51ab8b8a2cb167c964f9667", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/facebook/opt-30b", "type": "model_card" }, { "label": "Model license", "url": "https://github.com/facebookresearch/metaseq/blob/main/projects/OPT/MODEL_LICENSE.md", "type": "license" }, { "label": "Original OPT announcement", "url": "https://ai.meta.com/blog/democratizing-access-to-large-scale-language-models-with-opt-175b/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:10:00.648Z", "last_seen": "2026-10-29T10:10:00.648Z", "upstream_present": false, "availability": "public", "notes": "The initial repository lists this downloadable checkpoint; date follows the May 3 public family announcement. Nominal parameter count." }, { "id": "meta/opt-350m", "slug": "meta--opt-350m", "name": "OPT 350M", "developer": "Meta", "developer_id": "meta", "family": "opt", "description": null, "release_date": "2022-05-03", "date_precision": "day", "open_weights": true, "license": "OPT Model License", "parameters_b": 0.35, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/facebookresearch/metaseq/commit/f2cd36798793604cf51ab8b8a2cb167c964f9667", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://github.com/facebookresearch/metaseq/commit/f2cd36798793604cf51ab8b8a2cb167c964f9667", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/facebook/opt-350m", "type": "model_card" }, { "label": "Model license", "url": "https://github.com/facebookresearch/metaseq/blob/main/projects/OPT/MODEL_LICENSE.md", "type": "license" }, { "label": "Original OPT announcement", "url": "https://ai.meta.com/blog/democratizing-access-to-large-scale-language-models-with-opt-175b/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:10:00.648Z", "last_seen": "2026-10-29T10:10:00.648Z", "upstream_present": false, "availability": "public", "notes": "The initial repository lists this downloadable checkpoint; date follows the May 3 public family announcement. Nominal parameter count." }, { "id": "meta/opt-6.7b", "slug": "meta--opt-6.7b", "name": "OPT 6.7B", "developer": "Meta", "developer_id": "meta", "family": "opt", "description": "Open Pretrained Transformer research checkpoint.", "release_date": "2022-05-03", "date_precision": "day", "open_weights": true, "license": "OPT non-commercial research license", "parameters_b": 6.658473984, "active_parameters_b": 6.66, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://ai.meta.com/blog/democratizing-access-to-large-scale-language-models-with-opt-175b/", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://ai.meta.com/blog/democratizing-access-to-large-scale-language-models-with-opt-175b/", "type": "announcement" }, { "label": "OPT technical report", "url": "https://arxiv.org/html/2205.01068v4", "type": "primary" }, { "label": "Official OPT checkpoint configuration", "url": "https://huggingface.co/facebook/opt-6.7b/raw/a45aa65bbeb77c1558bc99bedc6779195462dab0/config.json", "type": "weights" }, { "label": "Official OPT checkpoint index", "url": "https://huggingface.co/facebook/opt-6.7b/raw/a45aa65bbeb77c1558bc99bedc6779195462dab0/pytorch_model.bin.index.json", "type": "weights" }, { "label": "OPT learned tensor definitions", "url": "https://raw.githubusercontent.com/huggingface/transformers/v4.21.0/src/transformers/models/opt/modeling_opt.py", "type": "primary" }, { "label": "OPT model license", "url": "https://raw.githubusercontent.com/facebookresearch/metaseq/main/projects/OPT/MODEL_LICENSE.md", "type": "primary" }, { "label": "MIDAS original comparison: OPT-6.7B activated parameters, Table 8", "url": "https://openreview.net/pdf?id=O0BRg90v1Y", "type": "primary" }, { "label": "OPT paper", "url": "https://arxiv.org/abs/2205.01068", "type": "paper" }, { "label": "Official OPT repository", "url": "https://github.com/facebookresearch/metaseq", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "OPT-6.7B is Meta’s nominal size label; the pinned official checkpoint reconciles to 6.658473984B learned parameters, counting tied embeddings once. MIDAS Table 8 reports 6.66B activated under its research comparison convention, including embeddings; this is rounded and is not a separate Meta disclosure. Public research weights use non-commercial terms.", "field_status": { "reasoning": "undisclosed", "tool_call": "undisclosed" }, "audited_at": "2026-10-27T15:23:03.163066Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/292.json" }, { "id": "bigscience/bloom-1b7", "slug": "bigscience--bloom-1b7", "name": "BLOOM-1b7", "developer": "BigScience", "developer_id": "bigscience", "family": "bloom", "description": null, "release_date": "2022-05", "date_precision": "month", "open_weights": true, "license": "BigScience RAIL License v1.0", "parameters_b": 1.72240896, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigscience/bloom-1b7", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/bigscience/bloom-1b7", "type": "model_card" }, { "label": "Model configuration", "url": "https://huggingface.co/bigscience/bloom-1b7/blob/main/config.json", "type": "documentation" }, { "label": "Model license", "url": "https://huggingface.co/spaces/bigscience/license", "type": "documentation" }, { "label": "Primary repository history", "url": "https://huggingface.co/bigscience/bloom-1b7/commits/main", "type": "documentation" }, { "label": "Original available checkpoint documentation", "url": "https://huggingface.co/bigscience/bloom-560m/blob/6a83bd49e2a5350051265a6f454069e28ae8d88a/README.md", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:26:25.648Z", "last_seen": "2026-10-29T10:26:25.648Z", "upstream_present": false, "availability": "public", "notes": "May 2022 publisher documentation recommends this already fully trained checkpoint under its earlier BLOOM-1B3 name. The 1.7B name includes embeddings. First availability is recorded at month precision; the copied July 11 estimate on the current card refers to the later main BLOOM launch." }, { "id": "salesforce/codegen-16b-multi", "slug": "salesforce--codegen-16b-multi", "name": "CodeGen 16B MULTI", "developer": "Salesforce", "developer_id": "salesforce", "family": "codegen", "description": null, "release_date": "2022-04-01", "date_precision": "day", "open_weights": true, "license": "BSD-3-Clause", "parameters_b": 16, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/salesforce/CodeGen/commit/2c8c869b34ce18ce22ddb51fa7448c1ab5e7ddd2", "source_type": "primary", "sources": [ { "label": "Publisher adds the remaining NL and MULTI checkpoints", "url": "https://github.com/salesforce/CodeGen/commit/2c8c869b34ce18ce22ddb51fa7448c1ab5e7ddd2", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Salesforce/codegen-16B-multi", "type": "model_card" }, { "label": "Original release announcement", "url": "https://www.salesforce.com/blog/codegen/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T09:44:55.983Z", "last_seen": "2026-10-29T09:44:55.983Z", "upstream_present": false, "availability": "public", "notes": "The April 1 publisher commit explicitly adds the remaining NL and MULTI checkpoints. The March 29 release enabled only MONO; its generic family description did not establish availability of this variant. Parameter count is nominal." }, { "id": "salesforce/codegen-16b-nl", "slug": "salesforce--codegen-16b-nl", "name": "CodeGen 16B NL", "developer": "Salesforce", "developer_id": "salesforce", "family": "codegen", "description": null, "release_date": "2022-04-01", "date_precision": "day", "open_weights": true, "license": "BSD-3-Clause", "parameters_b": 16, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/salesforce/CodeGen/commit/2c8c869b34ce18ce22ddb51fa7448c1ab5e7ddd2", "source_type": "primary", "sources": [ { "label": "Publisher adds the remaining NL and MULTI checkpoints", "url": "https://github.com/salesforce/CodeGen/commit/2c8c869b34ce18ce22ddb51fa7448c1ab5e7ddd2", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Salesforce/codegen-16B-nl", "type": "model_card" }, { "label": "Original release announcement", "url": "https://www.salesforce.com/blog/codegen/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T09:44:55.983Z", "last_seen": "2026-10-29T09:44:55.983Z", "upstream_present": false, "availability": "public", "notes": "The April 1 publisher commit explicitly adds the remaining NL and MULTI checkpoints. The March 29 release enabled only MONO; its generic family description did not establish availability of this variant. Parameter count is nominal." }, { "id": "salesforce/codegen-2b-multi", "slug": "salesforce--codegen-2b-multi", "name": "CodeGen 2B MULTI", "developer": "Salesforce", "developer_id": "salesforce", "family": "codegen", "description": null, "release_date": "2022-04-01", "date_precision": "day", "open_weights": true, "license": "BSD-3-Clause", "parameters_b": 2, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/salesforce/CodeGen/commit/2c8c869b34ce18ce22ddb51fa7448c1ab5e7ddd2", "source_type": "primary", "sources": [ { "label": "Publisher adds the remaining NL and MULTI checkpoints", "url": "https://github.com/salesforce/CodeGen/commit/2c8c869b34ce18ce22ddb51fa7448c1ab5e7ddd2", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Salesforce/codegen-2B-multi", "type": "model_card" }, { "label": "Original release announcement", "url": "https://www.salesforce.com/blog/codegen/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T09:44:55.983Z", "last_seen": "2026-10-29T09:44:55.983Z", "upstream_present": false, "availability": "public", "notes": "The April 1 publisher commit explicitly adds the remaining NL and MULTI checkpoints. The March 29 release enabled only MONO; its generic family description did not establish availability of this variant. Parameter count is nominal." }, { "id": "salesforce/codegen-2b-nl", "slug": "salesforce--codegen-2b-nl", "name": "CodeGen 2B NL", "developer": "Salesforce", "developer_id": "salesforce", "family": "codegen", "description": null, "release_date": "2022-04-01", "date_precision": "day", "open_weights": true, "license": "BSD-3-Clause", "parameters_b": 2, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/salesforce/CodeGen/commit/2c8c869b34ce18ce22ddb51fa7448c1ab5e7ddd2", "source_type": "primary", "sources": [ { "label": "Publisher adds the remaining NL and MULTI checkpoints", "url": "https://github.com/salesforce/CodeGen/commit/2c8c869b34ce18ce22ddb51fa7448c1ab5e7ddd2", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Salesforce/codegen-2B-nl", "type": "model_card" }, { "label": "Original release announcement", "url": "https://www.salesforce.com/blog/codegen/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T09:44:55.983Z", "last_seen": "2026-10-29T09:44:55.983Z", "upstream_present": false, "availability": "public", "notes": "The April 1 publisher commit explicitly adds the remaining NL and MULTI checkpoints. The March 29 release enabled only MONO; its generic family description did not establish availability of this variant. Parameter count is nominal." }, { "id": "salesforce/codegen-350m-multi", "slug": "salesforce--codegen-350m-multi", "name": "CodeGen 350M MULTI", "developer": "Salesforce", "developer_id": "salesforce", "family": "codegen", "description": null, "release_date": "2022-04-01", "date_precision": "day", "open_weights": true, "license": "BSD-3-Clause", "parameters_b": 0.35, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/salesforce/CodeGen/commit/2c8c869b34ce18ce22ddb51fa7448c1ab5e7ddd2", "source_type": "primary", "sources": [ { "label": "Publisher adds the remaining NL and MULTI checkpoints", "url": "https://github.com/salesforce/CodeGen/commit/2c8c869b34ce18ce22ddb51fa7448c1ab5e7ddd2", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Salesforce/codegen-350M-multi", "type": "model_card" }, { "label": "Original release announcement", "url": "https://www.salesforce.com/blog/codegen/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T09:44:55.983Z", "last_seen": "2026-10-29T09:44:55.983Z", "upstream_present": false, "availability": "public", "notes": "The April 1 publisher commit explicitly adds the remaining NL and MULTI checkpoints. The March 29 release enabled only MONO; its generic family description did not establish availability of this variant. Parameter count is nominal." }, { "id": "salesforce/codegen-350m-nl", "slug": "salesforce--codegen-350m-nl", "name": "CodeGen 350M NL", "developer": "Salesforce", "developer_id": "salesforce", "family": "codegen", "description": null, "release_date": "2022-04-01", "date_precision": "day", "open_weights": true, "license": "BSD-3-Clause", "parameters_b": 0.35, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/salesforce/CodeGen/commit/2c8c869b34ce18ce22ddb51fa7448c1ab5e7ddd2", "source_type": "primary", "sources": [ { "label": "Publisher adds the remaining NL and MULTI checkpoints", "url": "https://github.com/salesforce/CodeGen/commit/2c8c869b34ce18ce22ddb51fa7448c1ab5e7ddd2", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Salesforce/codegen-350M-nl", "type": "model_card" }, { "label": "Original release announcement", "url": "https://www.salesforce.com/blog/codegen/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T09:44:55.983Z", "last_seen": "2026-10-29T09:44:55.983Z", "upstream_present": false, "availability": "public", "notes": "The April 1 publisher commit explicitly adds the remaining NL and MULTI checkpoints. The March 29 release enabled only MONO; its generic family description did not establish availability of this variant. Parameter count is nominal." }, { "id": "salesforce/codegen-6b-multi", "slug": "salesforce--codegen-6b-multi", "name": "CodeGen 6B MULTI", "developer": "Salesforce", "developer_id": "salesforce", "family": "codegen", "description": null, "release_date": "2022-04-01", "date_precision": "day", "open_weights": true, "license": "BSD-3-Clause", "parameters_b": 6, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/salesforce/CodeGen/commit/2c8c869b34ce18ce22ddb51fa7448c1ab5e7ddd2", "source_type": "primary", "sources": [ { "label": "Publisher adds the remaining NL and MULTI checkpoints", "url": "https://github.com/salesforce/CodeGen/commit/2c8c869b34ce18ce22ddb51fa7448c1ab5e7ddd2", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Salesforce/codegen-6B-multi", "type": "model_card" }, { "label": "Original release announcement", "url": "https://www.salesforce.com/blog/codegen/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T09:44:55.983Z", "last_seen": "2026-10-29T09:44:55.983Z", "upstream_present": false, "availability": "public", "notes": "The April 1 publisher commit explicitly adds the remaining NL and MULTI checkpoints. The March 29 release enabled only MONO; its generic family description did not establish availability of this variant. Parameter count is nominal." }, { "id": "salesforce/codegen-6b-nl", "slug": "salesforce--codegen-6b-nl", "name": "CodeGen 6B NL", "developer": "Salesforce", "developer_id": "salesforce", "family": "codegen", "description": null, "release_date": "2022-04-01", "date_precision": "day", "open_weights": true, "license": "BSD-3-Clause", "parameters_b": 6, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/salesforce/CodeGen/commit/2c8c869b34ce18ce22ddb51fa7448c1ab5e7ddd2", "source_type": "primary", "sources": [ { "label": "Publisher adds the remaining NL and MULTI checkpoints", "url": "https://github.com/salesforce/CodeGen/commit/2c8c869b34ce18ce22ddb51fa7448c1ab5e7ddd2", "type": "repository" }, { "label": "Official model card", "url": "https://huggingface.co/Salesforce/codegen-6B-nl", "type": "model_card" }, { "label": "Original release announcement", "url": "https://www.salesforce.com/blog/codegen/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T09:44:55.983Z", "last_seen": "2026-10-29T09:44:55.983Z", "upstream_present": false, "availability": "public", "notes": "The April 1 publisher commit explicitly adds the remaining NL and MULTI checkpoints. The March 29 release enabled only MONO; its generic family description did not establish availability of this variant. Parameter count is nominal." }, { "id": "salesforce/codegen-16b-mono", "slug": "salesforce--codegen-16b-mono", "name": "CodeGen 16B MONO", "developer": "Salesforce", "developer_id": "salesforce", "family": "codegen", "description": null, "release_date": "2022-03-29", "date_precision": "day", "open_weights": true, "license": "BSD-3-Clause", "parameters_b": 16, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/salesforce/CodeGen/commit/cd29a1cc3bb7e7e978faa949b19a3ab61691bbfd", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://github.com/salesforce/CodeGen/commit/cd29a1cc3bb7e7e978faa949b19a3ab61691bbfd", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/Salesforce/codegen-16B-mono", "type": "model_card" }, { "label": "Original release announcement", "url": "https://www.salesforce.com/blog/codegen/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T09:44:55.983Z", "last_seen": "2026-10-29T09:44:55.983Z", "upstream_present": false, "availability": "public", "notes": "The original repository describes four sizes and three separately trained variants; counts are nominal. NL is pretrained, Multi continues on multilingual code, and Mono continues on Python." }, { "id": "salesforce/codegen-2b-mono", "slug": "salesforce--codegen-2b-mono", "name": "CodeGen 2B MONO", "developer": "Salesforce", "developer_id": "salesforce", "family": "codegen", "description": null, "release_date": "2022-03-29", "date_precision": "day", "open_weights": true, "license": "BSD-3-Clause", "parameters_b": 2, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/salesforce/CodeGen/commit/cd29a1cc3bb7e7e978faa949b19a3ab61691bbfd", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://github.com/salesforce/CodeGen/commit/cd29a1cc3bb7e7e978faa949b19a3ab61691bbfd", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/Salesforce/codegen-2B-mono", "type": "model_card" }, { "label": "Original release announcement", "url": "https://www.salesforce.com/blog/codegen/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T09:44:55.983Z", "last_seen": "2026-10-29T09:44:55.983Z", "upstream_present": false, "availability": "public", "notes": "The original repository describes four sizes and three separately trained variants; counts are nominal. NL is pretrained, Multi continues on multilingual code, and Mono continues on Python." }, { "id": "salesforce/codegen-350m-mono", "slug": "salesforce--codegen-350m-mono", "name": "CodeGen 350M MONO", "developer": "Salesforce", "developer_id": "salesforce", "family": "codegen", "description": null, "release_date": "2022-03-29", "date_precision": "day", "open_weights": true, "license": "BSD-3-Clause", "parameters_b": 0.35, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/salesforce/CodeGen/commit/cd29a1cc3bb7e7e978faa949b19a3ab61691bbfd", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://github.com/salesforce/CodeGen/commit/cd29a1cc3bb7e7e978faa949b19a3ab61691bbfd", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/Salesforce/codegen-350M-mono", "type": "model_card" }, { "label": "Original release announcement", "url": "https://www.salesforce.com/blog/codegen/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T09:44:55.983Z", "last_seen": "2026-10-29T09:44:55.983Z", "upstream_present": false, "availability": "public", "notes": "The original repository describes four sizes and three separately trained variants; counts are nominal. NL is pretrained, Multi continues on multilingual code, and Mono continues on Python." }, { "id": "salesforce/codegen-6b-mono", "slug": "salesforce--codegen-6b-mono", "name": "CodeGen 6B MONO", "developer": "Salesforce", "developer_id": "salesforce", "family": "codegen", "description": null, "release_date": "2022-03-29", "date_precision": "day", "open_weights": true, "license": "BSD-3-Clause", "parameters_b": 6, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/salesforce/CodeGen/commit/cd29a1cc3bb7e7e978faa949b19a3ab61691bbfd", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://github.com/salesforce/CodeGen/commit/cd29a1cc3bb7e7e978faa949b19a3ab61691bbfd", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/Salesforce/codegen-6B-mono", "type": "model_card" }, { "label": "Original release announcement", "url": "https://www.salesforce.com/blog/codegen/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T09:44:55.983Z", "last_seen": "2026-10-29T09:44:55.983Z", "upstream_present": false, "availability": "public", "notes": "The original repository describes four sizes and three separately trained variants; counts are nominal. NL is pretrained, Multi continues on multilingual code, and Mono continues on Python." }, { "id": "openai/code-davinci-002", "slug": "openai--code-davinci-002", "name": "code-davinci-002", "developer": "OpenAI", "developer_id": "openai", "family": "codex", "description": null, "release_date": "2022-03-15", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "unknown", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://openai.com/index/gpt-3-edit-insert/", "source_type": "primary", "sources": [ { "label": "Publisher release announcement", "url": "https://openai.com/index/gpt-3-edit-insert/", "type": "announcement" }, { "label": "Publisher historical model retirement list", "url": "https://openai.com/index/gpt-4-api-general-availability/", "type": "documentation" }, { "label": "Publisher deprecation history", "url": "https://developers.openai.com/api/docs/deprecations", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:38:48.065Z", "last_seen": "2026-10-29T10:38:48.065Z", "upstream_present": false, "availability": "public", "notes": "Historical OpenAI API model. Release is verified by the dated publisher announcement. The public Codex endpoint was retired March 23, 2023. Exact model parameters and context window have not been verified from the reviewed primary sources; GPT-3 paper sizes are not assigned to API model aliases." }, { "id": "openai/code-davinci-edit-001", "slug": "openai--code-davinci-edit-001", "name": "code-davinci-edit-001", "developer": "OpenAI", "developer_id": "openai", "family": "codex", "description": null, "release_date": "2022-03-15", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "unknown", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://openai.com/index/gpt-3-edit-insert/", "source_type": "primary", "sources": [ { "label": "Publisher release announcement", "url": "https://openai.com/index/gpt-3-edit-insert/", "type": "announcement" }, { "label": "Publisher historical model retirement list", "url": "https://openai.com/index/gpt-4-api-general-availability/", "type": "documentation" }, { "label": "Publisher deprecation history", "url": "https://developers.openai.com/api/docs/deprecations", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:38:48.065Z", "last_seen": "2026-10-29T10:38:48.065Z", "upstream_present": false, "availability": "public", "notes": "Historical OpenAI API model. Release is verified by the dated publisher announcement. Retired January 4, 2024. Exact model parameters and context window have not been verified from the reviewed primary sources; GPT-3 paper sizes are not assigned to API model aliases." }, { "id": "openai/text-davinci-002", "slug": "openai--text-davinci-002", "name": "text-davinci-002", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-3.5", "description": null, "release_date": "2022-03-15", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "unknown", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://openai.com/index/gpt-3-edit-insert/", "source_type": "primary", "sources": [ { "label": "Publisher release announcement", "url": "https://openai.com/index/gpt-3-edit-insert/", "type": "announcement" }, { "label": "Publisher historical model retirement list", "url": "https://openai.com/index/gpt-4-api-general-availability/", "type": "documentation" }, { "label": "Publisher deprecation history", "url": "https://developers.openai.com/api/docs/deprecations", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:38:48.065Z", "last_seen": "2026-10-29T10:38:48.065Z", "upstream_present": false, "availability": "public", "notes": "Historical OpenAI API model. Release is verified by the dated publisher announcement. Retired January 4, 2024. Exact model parameters and context window have not been verified from the reviewed primary sources; GPT-3 paper sizes are not assigned to API model aliases." }, { "id": "openai/text-davinci-edit-001", "slug": "openai--text-davinci-edit-001", "name": "text-davinci-edit-001", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-3", "description": null, "release_date": "2022-03-15", "date_precision": "day", "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "unknown", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://openai.com/index/gpt-3-edit-insert/", "source_type": "primary", "sources": [ { "label": "Publisher release announcement", "url": "https://openai.com/index/gpt-3-edit-insert/", "type": "announcement" }, { "label": "Publisher historical model retirement list", "url": "https://openai.com/index/gpt-4-api-general-availability/", "type": "documentation" }, { "label": "Publisher deprecation history", "url": "https://developers.openai.com/api/docs/deprecations", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-29T10:38:48.065Z", "last_seen": "2026-10-29T10:38:48.065Z", "upstream_present": false, "availability": "public", "notes": "Historical OpenAI API model. Release is verified by the dated publisher announcement. Retired January 4, 2024. Exact model parameters and context window have not been verified from the reviewed primary sources; GPT-3 paper sizes are not assigned to API model aliases." }, { "id": "eleutherai/gpt-neox-20b", "slug": "eleutherai--gpt-neox-20b", "name": "GPT-NeoX-20B", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "gpt-neox", "description": null, "release_date": "2022-02-02", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 20.55456768, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://blog.eleuther.ai/announcing-20b/", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://blog.eleuther.ai/announcing-20b/", "type": "announcement" }, { "label": "Official model card", "url": "https://huggingface.co/EleutherAI/gpt-neox-20b", "type": "model_card" }, { "label": "Compute partner release announcement", "url": "https://www.coreweave.com/blog/coreweave-unlocks-the-power-of-eleutherais-gpt-neox-20b", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T09:44:55.983Z", "last_seen": "2026-10-29T09:44:55.983Z", "upstream_present": false, "availability": "public", "notes": "First hosted availability on February 2; the original announcement records downloadable weights on February 9. Parameter count follows the developer model-card table." }, { "id": "eleutherai/gpt-j-6b", "slug": "eleutherai--gpt-j-6b", "name": "GPT-J 6B", "developer": "EleutherAI", "developer_id": "eleutherai", "family": "gpt-j", "description": "Autoregressive model trained on The Pile.", "release_date": "2021-06-09", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 6.050882784, "active_parameters_b": 6, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/kingoflolz/mesh-transformer-jax/commit/b0c18e5205ea5ff7f97a7047e964fc04a78ea42b", "source_type": "primary", "sources": [ { "label": "Official GPT-J model card", "url": "https://huggingface.co/EleutherAI/gpt-j-6b", "type": "weights" }, { "label": "Official checkpoint configuration", "url": "https://huggingface.co/EleutherAI/gpt-j-6b/blob/47e169305d2e8376be1d31e765533382721b2cc1/config.json", "type": "primary" }, { "label": "Official checkpoint tensor metadata", "url": "https://huggingface.co/EleutherAI/gpt-j-6b/blob/47e169305d2e8376be1d31e765533382721b2cc1/pytorch_model.bin", "type": "weights" }, { "label": "GPT-J parameter and buffer definitions", "url": "https://github.com/huggingface/transformers/blob/v4.18.0/src/transformers/models/gptj/modeling_gptj.py", "type": "primary" }, { "label": "Original sharded layer definitions", "url": "https://github.com/kingoflolz/mesh-transformer-jax/blob/b0c18e5205ea5ff7f97a7047e964fc04a78ea42b/mesh_transformer/layers.py", "type": "primary" }, { "label": "Original eight-shard configuration", "url": "https://github.com/kingoflolz/mesh-transformer-jax/blob/b0c18e5205ea5ff7f97a7047e964fc04a78ea42b/configs/6B_roto_256.json", "type": "primary" }, { "label": "Later Pythia parameter conventions", "url": "https://arxiv.org/html/2304.01373v2", "type": "primary" }, { "label": "Later GPT-NeoX comparison tables", "url": "https://arxiv.org/html/2204.06745v1", "type": "primary" }, { "label": "DeepSeekMoE activated-parameter comparison, Figure 7", "url": "https://aclanthology.org/2024.acl-long.70.pdf", "type": "primary" }, { "label": "Original release source", "url": "https://github.com/kingoflolz/mesh-transformer-jax/commit/b0c18e5205ea5ff7f97a7047e964fc04a78ea42b", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "The consolidated Hugging Face checkpoint has 6,050,882,784 learned parameters, with untied embeddings and output head counted separately and causal-mask buffers excluded. The original eight-shard JAX representation reports 6,053,381,344 elements because of replicated normalization and shard-specific biases. Approximately 6B active follows DeepSeek’s Figure 7 benchmark convention, not an EleutherAI active-count disclosure or an exact tensor count.", "field_status": { "reasoning": "undisclosed", "tool_call": "undisclosed" }, "audited_at": "2026-10-27T13:53:59.517708+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/184.json" }, { "id": "openai/gpt-3-175b", "slug": "openai--gpt-3-175b", "name": "GPT-3 175B", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-3", "description": "Largest original GPT-3 model, a 96-layer Transformer for text completion and learning from examples in its prompt.", "release_date": "2020-06-11", "date_precision": "day", "open_weights": false, "license": "Proprietary", "parameters_b": 175, "active_parameters_b": 175, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://openai.com/index/openai-api/", "source_type": "primary", "sources": [ { "label": "GPT-3 original technical report, revision 4", "url": "https://arxiv.org/html/2005.14165v4", "type": "technical_report" }, { "label": "Original release source", "url": "https://openai.com/index/openai-api/", "type": "announcement" }, { "label": "GPT-3 175B model card", "url": "https://github.com/openai/gpt-3/blob/d7a9bb505df6f630f9bab3b30c889e52f22eb9ea/model-card.md", "type": "primary" }, { "label": "GPT-3 API and licensing history", "url": "https://openai.com/index/openai-licenses-gpt-3-technology-to-microsoft/", "type": "announcement" }, { "label": "Original GPT-3 research announcement", "url": "https://openai.com/index/language-models-are-few-shot-learners/", "type": "announcement" }, { "label": "Google GLaM comparison: GPT-3 activated parameters", "url": "https://proceedings.mlr.press/v162/du22c/du22c.pdf", "type": "technical_report" }, { "label": "Google GLaM announcement: total and activated parameters", "url": "https://research.google/blog/more-efficient-in-context-learning-with-glam/", "type": "primary" }, { "label": "GPT-3 paper", "url": "https://arxiv.org/abs/2005.14165", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "June 11, 2020 marks the initial API private-beta announcement and access requests; the paper was announced May 28. OpenAI reports a nominal 175B total; Google's GLaM comparison reports a nominal 175B activated per input token. The original model has a 2048-token context and alternating dense/local sparse attention. The September 2020 retraining and later InstructGPT models are distinct versions.", "field_status": { "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T16:15:38.183467Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/401.json" }, { "id": "openai/gpt-2-1.5b", "slug": "openai--gpt-2-1.5b", "name": "GPT-2 1.5B", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-2", "description": "Largest GPT-2 checkpoint, a 48-layer dense Transformer trained on WebText for autoregressive text generation.", "release_date": "2019-11-05", "date_precision": "day", "open_weights": true, "license": "Modified MIT", "parameters_b": 1.5576112, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 1024, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://openai.com/index/gpt-2-1-5b-release/", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://openai.com/index/gpt-2-1-5b-release/", "type": "announcement" }, { "label": "GPT-2 model card, November 2019", "url": "https://github.com/openai/gpt-2/blob/9b63575ef42771a015060c964af2c3da4cf7c8ab/model_card.md", "type": "primary" }, { "label": "Original GPT-2 Modified MIT License", "url": "https://github.com/openai/gpt-2/blob/9b63575ef42771a015060c964af2c3da4cf7c8ab/LICENSE", "type": "primary" }, { "label": "Original OpenAI 1558M hyperparameters", "url": "https://openaipublic.blob.core.windows.net/gpt-2/models/1558M/hparams.json", "type": "primary" }, { "label": "Original GPT-2 model implementation", "url": "https://github.com/openai/gpt-2/blob/9b63575ef42771a015060c964af2c3da4cf7c8ab/src/model.py", "type": "primary" }, { "label": "Original OpenAI 1558M TensorFlow checkpoint index", "url": "https://openaipublic.blob.core.windows.net/gpt-2/models/1558M/model.ckpt.index", "type": "primary" }, { "label": "GPT-2 license history", "url": "https://api.github.com/repos/openai/gpt-2/commits?path=LICENSE&per_page=100", "type": "primary" }, { "label": "Language Models are Unsupervised Multitask Learners", "url": "https://cdn.openai.com/better-language-models/language_models_are_unsupervised_multitask_learners.pdf", "type": "primary" }, { "label": "Small Batch Size Training for Language Models: When Vanilla SGD Works, and Why Gradient Accumulation Is Wasteful", "url": "https://proceedings.neurips.cc/paper_files/paper/2025/file/dabfbf500318462382aa70a95466ad85-Paper-Conference.pdf", "type": "primary" }, { "label": "Too Big to Fail: Larger Language Models are Disproportionately Resilient to Induction of Dementia-Related Linguistic Anomalies", "url": "https://aclanthology.org/2024.findings-acl.380.pdf", "type": "primary" }, { "label": "GPT-2 model card", "url": "https://github.com/openai/gpt-2/blob/master/model_card.md", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "Released publicly on November 5, 2019. The original 1558M checkpoint contains 1,557,611,200 learned parameters with tied token embeddings counted once; OpenAI labels it nominally 1.5B. A separate active-parameter count remains undisclosed. The release originally used MIT; the repository changed to Modified MIT in January 2020, which remains its current published license.", "field_status": { "active_parameters_b": "undisclosed", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T16:12:56.550932Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/397.json" }, { "id": "google/t5-11b", "slug": "google--t5-11b", "name": "T5 11B", "developer": "Google", "developer_id": "google", "family": "t5", "description": "Original T5 encoder-decoder model pretrained with span corruption and supervised text-to-text tasks.", "release_date": "2019-10-23", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 11.307321344, "active_parameters_b": 11, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/google-research/text-to-text-transfer-transformer/blob/9fd7b14a769417be33bc6c850f9598764913c833/README.md", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://github.com/google-research/text-to-text-transfer-transformer/blob/9fd7b14a769417be33bc6c850f9598764913c833/README.md", "type": "announcement" }, { "label": "Original T5 release commit metadata", "url": "https://api.github.com/repos/google-research/text-to-text-transfer-transformer/commits/9fd7b14a769417be33bc6c850f9598764913c833", "type": "primary" }, { "label": "Original Google T5-11B checkpoint tensor index", "url": "https://storage.googleapis.com/t5-data/pretrained_models/11B/model.ckpt-1000000.index", "type": "weights" }, { "label": "Original T5-11B operative configuration", "url": "https://storage.googleapis.com/t5-data/pretrained_models/11B/operative_config.gin", "type": "weights" }, { "label": "Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer", "url": "https://jmlr.org/papers/volume21/20-074/20-074.pdf", "type": "primary" }, { "label": "Pinned original T5-11B Transformers configuration", "url": "https://huggingface.co/google-t5/t5-11b/resolve/90f37703b3334dfe9d2b009bfcbfbf1ac9d28ea3/config.json", "type": "weights" }, { "label": "T5 model implementation", "url": "https://github.com/huggingface/transformers/blob/a672eb5f474b8b30173e9f96eab66adf292da635/src/transformers/models/t5/modeling_t5.py", "type": "primary" }, { "label": "T5 11B model card", "url": "https://huggingface.co/google-t5/t5-11b/blob/90f37703b3334dfe9d2b009bfcbfbf1ac9d28ea3/README.md", "type": "primary" }, { "label": "Original T5 Apache License 2.0", "url": "https://github.com/google-research/text-to-text-transfer-transformer/blob/9fd7b14a769417be33bc6c850f9598764913c833/LICENSE", "type": "primary" }, { "label": "T5X released-model documentation", "url": "https://github.com/google-research/t5x/blob/main/docs/models.md", "type": "primary" }, { "label": "Entities: what are they, where are they and how do we model them? — TAC keynote", "url": "https://tac.nist.gov/publications/2020/presentations/TAC2020.KEYNOTE.Bikel.presentation.pdf", "type": "primary" }, { "label": "How Much Knowledge Can You Pack Into the Parameters of a Language Model?", "url": "https://aclanthology.org/2020.emnlp-main.437.pdf", "type": "primary" }, { "label": "Entities as Experts: Sparse Memory Access with Entity Supervision", "url": "https://aclanthology.org/2020.emnlp-main.400.pdf", "type": "primary" }, { "label": "T5 research paper", "url": "https://arxiv.org/abs/1910.10683", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "The original checkpoint contains 11,307,321,344 unique parameters, with shared token/output embeddings counted once and optimizer state excluded. Active 11B follows Google Research’s nominal per-example QA comparison, transferred through the documented original T5 fine-tuning relationship; it is not an exact decoder-token count. The 512-token training/default length is not a verified hard context limit. Release date follows the original repository snapshot.", "field_status": { "context_window": "unverified", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T13:32:16.579331+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/229.json" }, { "id": "google/t5-3b", "slug": "google--t5-3b", "name": "T5 3B", "developer": "Google", "developer_id": "google", "family": "t5", "description": "Google's original dense encoder-decoder model for text-to-text tasks, pretrained on a mixture of span-corrupted text and supervised NLP tasks.", "release_date": "2019-10-23", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 2.851598336, "active_parameters_b": 3, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/google-research/text-to-text-transfer-transformer/blob/9fd7b14a769417be33bc6c850f9598764913c833/README.md", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://github.com/google-research/text-to-text-transfer-transformer/blob/9fd7b14a769417be33bc6c850f9598764913c833/README.md", "type": "announcement" }, { "label": "Original T5 release commit metadata", "url": "https://api.github.com/repos/google-research/text-to-text-transfer-transformer/commits/9fd7b14a769417be33bc6c850f9598764913c833", "type": "primary" }, { "label": "Original Google T5-3B operative configuration", "url": "https://storage.googleapis.com/t5-data/pretrained_models/3B/operative_config.gin", "type": "primary" }, { "label": "Original Google T5-3B checkpoint index", "url": "https://storage.googleapis.com/t5-data/pretrained_models/3B/model.ckpt-1000000.index", "type": "weights" }, { "label": "T5-3B checkpoint model card", "url": "https://huggingface.co/google-t5/t5-3b/blob/bed96aab9ee46012a5046386105ee5fd0ac572f0/README.md", "type": "primary" }, { "label": "Pinned T5-3B configuration", "url": "https://huggingface.co/google-t5/t5-3b/resolve/bed96aab9ee46012a5046386105ee5fd0ac572f0/config.json", "type": "weights" }, { "label": "Pinned converted T5-3B safetensors header", "url": "https://huggingface.co/google-t5/t5-3b/resolve/bed96aab9ee46012a5046386105ee5fd0ac572f0/model.safetensors", "type": "weights" }, { "label": "Pinned T5 implementation", "url": "https://github.com/huggingface/transformers/blob/8cb5963cc22174954e7dca2c0a3320b7dc2f4edc/src/transformers/models/t5/modeling_t5.py", "type": "primary" }, { "label": "T5 research paper", "url": "https://arxiv.org/abs/1910.10683", "type": "paper" }, { "label": "Original T5 Apache 2.0 license", "url": "https://github.com/google-research/text-to-text-transfer-transformer/blob/9fd7b14a769417be33bc6c850f9598764913c833/LICENSE", "type": "primary" }, { "label": "Google Research TAC keynote: activated-parameter comparison", "url": "https://tac.nist.gov/publications/2020/presentations/TAC2020.KEYNOTE.Bikel.presentation.pdf", "type": "primary" }, { "label": "Original T5 QA fine-tune lineage", "url": "https://aclanthology.org/2020.emnlp-main.437.pdf", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "The original checkpoint has 2,851,598,336 parameters, counting both stacks and shared embeddings once; the conversion adds an unused 1,024-element bias. Nominal active size 3B follows Google Research’s TAC 2020 benchmark comparison, linked to the original model through its documented QA fine-tune. This rounded convention is not a phase-specific or per-token tensor census. Training used 512-token inputs and targets; a universal hard context maximum is not established.", "field_status": { "context_window": "unverified", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T13:32:44.880636+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/230.json" }, { "id": "google/t5-base", "slug": "google--t5-base", "name": "T5 Base", "developer": "Google", "developer_id": "google", "family": "t5", "description": "Google’s encoder-decoder T5 model trained on text denoising and supervised text-to-text tasks.", "release_date": "2019-10-23", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.222903552, "active_parameters_b": 0.22, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/google-research/text-to-text-transfer-transformer/blob/9fd7b14a769417be33bc6c850f9598764913c833/README.md", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://github.com/google-research/text-to-text-transfer-transformer/blob/9fd7b14a769417be33bc6c850f9598764913c833/README.md", "type": "announcement" }, { "label": "Original release commit metadata", "url": "https://api.github.com/repos/google-research/text-to-text-transfer-transformer/commits/9fd7b14a769417be33bc6c850f9598764913c833", "type": "primary" }, { "label": "T5 research paper", "url": "https://arxiv.org/abs/1910.10683", "type": "paper" }, { "label": "Original T5-base operative configuration", "url": "https://storage.googleapis.com/t5-data/pretrained_models/base/operative_config.gin", "type": "primary" }, { "label": "Original T5-base TensorFlow checkpoint index", "url": "https://storage.googleapis.com/t5-data/pretrained_models/base/model.ckpt-999900.index", "type": "weights" }, { "label": "Exact converted T5-base configuration", "url": "https://huggingface.co/google-t5/t5-base/resolve/a9723ea7f1b39c1eae772870f3b547bf6ef7e6c1/config.json", "type": "weights" }, { "label": "Converted T5-base tensor header", "url": "https://huggingface.co/google-t5/t5-base/resolve/a9723ea7f1b39c1eae772870f3b547bf6ef7e6c1/model.safetensors", "type": "weights" }, { "label": "Pinned Transformers T5 implementation", "url": "https://github.com/huggingface/transformers/blob/a672eb5f474b8b30173e9f96eab66adf292da635/src/transformers/models/t5/modeling_t5.py", "type": "primary" }, { "label": "T5 Base model card", "url": "https://huggingface.co/google-t5/t5-base/blob/a9723ea7f1b39c1eae772870f3b547bf6ef7e6c1/README.md", "type": "primary" }, { "label": "Original T5 Apache2.0 license", "url": "https://github.com/google-research/text-to-text-transfer-transformer/blob/9fd7b14a769417be33bc6c850f9598764913c833/LICENSE", "type": "primary" }, { "label": "TensorFlow checkpoint index schema", "url": "https://github.com/tensorflow/tensorflow/blob/v2.20.0/tensorflow/core/protobuf/tensor_bundle.proto", "type": "primary" }, { "label": "MC-SMoE paper: nominal activated-parameter comparison", "url": "https://proceedings.iclr.cc/paper_files/paper/2024/file/3d09a88c3372cdb79401fde592ca4db8-Paper-Conference.pdf", "type": "primary" }, { "label": "MC-SMoE exact T5 baseline implementation", "url": "https://github.com/UNITES-Lab/MC-SMoE/blob/99b400f0d5b3be0015ebec4a5a80a7821be1aebd/mcsmoe/t5-finetune-on-downstream.py", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "The original checkpoint has 222,903,552 parameters across encoder, decoder and shared embedding; tied input/output weights count once. Nominal 220M active follows Li et al.’s MC-SMoE comparison convention, without separate encoder/decode-phase counts. The unused 384-value HF cross-attention bias is excluded from the total. Training/default length is 512 tokens per input/target, not a fixed maximum. Thinking mode and tool calling remain unverified. Original release: October 23, 2019.", "field_status": { "context_window": "unverified", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T13:34:54.499708+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/231.json" }, { "id": "google/t5-large", "slug": "google--t5-large", "name": "T5 Large", "developer": "Google", "developer_id": "google", "family": "t5", "description": "Google's original dense T5 encoder-decoder model for text-to-text tasks.", "release_date": "2019-10-23", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.737668096, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/google-research/text-to-text-transfer-transformer/blob/9fd7b14a769417be33bc6c850f9598764913c833/README.md", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://github.com/google-research/text-to-text-transfer-transformer/blob/9fd7b14a769417be33bc6c850f9598764913c833/README.md", "type": "announcement" }, { "label": "T5 research paper", "url": "https://arxiv.org/abs/1910.10683", "type": "paper" }, { "label": "Original Google T5-Large TensorFlow checkpoint index", "url": "https://storage.googleapis.com/t5-data/pretrained_models/large/model.ckpt-1000700.index", "type": "weights" }, { "label": "Original T5-Large operative configuration", "url": "https://storage.googleapis.com/t5-data/pretrained_models/large/operative_config.gin", "type": "weights" }, { "label": "Pinned T5-Large converted configuration", "url": "https://huggingface.co/google-t5/t5-large/resolve/150ebc2c4b72291e770f58e6057481c8d2ed331a/config.json", "type": "weights" }, { "label": "Pinned converted T5-Large safetensors header", "url": "https://huggingface.co/google-t5/t5-large/resolve/150ebc2c4b72291e770f58e6057481c8d2ed331a/model.safetensors", "type": "weights" }, { "label": "Pinned T5 implementation", "url": "https://github.com/huggingface/transformers/blob/a672eb5f474b8b30173e9f96eab66adf292da635/src/transformers/models/t5/modeling_t5.py", "type": "primary" }, { "label": "Pinned T5 Large model card", "url": "https://huggingface.co/google-t5/t5-large/blob/150ebc2c4b72291e770f58e6057481c8d2ed331a/README.md", "type": "primary" }, { "label": "Original T5 Apache License 2.0", "url": "https://github.com/google-research/text-to-text-transfer-transformer/blob/9fd7b14a769417be33bc6c850f9598764913c833/LICENSE", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "The original checkpoint has 737,668,096 unique parameters across encoder, decoder and shared embeddings; Google’s 770M label is nominal. Tied embeddings count once; optimizer state and an unused 512-value conversion bias are excluded. No separate active-parameter count was found in the checked primary sources. A 512-token training/default length does not establish a fixed maximum context. Release date follows the original checkpoint release README and paper.", "field_status": { "active_parameters_b": "undisclosed", "context_window": "unverified", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T13:35:32.500376+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/232.json" }, { "id": "google/t5-small", "slug": "google--t5-small", "name": "T5 Small", "developer": "Google", "developer_id": "google", "family": "t5", "description": "Original T5 encoder-decoder model pretrained with span corruption and supervised text-to-text tasks.", "release_date": "2019-10-23", "date_precision": "day", "open_weights": true, "license": "Apache-2.0", "parameters_b": 0.060506624, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/google-research/text-to-text-transfer-transformer/blob/9fd7b14a769417be33bc6c850f9598764913c833/README.md", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://github.com/google-research/text-to-text-transfer-transformer/blob/9fd7b14a769417be33bc6c850f9598764913c833/README.md", "type": "announcement" }, { "label": "Original T5-Small tensor index", "url": "https://storage.googleapis.com/t5-data/pretrained_models/small/model.ckpt-1000000.index", "type": "weights" }, { "label": "Original T5-Small operative configuration", "url": "https://storage.googleapis.com/t5-data/pretrained_models/small/operative_config.gin", "type": "primary" }, { "label": "T5 research paper", "url": "https://jmlr.org/papers/volume21/20-074/20-074.pdf", "type": "primary" }, { "label": "T5 Small model card", "url": "https://huggingface.co/google-t5/t5-small/blob/df1b051c49625cf57a3d0d8d3863ed4d13564fe4/README.md", "type": "weights" }, { "label": "T5 research paper", "url": "https://arxiv.org/abs/1910.10683", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "Google reports a nominal 60M total; the original checkpoint contains 60,506,624 unique learned parameters, counting shared encoder/decoder/output embeddings once. An unused 256-element bias in a converted checkpoint is excluded. No separate active-parameter count was found in the checked primary sources. Length 512 is a training/default sequence length, not an established maximum. Original public release: October 23, 2019.", "field_status": { "active_parameters_b": "undisclosed", "context_window": "unverified", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T13:35:47.717796+00:00", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/233.json" }, { "id": "openai/gpt-2-774m", "slug": "openai--gpt-2-774m", "name": "GPT-2 774M", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-2", "description": "Large GPT-2 checkpoint, a 36-layer dense Transformer trained on WebText for autoregressive text generation.", "release_date": "2019-08-20", "date_precision": "day", "open_weights": true, "license": "Modified MIT", "parameters_b": 0.77403008, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 1024, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://openai.com/index/gpt-2-6-month-follow-up/", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://openai.com/index/gpt-2-6-month-follow-up/", "type": "announcement" }, { "label": "Original 774M release commit", "url": "https://github.com/openai/gpt-2/commit/f35fa1d920e9d2d0690f66d03aa3f76b3c59230e", "type": "primary" }, { "label": "Original 774M checkpoint hyperparameters", "url": "https://openaipublic.blob.core.windows.net/gpt-2/models/774M/hparams.json", "type": "primary" }, { "label": "Original GPT-2 model implementation", "url": "https://github.com/openai/gpt-2/blob/9b63575ef42771a015060c964af2c3da4cf7c8ab/src/model.py", "type": "primary" }, { "label": "Original 774M TensorFlow checkpoint index", "url": "https://openaipublic.blob.core.windows.net/gpt-2/models/774M/model.ckpt.index", "type": "primary" }, { "label": "Current original-repository Modified MIT License", "url": "https://github.com/openai/gpt-2/blob/9b63575ef42771a015060c964af2c3da4cf7c8ab/LICENSE", "type": "primary" }, { "label": "License at the August 2019 public release", "url": "https://raw.githubusercontent.com/openai/gpt-2/f35fa1d920e9d2d0690f66d03aa3f76b3c59230e/LICENSE", "type": "primary" }, { "label": "January 2020 license update", "url": "https://github.com/openai/gpt-2/commit/0f97760ebe1ccad5778a6d5957b2be9800e4efd8", "type": "primary" }, { "label": "Language Models are Unsupervised Multitask Learners", "url": "https://cdn.openai.com/better-language-models/language_models_are_unsupervised_multitask_learners.pdf", "type": "primary" }, { "label": "GPT-2 model card", "url": "https://github.com/openai/gpt-2/blob/master/model_card.md", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "Publicly released on August 20, 2019 after earlier selected-partner sharing. The original configuration and implementation support exactly 774,030,080 learned parameters, counting tied token embeddings once; 774M is OpenAI's corrected nominal label, replacing 762M. Active parameters are not separately established for this checkpoint. The release originally carried MIT; the current official repository publishes Modified MIT terms introduced in January 2020.", "field_status": { "active_parameters_b": "undisclosed", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T16:14:34.557948Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/400.json" }, { "id": "openai/gpt-2-355m", "slug": "openai--gpt-2-355m", "name": "GPT-2 355M", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-2", "description": "Medium GPT-2 text model, a 24-layer decoder-only Transformer trained on WebText.", "release_date": "2019-05-03", "date_precision": "day", "open_weights": true, "license": "Modified MIT", "parameters_b": 0.354823168, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 1024, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/openai/gpt-2/commit/0503b1b24969ccbaf93a5d05f2c2588dabff1c46", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://github.com/openai/gpt-2/commit/0503b1b24969ccbaf93a5d05f2c2588dabff1c46", "type": "announcement" }, { "label": "Correction of model size labels", "url": "https://github.com/openai/gpt-2/commit/f35fa1d920e9d2d0690f66d03aa3f76b3c59230e", "type": "repository" }, { "label": "GPT-2 model card", "url": "https://github.com/openai/gpt-2/blob/9b63575ef42771a015060c964af2c3da4cf7c8ab/model_card.md", "type": "model_card" }, { "label": "Exact medium-model configuration", "url": "https://openaipublic.blob.core.windows.net/gpt-2/models/355M/hparams.json", "type": "configuration" }, { "label": "Original medium checkpoint index", "url": "https://openaipublic.blob.core.windows.net/gpt-2/models/355M/model.ckpt.index", "type": "weights" }, { "label": "Original architecture implementation", "url": "https://raw.githubusercontent.com/openai/gpt-2/9b63575ef42771a015060c964af2c3da4cf7c8ab/src/model.py", "type": "repository" }, { "label": "Current model license", "url": "https://raw.githubusercontent.com/openai/gpt-2/9b63575ef42771a015060c964af2c3da4cf7c8ab/LICENSE", "type": "license" }, { "label": "Original GPT-2 technical report", "url": "https://cdn.openai.com/better-language-models/language_models_are_unsupervised_multitask_learners.pdf", "type": "paper" }, { "label": "GPT-2 model card", "url": "https://github.com/openai/gpt-2/blob/master/model_card.md", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "The medium GPT-2 checkpoint was released on May 3, 2019 under the erroneous 345M label, corrected to 355M in August. Its complete learned count is 354,823,168, with tied output embeddings counted once. Current repository terms are Modified MIT; the original release used MIT. Active parameters are not separately defined for this dense model.", "field_status": { "active_parameters_b": "not_applicable", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T01:18:39.494300Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/399.json" }, { "id": "openai/gpt-2-124m", "slug": "openai--gpt-2-124m", "name": "GPT-2 124M", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-2", "description": "Smallest GPT-2 model, a 12-layer causal Transformer pretrained on WebText for next-token prediction.", "release_date": "2019-02-14", "date_precision": "day", "open_weights": true, "license": "Modified MIT", "parameters_b": 0.124439808, "active_parameters_b": 0.1244, "parameters_source": "primary_source", "architecture": "dense", "context_window": 1024, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://openai.com/index/better-language-models/", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://openai.com/index/better-language-models/", "type": "announcement" }, { "label": "GPT-2 count correction and release repository", "url": "https://github.com/openai/gpt-2/blob/9b63575ef42771a015060c964af2c3da4cf7c8ab/README.md", "type": "primary" }, { "label": "GPT-2 Modified MIT license", "url": "https://github.com/openai/gpt-2/blob/9b63575ef42771a015060c964af2c3da4cf7c8ab/LICENSE", "type": "primary" }, { "label": "Original GPT-2 implementation", "url": "https://github.com/openai/gpt-2/blob/9b63575ef42771a015060c964af2c3da4cf7c8ab/src/model.py", "type": "primary" }, { "label": "Original 124M hyperparameters", "url": "https://openaipublic.blob.core.windows.net/gpt-2/models/124M/hparams.json", "type": "primary" }, { "label": "Original 124M checkpoint index", "url": "https://openaipublic.blob.core.windows.net/gpt-2/models/124M/model.ckpt.index", "type": "weights" }, { "label": "GPT-2 technical paper", "url": "https://cdn.openai.com/better-language-models/language_models_are_unsupervised_multitask_learners.pdf", "type": "technical_report" }, { "label": "GPT-2 model card", "url": "https://github.com/openai/gpt-2/blob/9b63575ef42771a015060c964af2c3da4cf7c8ab/model_card.md", "type": "primary" }, { "label": "Original MIT license, February 2019", "url": "https://github.com/openai/gpt-2/blob/aae26abd600ad32771fadf6284f339f07b9d8ea1/LICENSE", "type": "primary" }, { "label": "Tara 1.4 original GPT-2 Small active-parameter comparison", "url": "https://huggingface.co/aungkomyint/tara1.4-base/blob/346c354dce56e32922479fee40494d8a6ef1a4a7/README.md", "type": "primary" }, { "label": "GPT-2 model card", "url": "https://github.com/openai/gpt-2/blob/master/model_card.md", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "Released in February 2019 as the small GPT-2 model. The original 117M label was a counting error later corrected to 124M; the complete checkpoint contains 124,439,808 learned parameters, with tied token/output embeddings. Aung Ko Myint's Tara 1.4 comparison reports nominal 124.4M active parameters; this is its research benchmark convention, not a separate OpenAI disclosure. The repository first added an MIT license on February 20, 2019 and changed its published text to Modified MIT in January 2020.", "field_status": { "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T16:14:22.811236Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/398.json" }, { "id": "openai/gpt-1", "slug": "openai--gpt-1", "name": "GPT-1", "developer": "OpenAI", "developer_id": "openai", "family": "gpt", "description": "OpenAI's original 12-layer decoder-only Transformer, pretrained on BooksCorpus.", "release_date": "2018-06-11", "date_precision": "day", "open_weights": true, "license": "MIT", "parameters_b": 0.116534784, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 512, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://openai.com/index/language-unsupervised/", "source_type": "primary", "sources": [ { "label": "Original release source", "url": "https://openai.com/index/language-unsupervised/", "type": "announcement" }, { "label": "Original technical report", "url": "https://cdn.openai.com/research-covers/language-unsupervised/language_understanding_paper.pdf", "type": "paper" }, { "label": "Original checkpoint tensor shapes", "url": "https://raw.githubusercontent.com/openai/finetune-transformer-lm/a69b5c43b0452462890bca8ff92fb75dee9290cf/model/params_shapes.json", "type": "weights" }, { "label": "Original architecture and checkpoint loader", "url": "https://raw.githubusercontent.com/openai/finetune-transformer-lm/a69b5c43b0452462890bca8ff92fb75dee9290cf/train.py", "type": "repository" }, { "label": "Original checkpoint files", "url": "https://github.com/openai/finetune-transformer-lm/tree/a69b5c43b0452462890bca8ff92fb75dee9290cf/model", "type": "weights" }, { "label": "Original model release license", "url": "https://raw.githubusercontent.com/openai/finetune-transformer-lm/f7c13308472857da7317a8b6628e9e628d9f6a73/LICENSE", "type": "license" }, { "label": "Code, pretrained model and license", "url": "https://github.com/openai/finetune-transformer-lm", "type": "weights" } ], "aliases": [], "first_seen": "2026-10-26T08:55:12.627Z", "last_seen": "2026-10-26T08:55:12.627Z", "upstream_present": false, "notes": "The original pretrained checkpoint contains exactly 116,534,784 parameters (about 117 million). The count includes learned token and position embeddings and all twelve Transformer blocks, with tied output weights counted once. Fine-tuning adds three token embeddings and a classifier; these are excluded. Active parameters are not separately defined for this dense model. Commonsense reasoning benchmark results do not establish a dedicated reasoning mode.", "field_status": { "active_parameters_b": "not_applicable", "reasoning": "unverified", "tool_call": "unverified" }, "audited_at": "2026-10-27T01:13:46.889506Z", "audit_url": "https://github.com/kiankyars/model-tracker/blob/main/research/model-audit/2026-10-26/reports/396.json" }, { "id": "ai21/j1-grande", "slug": "ai21--j1-grande", "name": "Jurassic-1 Grande", "developer": "AI21 Labs", "developer_id": "ai21", "family": "jurassic-1", "description": null, "release_date": null, "date_precision": null, "open_weights": false, "license": "Proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.ai21.com/blog/introducing-j1-grande/", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://www.ai21.com/blog/introducing-j1-grande/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:10:00.648Z", "last_seen": "2026-10-29T10:10:00.648Z", "upstream_present": false, "availability": "public", "notes": "The May 2 announcement says Studio users had already tested Grande during preceding weeks. The exact first external availability date and parameter count remain unverified; May 2 is not substituted for the earlier access date.", "announcement_date": "2022-05-02" }, { "id": "bigcode/octocoder", "slug": "bigcode--octocoder", "name": "OctoCoder", "developer": "BigCode", "developer_id": "bigcode", "family": "starcoder", "description": null, "release_date": null, "date_precision": null, "open_weights": true, "license": "BigCode OpenRAIL-M", "parameters_b": 15.5, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigcode/octocoder", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/bigcode/octocoder/blob/0f863c63e38ba80fc2c4010f34a7f46d537a9eee/README.md", "type": "model_card" }, { "label": "Developer paper describing OctoCoder", "url": "https://arxiv.org/abs/2308.07124", "type": "paper" }, { "label": "Official training repository", "url": "https://github.com/bigcode-project/octopack", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-26T09:05:47.987Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": false, "availability": "public", "notes": "BigCode instruction-tuned its own StarCoder on CommitPackFT and OASST. The official card gives 15.5B parameters. Public weights are verified; the first public release date remains unverified. The August 14 paper is not used as a release date because repository activity predates it." }, { "id": "bigcode/starcoderbase-1b", "slug": "bigcode--starcoderbase-1b", "name": "StarCoderBase 1B", "developer": "BigCode", "developer_id": "bigcode", "family": "starcoder", "description": null, "release_date": null, "date_precision": null, "open_weights": true, "license": "BigCode OpenRAIL-M", "parameters_b": 1, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigcode/starcoderbase-1b", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/bigcode/starcoderbase-1b", "type": "model_card" }, { "label": "BigCode developer confirmation of smaller-model availability", "url": "https://huggingface.co/bigcode/starcoder/discussions/53", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-26T09:05:47.987Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": false, "availability": "public", "notes": "The official card verifies this separately pretrained model. The first public release date remains unverified. On October 5, 2023 a BigCode developer confirmed the 1B, 3B and 7B models had been released a few months earlier; the May release date of the 15.5B model is not reused." }, { "id": "bigcode/starcoderbase-3b", "slug": "bigcode--starcoderbase-3b", "name": "StarCoderBase 3B", "developer": "BigCode", "developer_id": "bigcode", "family": "starcoder", "description": null, "release_date": null, "date_precision": null, "open_weights": true, "license": "BigCode OpenRAIL-M", "parameters_b": 3, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigcode/starcoderbase-3b", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/bigcode/starcoderbase-3b", "type": "model_card" }, { "label": "BigCode developer confirmation of smaller-model availability", "url": "https://huggingface.co/bigcode/starcoder/discussions/53", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-26T09:05:47.987Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": false, "availability": "public", "notes": "The official card verifies this separately pretrained model. The first public release date remains unverified. On October 5, 2023 a BigCode developer confirmed the 1B, 3B and 7B models had been released a few months earlier; the May release date of the 15.5B model is not reused." }, { "id": "bigcode/starcoderbase-7b", "slug": "bigcode--starcoderbase-7b", "name": "StarCoderBase 7B", "developer": "BigCode", "developer_id": "bigcode", "family": "starcoder", "description": null, "release_date": null, "date_precision": null, "open_weights": true, "license": "BigCode OpenRAIL-M", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigcode/starcoderbase-7b", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/bigcode/starcoderbase-7b", "type": "model_card" }, { "label": "BigCode developer confirmation of smaller-model availability", "url": "https://huggingface.co/bigcode/starcoder/discussions/53", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-26T09:05:47.987Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": false, "availability": "public", "notes": "The official card verifies this separately pretrained model. The first public release date remains unverified. On October 5, 2023 a BigCode developer confirmed the 1B, 3B and 7B models had been released a few months earlier; the May release date of the 15.5B model is not reused." }, { "id": "bigcode/starcoderplus", "slug": "bigcode--starcoderplus", "name": "StarCoderPlus", "developer": "BigCode", "developer_id": "bigcode", "family": "starcoder", "description": null, "release_date": null, "date_precision": null, "open_weights": true, "license": "BigCode OpenRAIL-M", "parameters_b": 15.5, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 8192, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/bigcode/starcoderplus", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/bigcode/starcoderplus", "type": "model_card" } ], "aliases": [], "first_seen": "2026-10-26T09:05:47.987Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": false, "availability": "public", "notes": "BigCode continued pretraining its own StarCoderBase on 600B additional English and code tokens. Public weights and specifications are verified; the first public release date remains unverified." }, { "id": "google/chinchilla-70b", "slug": "google--chinchilla-70b", "name": "Chinchilla 70B", "developer": "Google", "developer_id": "google", "family": "chinchilla", "description": null, "release_date": null, "date_precision": null, "open_weights": false, "license": null, "parameters_b": 70, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://arxiv.org/abs/2203.15556", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://arxiv.org/abs/2203.15556", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T10:17:34.586Z", "last_seen": "2026-10-29T10:17:34.586Z", "upstream_present": false, "availability": "announced", "notes": "DeepMind research model. The paper announces a nominal 70B model; external access to this checkpoint was not verified. The paper date is recorded as an announcement, not a release.", "announcement_date": "2022-03-29" }, { "id": "google/flan-palm-540b", "slug": "google--flan-palm-540b", "name": "Flan-PaLM 540B", "developer": "Google", "developer_id": "google", "family": "flan-palm", "description": null, "release_date": null, "date_precision": null, "open_weights": false, "license": null, "parameters_b": 540, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://arxiv.org/html/2210.11416v5", "source_type": "primary", "sources": [ { "label": "Developer technical report", "url": "https://arxiv.org/html/2210.11416v5", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:17:34.586Z", "last_seen": "2026-10-29T10:17:34.586Z", "upstream_present": false, "availability": "announced", "notes": "Research model announced by its developer. External availability of this exact checkpoint was not verified; this record is excluded from release counts. Parameter count follows the paper table; later commercial API aliases are not assumed equivalent.", "announcement_date": "2022-10-20" }, { "id": "google/flan-palm-62b", "slug": "google--flan-palm-62b", "name": "Flan-PaLM 62B", "developer": "Google", "developer_id": "google", "family": "flan-palm", "description": null, "release_date": null, "date_precision": null, "open_weights": false, "license": null, "parameters_b": 62, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://arxiv.org/html/2210.11416v5", "source_type": "primary", "sources": [ { "label": "Developer technical report", "url": "https://arxiv.org/html/2210.11416v5", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:17:34.586Z", "last_seen": "2026-10-29T10:17:34.586Z", "upstream_present": false, "availability": "announced", "notes": "Research model announced by its developer. External availability of this exact checkpoint was not verified; this record is excluded from release counts. Parameter count follows the paper table; later commercial API aliases are not assumed equivalent.", "announcement_date": "2022-10-20" }, { "id": "google/flan-palm-8b", "slug": "google--flan-palm-8b", "name": "Flan-PaLM 8B", "developer": "Google", "developer_id": "google", "family": "flan-palm", "description": null, "release_date": null, "date_precision": null, "open_weights": false, "license": null, "parameters_b": 8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://arxiv.org/html/2210.11416v5", "source_type": "primary", "sources": [ { "label": "Developer technical report", "url": "https://arxiv.org/html/2210.11416v5", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:17:34.586Z", "last_seen": "2026-10-29T10:17:34.586Z", "upstream_present": false, "availability": "announced", "notes": "Research model announced by its developer. External availability of this exact checkpoint was not verified; this record is excluded from release counts. Parameter count follows the paper table; later commercial API aliases are not assumed equivalent.", "announcement_date": "2022-10-20" }, { "id": "google/gemini-1.0-nano-1", "slug": "google--gemini-1.0-nano-1", "name": "Gemini 1.0 Nano-1", "developer": "Google", "developer_id": "google", "family": "gemini", "description": null, "release_date": null, "date_precision": null, "open_weights": false, "license": null, "parameters_b": 1.8, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://blog.google/technology/ai/google-gemini-ai/", "source_type": "primary", "sources": [ { "label": "Dated launch and product availability", "url": "https://blog.google/technology/ai/google-gemini-ai/", "type": "announcement" }, { "label": "Developer technical report", "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_1_report.pdf", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:26:10.259Z", "last_seen": "2026-10-29T10:26:10.259Z", "upstream_present": false, "availability": "announced", "notes": "The technical report names this trained size. Pixel availability is documented for the Nano family without mapping it to Nano-1 or Nano-2; variant-specific availability remains unverified.", "announcement_date": "2023-12-06" }, { "id": "google/gemini-1.0-nano-2", "slug": "google--gemini-1.0-nano-2", "name": "Gemini 1.0 Nano-2", "developer": "Google", "developer_id": "google", "family": "gemini", "description": null, "release_date": null, "date_precision": null, "open_weights": false, "license": null, "parameters_b": 3.25, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": null, "context_window": 32768, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://blog.google/technology/ai/google-gemini-ai/", "source_type": "primary", "sources": [ { "label": "Dated launch and product availability", "url": "https://blog.google/technology/ai/google-gemini-ai/", "type": "announcement" }, { "label": "Developer technical report", "url": "https://storage.googleapis.com/deepmind-media/gemini/gemini_1_report.pdf", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:26:10.259Z", "last_seen": "2026-10-29T10:26:10.259Z", "upstream_present": false, "availability": "announced", "notes": "The technical report names this trained size. Pixel availability is documented for the Nano family without mapping it to Nano-1 or Nano-2; variant-specific availability remains unverified.", "announcement_date": "2023-12-06" }, { "id": "google/gemini-4-carbon", "slug": "google--gemini-4-carbon", "name": "Gemini 4 Carbon", "developer": "Google", "developer_id": "google", "family": "gemini", "description": null, "release_date": null, "date_precision": null, "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/", "type": "announcement" } ], "aliases": [], "first_seen": "2026-10-29T09:44:55.983Z", "last_seen": "2026-10-29T09:44:55.983Z", "upstream_present": false, "availability": "announced", "notes": "Google announced the model on October 28 and described an upcoming restricted rollout through Fairwind. The first external availability date, parameter counts, context length and specific non-text modalities were not verified. Evaluation access does not establish general availability.", "announcement_date": "2026-10-28" }, { "id": "google/minerva-540b", "slug": "google--minerva-540b", "name": "Minerva 540B", "developer": "Google", "developer_id": "google", "family": "minerva", "description": null, "release_date": null, "date_precision": null, "open_weights": false, "license": null, "parameters_b": 540.35, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://arxiv.org/html/2206.14858v2", "source_type": "primary", "sources": [ { "label": "Developer technical report", "url": "https://arxiv.org/html/2206.14858v2", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:17:34.586Z", "last_seen": "2026-10-29T10:17:34.586Z", "upstream_present": false, "availability": "announced", "notes": "Research model announced by its developer. External availability of this exact checkpoint was not verified; this record is excluded from release counts. Parameter count follows the paper table; later commercial API aliases are not assumed equivalent.", "announcement_date": "2022-06-29" }, { "id": "google/minerva-62b", "slug": "google--minerva-62b", "name": "Minerva 62B", "developer": "Google", "developer_id": "google", "family": "minerva", "description": null, "release_date": null, "date_precision": null, "open_weights": false, "license": null, "parameters_b": 62.5, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://arxiv.org/html/2206.14858v2", "source_type": "primary", "sources": [ { "label": "Developer technical report", "url": "https://arxiv.org/html/2206.14858v2", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:17:34.586Z", "last_seen": "2026-10-29T10:17:34.586Z", "upstream_present": false, "availability": "announced", "notes": "Research model announced by its developer. External availability of this exact checkpoint was not verified; this record is excluded from release counts. Parameter count follows the paper table; later commercial API aliases are not assumed equivalent.", "announcement_date": "2022-06-29" }, { "id": "google/minerva-8b", "slug": "google--minerva-8b", "name": "Minerva 8B", "developer": "Google", "developer_id": "google", "family": "minerva", "description": null, "release_date": null, "date_precision": null, "open_weights": false, "license": null, "parameters_b": 8.63, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://arxiv.org/html/2206.14858v2", "source_type": "primary", "sources": [ { "label": "Developer technical report", "url": "https://arxiv.org/html/2206.14858v2", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:17:34.586Z", "last_seen": "2026-10-29T10:17:34.586Z", "upstream_present": false, "availability": "announced", "notes": "Research model announced by its developer. External availability of this exact checkpoint was not verified; this record is excluded from release counts. Parameter count follows the paper table; later commercial API aliases are not assumed equivalent.", "announcement_date": "2022-06-29" }, { "id": "google/palm-540b", "slug": "google--palm-540b", "name": "PaLM 540B", "developer": "Google", "developer_id": "google", "family": "palm", "description": null, "release_date": null, "date_precision": null, "open_weights": false, "license": null, "parameters_b": 540.35, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://research.google/blog/pathways-language-model-palm-scaling-to-540-billion-parameters-for-breakthrough-performance/", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://research.google/blog/pathways-language-model-palm-scaling-to-540-billion-parameters-for-breakthrough-performance/", "type": "announcement" }, { "label": "Developer technical report", "url": "https://arxiv.org/html/2204.02311v5", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:17:34.586Z", "last_seen": "2026-10-29T10:17:34.586Z", "upstream_present": false, "availability": "announced", "notes": "Research model announced by its developer. External availability of this exact checkpoint was not verified; this record is excluded from release counts. Parameter count follows the paper table; later commercial API aliases are not assumed equivalent.", "announcement_date": "2022-04-04" }, { "id": "google/palm-62b", "slug": "google--palm-62b", "name": "PaLM 62B", "developer": "Google", "developer_id": "google", "family": "palm", "description": null, "release_date": null, "date_precision": null, "open_weights": false, "license": null, "parameters_b": 62.5, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://research.google/blog/pathways-language-model-palm-scaling-to-540-billion-parameters-for-breakthrough-performance/", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://research.google/blog/pathways-language-model-palm-scaling-to-540-billion-parameters-for-breakthrough-performance/", "type": "announcement" }, { "label": "Developer technical report", "url": "https://arxiv.org/html/2204.02311v5", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:17:34.586Z", "last_seen": "2026-10-29T10:17:34.586Z", "upstream_present": false, "availability": "announced", "notes": "Research model announced by its developer. External availability of this exact checkpoint was not verified; this record is excluded from release counts. Parameter count follows the paper table; later commercial API aliases are not assumed equivalent.", "announcement_date": "2022-04-04" }, { "id": "google/palm-8b", "slug": "google--palm-8b", "name": "PaLM 8B", "developer": "Google", "developer_id": "google", "family": "palm", "description": null, "release_date": null, "date_precision": null, "open_weights": false, "license": null, "parameters_b": 8.63, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://research.google/blog/pathways-language-model-palm-scaling-to-540-billion-parameters-for-breakthrough-performance/", "source_type": "primary", "sources": [ { "label": "Release evidence", "url": "https://research.google/blog/pathways-language-model-palm-scaling-to-540-billion-parameters-for-breakthrough-performance/", "type": "announcement" }, { "label": "Developer technical report", "url": "https://arxiv.org/html/2204.02311v5", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:17:34.586Z", "last_seen": "2026-10-29T10:17:34.586Z", "upstream_present": false, "availability": "announced", "notes": "Research model announced by its developer. External availability of this exact checkpoint was not verified; this record is excluded from release counts. Parameter count follows the paper table; later commercial API aliases are not assumed equivalent.", "announcement_date": "2022-04-04" }, { "id": "huggingface/carbon-x", "slug": "huggingface--carbon-x", "name": "Carbon-X", "developer": "Hugging Face and institutional partners", "developer_id": "huggingface", "family": "carbon", "description": null, "release_date": null, "date_precision": null, "open_weights": true, "license": "Apache-2.0", "parameters_b": 14.82204672, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 40960, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/HuggingFaceBio/Carbon-X", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/HuggingFaceBio/Carbon-X/blob/17de89165e25c49e7de4f845d0e5d464469de359/README.md", "type": "model_card" }, { "label": "Official configuration", "url": "https://huggingface.co/HuggingFaceBio/Carbon-X/blob/17de89165e25c49e7de4f845d0e5d464469de359/config.json", "type": "config" }, { "label": "Official model tensor metadata", "url": "https://huggingface.co/api/models/HuggingFaceBio/Carbon-X", "type": "documentation" }, { "label": "Publisher release article", "url": "https://huggingface.co/spaces/HuggingFaceBio/Carbon-X-Release/blob/4048e8e471dfaf004a0f0fc3a8853a43872e99e7/index.html", "type": "primary" }, { "label": "Hugging Face institutional collaboration", "url": "https://github.com/huggingface/carbon", "type": "primary" }, { "label": "Publisher discussion and original checkpoint history", "url": "https://huggingface.co/HuggingFaceBio/Carbon-X/discussions/2", "type": "documentation" } ], "aliases": [ "HuggingFaceBio/Carbon-X" ], "first_seen": "2026-10-26T09:05:47.987Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": false, "availability": "public", "notes": "Hugging Face and institutional partners continued training Qwen3 on DNA and natural-language text, then applied SFT and RL. It produces English explanations and conversations, unlike the DNA-only Carbon500M/3B/8B models. Total is the official tensor count. The model card explicitly specifies40960 tokens. Thinking is disabled in post-training and evaluation; no separate reasoning or tool-call capability is asserted. First external availability has conflicting October21/22 evidence: the retained branch includes October21 weights, while main history was squashed onOctober22. Exact date is left unresolved rather than treating either upload timestamp as the launch." }, { "id": "inflection/inflection-1", "slug": "inflection--inflection-1", "name": "Inflection-1", "developer": "Inflection AI", "developer_id": "inflection", "family": "inflection", "description": null, "release_date": null, "date_precision": null, "open_weights": false, "license": "proprietary", "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": "unknown", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://web.archive.org/web/20230622173045/https://inflection.ai/inflection-1", "source_type": "primary", "sources": [ { "label": "Publisher Pi launch", "url": "https://inflection.ai/blog/pi", "type": "announcement" }, { "label": "Original publisher Inflection-1 announcement", "url": "https://web.archive.org/web/20230622173045/https://inflection.ai/inflection-1", "type": "announcement" }, { "label": "Inflection-1 technical memo, June 22 2023", "url": "https://web.archive.org/web/20230622173107/https://inflection.ai/assets/Inflection-1.pdf", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:39:55.111Z", "last_seen": "2026-10-29T10:39:55.111Z", "upstream_present": false, "availability": "announced", "notes": "The June 22 primary technical memo introduces Inflection-1 as the foundation model powering Pi and explicitly distinguishes its benchmarked base checkpoint from the adapted Pi product model. The announcement says a conversational API will be available soon; it does not verify external access to the base checkpoint. Pi was released May 2, but its date is not assigned to the unadapted base model.", "announcement_date": "2023-06-22" }, { "id": "lelapa/inkubalm-0.4b", "slug": "lelapa--inkubalm-0.4b", "name": "InkubaLM-0.4B", "developer": "Lelapa AI", "developer_id": "lelapa", "family": "InkubaLM", "description": null, "release_date": null, "date_precision": null, "open_weights": true, "license": "cc-by-nc-4.0", "parameters_b": 0.422, "active_parameters_b": null, "parameters_source": "primary", "architecture": null, "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://arxiv.org/abs/2408.17024", "source_type": "primary", "sources": [ { "label": "Official release record", "url": "https://arxiv.org/abs/2408.17024", "type": "announcement" }, { "label": "Developer model card", "url": "https://huggingface.co/lelapa/InkubaLM-0.4B", "type": "model_card" }, { "label": "Developer architecture table (Table 4)", "url": "https://arxiv.org/html/2408.17024v1", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-26T09:05:47.987Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": false, "availability": "public", "notes": "The developer technical report verifies the 0.4B model and public weights. The first external-release date remains unresolved: repository creation in May is not release evidence, and the prior developer announcement URL now returns 404." }, { "id": "maincode/matilda-platypus-1.5b-base", "slug": "maincode--matilda-platypus-1.5b-base", "name": "Matilda Platypus 1.5B Base", "developer": "Maincode", "developer_id": "maincode", "family": "matilda-platypus", "description": null, "release_date": null, "date_precision": null, "open_weights": true, "license": "Apache-2.0", "parameters_b": 1.53, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/Maincode/Matilda-Platypus-1.5B-Base", "source_type": "primary", "sources": [ { "label": "Official trained model card", "url": "https://huggingface.co/Maincode/Matilda-Platypus-1.5B-Base/blob/8a84cddb44fbee43e8596ad17725ccd2ab7d5434/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/Maincode/Matilda-Platypus-1.5B-Base/blob/8a84cddb44fbee43e8596ad17725ccd2ab7d5434/config.json", "type": "config" }, { "label": "Maincode technical report", "url": "https://huggingface.co/Maincode/Matilda-Platypus-1.5B-Base/resolve/8a84cddb44fbee43e8596ad17725ccd2ab7d5434/technical_report.pdf", "type": "paper" }, { "label": "Publisher homepage links the open-weight release", "url": "https://www.maincode.com/", "type": "primary" }, { "label": "Company model collection", "url": "https://huggingface.co/collections/Maincode/matilda-platypus", "type": "primary" } ], "aliases": [ "Maincode/Matilda-Platypus-1.5B-Base" ], "first_seen": "2026-10-26T09:05:47.987Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": false, "availability": "public", "notes": "Maincode trains a representation bridge and adapts retained Qwen3 layers, then applies reinforcement learning. This is a distinct trained base model with downloadable bridge and upper-block weights; the unchanged lower block is loaded from its licensed Qwen3 base. Total parameters are the publisher’s combined-model figure, not just this repository’s tensors. The October 2026 technical report and public collection verify the model; the exact first external availability is unresolved. October 28 repository upload timestamps are not used as release dates. Context remains unverified." }, { "id": "maincode/matilda-platypus-1.6b-base", "slug": "maincode--matilda-platypus-1.6b-base", "name": "Matilda Platypus 1.6B Base", "developer": "Maincode", "developer_id": "maincode", "family": "matilda-platypus", "description": null, "release_date": null, "date_precision": null, "open_weights": true, "license": "Apache-2.0", "parameters_b": 1.6, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/Maincode/Matilda-Platypus-1.6B-Base", "source_type": "primary", "sources": [ { "label": "Official trained model card", "url": "https://huggingface.co/Maincode/Matilda-Platypus-1.6B-Base/blob/53131bfc7ef529d25db629b987295af4cf6c8841/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/Maincode/Matilda-Platypus-1.6B-Base/blob/53131bfc7ef529d25db629b987295af4cf6c8841/config.json", "type": "config" }, { "label": "Maincode technical report", "url": "https://huggingface.co/Maincode/Matilda-Platypus-1.6B-Base/resolve/53131bfc7ef529d25db629b987295af4cf6c8841/technical_report.pdf", "type": "paper" }, { "label": "Publisher homepage links the open-weight release", "url": "https://www.maincode.com/", "type": "primary" }, { "label": "Company model collection", "url": "https://huggingface.co/collections/Maincode/matilda-platypus", "type": "primary" } ], "aliases": [ "Maincode/Matilda-Platypus-1.6B-Base" ], "first_seen": "2026-10-26T09:05:47.987Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": false, "availability": "public", "notes": "Maincode trains a representation bridge and adapts retained Qwen3 layers, then applies reinforcement learning. This is a distinct trained base model with downloadable bridge, upper-block and adapted lower-block weights; unlike the other sizes, its lower block also receives low-rank updates and ships in lower_block/. Total parameters are the publisher’s combined-model figure, not just this repository’s tensors. The October 2026 technical report and public collection verify the model; the exact first external availability is unresolved. October 28 repository upload timestamps are not used as release dates. Context remains unverified." }, { "id": "maincode/matilda-platypus-11b-base", "slug": "maincode--matilda-platypus-11b-base", "name": "Matilda Platypus 11B Base", "developer": "Maincode", "developer_id": "maincode", "family": "matilda-platypus", "description": null, "release_date": null, "date_precision": null, "open_weights": true, "license": "Apache-2.0", "parameters_b": 11.42, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/Maincode/Matilda-Platypus-11B-Base", "source_type": "primary", "sources": [ { "label": "Official trained model card", "url": "https://huggingface.co/Maincode/Matilda-Platypus-11B-Base/blob/0a18a8760e332d6f64f3b998859ea68bb98f2994/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/Maincode/Matilda-Platypus-11B-Base/blob/0a18a8760e332d6f64f3b998859ea68bb98f2994/config.json", "type": "config" }, { "label": "Maincode technical report", "url": "https://huggingface.co/Maincode/Matilda-Platypus-11B-Base/resolve/0a18a8760e332d6f64f3b998859ea68bb98f2994/technical_report.pdf", "type": "paper" }, { "label": "Publisher homepage links the open-weight release", "url": "https://www.maincode.com/", "type": "primary" }, { "label": "Company model collection", "url": "https://huggingface.co/collections/Maincode/matilda-platypus", "type": "primary" } ], "aliases": [ "Maincode/Matilda-Platypus-11B-Base" ], "first_seen": "2026-10-26T09:05:47.987Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": false, "availability": "public", "notes": "Maincode trains a representation bridge and adapts retained Qwen3 layers, then applies reinforcement learning. This is a distinct trained base model with downloadable bridge and upper-block weights; the unchanged lower block is loaded from its licensed Qwen3 base. Total parameters are the publisher’s combined-model figure, not just this repository’s tensors. The October 2026 technical report and public collection verify the model; the exact first external availability is unresolved. October 28 repository upload timestamps are not used as release dates. Context remains unverified." }, { "id": "maincode/matilda-platypus-12b-base", "slug": "maincode--matilda-platypus-12b-base", "name": "Matilda Platypus 12B Base", "developer": "Maincode", "developer_id": "maincode", "family": "matilda-platypus", "description": null, "release_date": null, "date_precision": null, "open_weights": true, "license": "Apache-2.0", "parameters_b": 12.57, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/Maincode/Matilda-Platypus-12B-Base", "source_type": "primary", "sources": [ { "label": "Official trained model card", "url": "https://huggingface.co/Maincode/Matilda-Platypus-12B-Base/blob/a273d40962e6d13b5fdd199a1dac15e5ebdb1081/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/Maincode/Matilda-Platypus-12B-Base/blob/a273d40962e6d13b5fdd199a1dac15e5ebdb1081/config.json", "type": "config" }, { "label": "Maincode technical report", "url": "https://huggingface.co/Maincode/Matilda-Platypus-12B-Base/resolve/a273d40962e6d13b5fdd199a1dac15e5ebdb1081/technical_report.pdf", "type": "paper" }, { "label": "Publisher homepage links the open-weight release", "url": "https://www.maincode.com/", "type": "primary" }, { "label": "Company model collection", "url": "https://huggingface.co/collections/Maincode/matilda-platypus", "type": "primary" } ], "aliases": [ "Maincode/Matilda-Platypus-12B-Base" ], "first_seen": "2026-10-26T09:05:47.987Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": false, "availability": "public", "notes": "Maincode trains a representation bridge and adapts retained Qwen3 layers, then applies reinforcement learning. This is a distinct trained base model with downloadable bridge and upper-block weights; the unchanged lower block is loaded from its licensed Qwen3 base. Total parameters are the publisher’s combined-model figure, not just this repository’s tensors. The October 2026 technical report and public collection verify the model; the exact first external availability is unresolved. October 28 repository upload timestamps are not used as release dates. Context remains unverified." }, { "id": "maincode/matilda-platypus-13b-base", "slug": "maincode--matilda-platypus-13b-base", "name": "Matilda Platypus 13B Base", "developer": "Maincode", "developer_id": "maincode", "family": "matilda-platypus", "description": null, "release_date": null, "date_precision": null, "open_weights": true, "license": "Apache-2.0", "parameters_b": 13.51, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/Maincode/Matilda-Platypus-13B-Base", "source_type": "primary", "sources": [ { "label": "Official trained model card", "url": "https://huggingface.co/Maincode/Matilda-Platypus-13B-Base/blob/fa393417bc7a67256965fca50853aa60ef2915a7/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/Maincode/Matilda-Platypus-13B-Base/blob/fa393417bc7a67256965fca50853aa60ef2915a7/config.json", "type": "config" }, { "label": "Maincode technical report", "url": "https://huggingface.co/Maincode/Matilda-Platypus-13B-Base/resolve/fa393417bc7a67256965fca50853aa60ef2915a7/technical_report.pdf", "type": "paper" }, { "label": "Publisher homepage links the open-weight release", "url": "https://www.maincode.com/", "type": "primary" }, { "label": "Company model collection", "url": "https://huggingface.co/collections/Maincode/matilda-platypus", "type": "primary" } ], "aliases": [ "Maincode/Matilda-Platypus-13B-Base" ], "first_seen": "2026-10-26T09:05:47.987Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": false, "availability": "public", "notes": "Maincode trains a representation bridge and adapts retained Qwen3 layers, then applies reinforcement learning. This is a distinct trained base model with downloadable bridge and upper-block weights; the unchanged lower block is loaded from its licensed Qwen3 base. Total parameters are the publisher’s combined-model figure, not just this repository’s tensors. The October 2026 technical report and public collection verify the model; the exact first external availability is unresolved. October 28 repository upload timestamps are not used as release dates. Context remains unverified." }, { "id": "maincode/matilda-platypus-5b-base", "slug": "maincode--matilda-platypus-5b-base", "name": "Matilda Platypus 5B Base", "developer": "Maincode", "developer_id": "maincode", "family": "matilda-platypus", "description": null, "release_date": null, "date_precision": null, "open_weights": true, "license": "Apache-2.0", "parameters_b": 5.48, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/Maincode/Matilda-Platypus-5B-Base", "source_type": "primary", "sources": [ { "label": "Official trained model card", "url": "https://huggingface.co/Maincode/Matilda-Platypus-5B-Base/blob/dae3b85f85cc58c945d6f80e09519d64ccebbe7c/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/Maincode/Matilda-Platypus-5B-Base/blob/dae3b85f85cc58c945d6f80e09519d64ccebbe7c/config.json", "type": "config" }, { "label": "Maincode technical report", "url": "https://huggingface.co/Maincode/Matilda-Platypus-5B-Base/resolve/dae3b85f85cc58c945d6f80e09519d64ccebbe7c/technical_report.pdf", "type": "paper" }, { "label": "Publisher homepage links the open-weight release", "url": "https://www.maincode.com/", "type": "primary" }, { "label": "Company model collection", "url": "https://huggingface.co/collections/Maincode/matilda-platypus", "type": "primary" } ], "aliases": [ "Maincode/Matilda-Platypus-5B-Base" ], "first_seen": "2026-10-26T09:05:47.987Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": false, "availability": "public", "notes": "Maincode trains a representation bridge and adapts retained Qwen3 layers, then applies reinforcement learning. This is a distinct trained base model with downloadable bridge and upper-block weights; the unchanged lower block is loaded from its licensed Qwen3 base. Total parameters are the publisher’s combined-model figure, not just this repository’s tensors. The October 2026 technical report and public collection verify the model; the exact first external availability is unresolved. October 28 repository upload timestamps are not used as release dates. Context remains unverified." }, { "id": "maincode/matilda-platypus-6b-base", "slug": "maincode--matilda-platypus-6b-base", "name": "Matilda Platypus 6B Base", "developer": "Maincode", "developer_id": "maincode", "family": "matilda-platypus", "description": null, "release_date": null, "date_precision": null, "open_weights": true, "license": "Apache-2.0", "parameters_b": 6.22, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/Maincode/Matilda-Platypus-6B-Base", "source_type": "primary", "sources": [ { "label": "Official trained model card", "url": "https://huggingface.co/Maincode/Matilda-Platypus-6B-Base/blob/d190286406c846612f957a8666d207f0f4022e0c/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/Maincode/Matilda-Platypus-6B-Base/blob/d190286406c846612f957a8666d207f0f4022e0c/config.json", "type": "config" }, { "label": "Maincode technical report", "url": "https://huggingface.co/Maincode/Matilda-Platypus-6B-Base/resolve/d190286406c846612f957a8666d207f0f4022e0c/technical_report.pdf", "type": "paper" }, { "label": "Publisher homepage links the open-weight release", "url": "https://www.maincode.com/", "type": "primary" }, { "label": "Company model collection", "url": "https://huggingface.co/collections/Maincode/matilda-platypus", "type": "primary" } ], "aliases": [ "Maincode/Matilda-Platypus-6B-Base" ], "first_seen": "2026-10-26T09:05:47.987Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": false, "availability": "public", "notes": "Maincode trains a representation bridge and adapts retained Qwen3 layers, then applies reinforcement learning. This is a distinct trained base model with downloadable bridge and upper-block weights; the unchanged lower block is loaded from its licensed Qwen3 base. Total parameters are the publisher’s combined-model figure, not just this repository’s tensors. The October 2026 technical report and public collection verify the model; the exact first external availability is unresolved. October 28 repository upload timestamps are not used as release dates. Context remains unverified." }, { "id": "maincode/matilda-platypus-7b-base", "slug": "maincode--matilda-platypus-7b-base", "name": "Matilda Platypus 7B Base", "developer": "Maincode", "developer_id": "maincode", "family": "matilda-platypus", "description": null, "release_date": null, "date_precision": null, "open_weights": true, "license": "Apache-2.0", "parameters_b": 6.96, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/Maincode/Matilda-Platypus-7B-Base", "source_type": "primary", "sources": [ { "label": "Official trained model card", "url": "https://huggingface.co/Maincode/Matilda-Platypus-7B-Base/blob/b2708113cdffabdef29e03ae903d14a80b0a3ffa/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/Maincode/Matilda-Platypus-7B-Base/blob/b2708113cdffabdef29e03ae903d14a80b0a3ffa/config.json", "type": "config" }, { "label": "Maincode technical report", "url": "https://huggingface.co/Maincode/Matilda-Platypus-7B-Base/resolve/b2708113cdffabdef29e03ae903d14a80b0a3ffa/technical_report.pdf", "type": "paper" }, { "label": "Publisher homepage links the open-weight release", "url": "https://www.maincode.com/", "type": "primary" }, { "label": "Company model collection", "url": "https://huggingface.co/collections/Maincode/matilda-platypus", "type": "primary" } ], "aliases": [ "Maincode/Matilda-Platypus-7B-Base" ], "first_seen": "2026-10-26T09:05:47.987Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": false, "availability": "public", "notes": "Maincode trains a representation bridge and adapts retained Qwen3 layers, then applies reinforcement learning. This is a distinct trained base model with downloadable bridge and upper-block weights; the unchanged lower block is loaded from its licensed Qwen3 base. Total parameters are the publisher’s combined-model figure, not just this repository’s tensors. The October 2026 technical report and public collection verify the model; the exact first external availability is unresolved. October 28 repository upload timestamps are not used as release dates. Context remains unverified." }, { "id": "meta/opt-iml-175b", "slug": "meta--opt-iml-175b", "name": "OPT-IML 175B", "developer": "Meta", "developer_id": "meta", "family": "opt", "description": null, "release_date": null, "date_precision": null, "open_weights": false, "license": "OPT-175B License Agreement", "parameters_b": 175, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/facebookresearch/metaseq/blob/e6f2ad17f5756ebce769e2a53ade0e5d93b94008/projects/OPT-IML/README.md", "source_type": "primary", "sources": [ { "label": "Original December 22 publisher release page", "url": "https://github.com/facebookresearch/metaseq/blob/e6f2ad17f5756ebce769e2a53ade0e5d93b94008/projects/OPT-IML/README.md", "type": "documentation" }, { "label": "Original model paper, variants and training sequence length", "url": "https://arxiv.org/html/2212.12017v1", "type": "paper" }, { "label": "January 21 instructions identify both restricted 175B checkpoints", "url": "https://github.com/facebookresearch/metaseq/blob/1de510e3b714384d4ebaf9782216e45e361dbaab/projects/OPT-IML/download_optiml175b.md", "type": "documentation" }, { "label": "January 24 public request form addition", "url": "https://github.com/facebookresearch/metaseq/commit/f4edbe4cf62f8abaa9c1ef51bc03590a2f9d2440", "type": "documentation" }, { "label": "Model weights license linked by publisher", "url": "https://github.com/facebookresearch/metaseq/blob/main/projects/OPT/MODEL_LICENSE.md", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:57:14.174Z", "last_seen": "2026-10-29T10:57:14.174Z", "upstream_present": false, "availability": "limited", "notes": "Original publisher instruction-tuned OPT checkpoint. December 22 documentation says 175B request access is coming soon. January 21 download instructions identify both regular and Max 175B weights; the request form appears January 24. These sources verify the model and restricted weight-distribution process, but do not establish when the first external recipient obtained access, so release_date remains null. Ordinary OPT-IML uses approximately 1,500 tasks. Parameters and 2048-token training sequence are documented by the original paper.", "announcement_date": "2022-12-22" }, { "id": "meta/opt-iml-max-175b", "slug": "meta--opt-iml-max-175b", "name": "OPT-IML-Max 175B", "developer": "Meta", "developer_id": "meta", "family": "opt", "description": null, "release_date": null, "date_precision": null, "open_weights": false, "license": "OPT-175B License Agreement", "parameters_b": 175, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://github.com/facebookresearch/metaseq/blob/e6f2ad17f5756ebce769e2a53ade0e5d93b94008/projects/OPT-IML/README.md", "source_type": "primary", "sources": [ { "label": "Original December 22 publisher release page", "url": "https://github.com/facebookresearch/metaseq/blob/e6f2ad17f5756ebce769e2a53ade0e5d93b94008/projects/OPT-IML/README.md", "type": "documentation" }, { "label": "Original model paper, variants and training sequence length", "url": "https://arxiv.org/html/2212.12017v1", "type": "paper" }, { "label": "January 21 instructions identify both restricted 175B checkpoints", "url": "https://github.com/facebookresearch/metaseq/blob/1de510e3b714384d4ebaf9782216e45e361dbaab/projects/OPT-IML/download_optiml175b.md", "type": "documentation" }, { "label": "January 24 public request form addition", "url": "https://github.com/facebookresearch/metaseq/commit/f4edbe4cf62f8abaa9c1ef51bc03590a2f9d2440", "type": "documentation" }, { "label": "Model weights license linked by publisher", "url": "https://github.com/facebookresearch/metaseq/blob/main/projects/OPT/MODEL_LICENSE.md", "type": "license" } ], "aliases": [], "first_seen": "2026-10-29T10:57:14.174Z", "last_seen": "2026-10-29T10:57:14.174Z", "upstream_present": false, "availability": "limited", "notes": "Original publisher instruction-tuned OPT checkpoint. December 22 documentation says 175B request access is coming soon. January 21 download instructions identify both regular and Max 175B weights; the request form appears January 24. These sources verify the model and restricted weight-distribution process, but do not establish when the first external recipient obtained access, so release_date remains null. Max uses approximately 2,000 tasks. Parameters and 2048-token training sequence are documented by the original paper.", "announcement_date": "2022-12-22" }, { "id": "microsoft/mai-base-2", "slug": "microsoft--mai-base-2", "name": "MAI Base 2", "developer": "Microsoft", "developer_id": "microsoft", "family": "mai-base", "description": "Microsoft pretrained mixture-of-experts language model undergoing post-training.", "release_date": null, "date_precision": null, "open_weights": false, "license": null, "parameters_b": 2800, "active_parameters_b": 80, "parameters_source": "primary_source", "architecture": "moe", "context_window": null, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://microsoft.ai/news/frontier-pre-training-at-microsoft-ai/", "source_type": "primary", "sources": [ { "label": "Publisher announcement", "url": "https://microsoft.ai/news/frontier-pre-training-at-microsoft-ai/", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:13:05.399Z", "last_seen": "2026-10-30T07:13:05.399Z", "upstream_present": false, "availability": "announced", "notes": "Microsoft announced completed pretraining on October15,2026, with2.8T total and80B active parameters. The announcement says post-training is underway and discusses future partner weight licensing; first external availability and a downloadablecheckpoint were not verified. Context, license and tool/thinking capabilities remain unverified.", "announcement_date": "2026-10-15" }, { "id": "microsoft/phi-1", "slug": "microsoft--phi-1", "name": "Phi-1", "developer": "Microsoft", "developer_id": "microsoft", "family": "phi", "description": null, "release_date": null, "date_precision": null, "open_weights": true, "license": "MIT", "parameters_b": 1.3, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/microsoft/phi-1", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/microsoft/phi-1/blob/d4c0adcb065e84e00ca814e35cba3012ea9841ab/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/microsoft/phi-1/blob/d4c0adcb065e84e00ca814e35cba3012ea9841ab/config.json", "type": "config" }, { "label": "Original developer paper", "url": "https://arxiv.org/abs/2306.11644v1", "type": "paper" } ], "aliases": [], "first_seen": "2026-10-29T10:36:54.148Z", "last_seen": "2026-10-29T10:36:54.148Z", "upstream_present": false, "availability": "public", "notes": "Microsoft pretrained this Python code model. The official card states 1.3B parameters. Public weights are verified, but their first release date remains unverified. The June 20, 2023 paper announced planned open sourcing, so its submission date is recorded only as the announcement.", "announcement_date": "2023-06-20" }, { "id": "openai/gpt-audio-2025-08-28", "slug": "openai--gpt-audio-2025-08-28", "name": "GPT-audio-2025-08-28", "developer": "OpenAI", "developer_id": "openai", "family": "gpt-audio", "description": null, "release_date": null, "date_precision": null, "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": 128000, "modalities": { "input": [ "text", "audio" ], "output": [ "text", "audio" ] }, "reasoning": null, "tool_call": null, "source_url": "https://developers.openai.com/api/docs/changelog", "source_type": "primary", "sources": [ { "label": "Dated publisher API releases", "url": "https://developers.openai.com/api/docs/changelog", "type": "documentation" }, { "label": "Publisher model documentation and snapshot mapping", "url": "https://developers.openai.com/api/docs/models/gpt-audio", "type": "documentation" } ], "aliases": [ "openai/gpt-audio" ], "first_seen": "2026-10-26T09:05:47.987Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": false, "availability": "public", "notes": "Original publisher conversational audio model. The exact dated snapshot is documented, but the snapshot suffix is not treated as proof of public release day." }, { "id": "swiss-ai/apertus-v1.5-8b-base", "slug": "swiss-ai--apertus-v1.5-8b-base", "name": "Apertus 1.5 8B Base", "developer": "Swiss AI Initiative", "developer_id": "swiss-ai", "family": "apertus", "description": null, "release_date": null, "date_precision": null, "open_weights": true, "license": "apache-2.0", "parameters_b": 8.839738945, "active_parameters_b": null, "parameters_source": "https://huggingface.co/api/models/swiss-ai/Apertus-v1.5-8B-base/revision/1e28c426548b1853c11e9dbd386e27accbdad19f", "architecture": "transformer", "context_window": 262144, "modalities": { "input": [ "text", "image", "audio" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/swiss-ai/Apertus-v1.5-8B-base", "source_type": "primary", "sources": [ { "label": "Publisher release notes", "url": "https://huggingface.co/blog/swiss-ai/apertus-v15-release-notes", "type": "primary" }, { "label": "Official model card", "url": "https://huggingface.co/swiss-ai/Apertus-v1.5-8B-base/blob/1e28c426548b1853c11e9dbd386e27accbdad19f/README.md", "type": "primary" }, { "label": "Official checkpoint parameter metadata", "url": "https://huggingface.co/api/models/swiss-ai/Apertus-v1.5-8B-base/revision/1e28c426548b1853c11e9dbd386e27accbdad19f", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-26T09:05:47.987Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": false, "availability": "public", "notes": "Total is the official safetensors parameter count, including released multimodal components, rather than the nominal size in the name. Model card confirms 262144 context and text/image/audio inputs. Downloads require accepting the publisher's license and usage policy; raw config retrieval returned401 during review. Base checkpoint supports plain-text completion without a chat template. Its exact first external release date is unresolved; the family's July24 date is not inherited because this repository was created later." }, { "id": "tencent/hy4", "slug": "tencent--hy4", "name": "Hy4", "developer": "Tencent", "developer_id": "tencent", "family": "hy4", "description": "Upcoming Hunyuan model for long-running productivity tasks with visual understanding.", "release_date": null, "date_precision": null, "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": true, "tool_call": null, "source_url": "https://www.tencent.com/tencent-cloud-accelerates-international-growth-advances-enterprise-ai-and-expands-global-partnerships/", "source_type": "primary", "sources": [ { "label": "Dated Tencent announcement", "url": "https://www.tencent.com/tencent-cloud-accelerates-international-growth-advances-enterprise-ai-and-expands-global-partnerships/", "type": "primary" }, { "label": "Original Chinese announcement", "url": "https://www.tencent.com/zh-cn/tencent-cloud-accelerates-international-growth-advances-enterprise-ai-and-expands-global-partnerships/", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:13:45.595Z", "last_seen": "2026-10-30T07:13:45.595Z", "upstream_present": false, "announcement_date": "2026-10-29", "availability": "announced", "notes": "Tencent’s October 29 announcement describes the general Hy4 and Hy4 Flash as forthcoming. External access, parameters, context and weight license remain unverified. Do not inherit the August Hy4 preview checkpoint specifications. Image input records the explicitly announced visual understanding; video support is not established." }, { "id": "tencent/hy4-flash", "slug": "tencent--hy4-flash", "name": "Hy4 Flash", "developer": "Tencent", "developer_id": "tencent", "family": "hy4", "description": "Upcoming Hunyuan model focused on lower-cost visual and agent tasks.", "release_date": null, "date_precision": null, "open_weights": false, "license": null, "parameters_b": null, "active_parameters_b": null, "parameters_source": null, "architecture": null, "context_window": null, "modalities": { "input": [ "text", "image" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://www.tencent.com/tencent-cloud-accelerates-international-growth-advances-enterprise-ai-and-expands-global-partnerships/", "source_type": "primary", "sources": [ { "label": "Dated Tencent announcement", "url": "https://www.tencent.com/tencent-cloud-accelerates-international-growth-advances-enterprise-ai-and-expands-global-partnerships/", "type": "primary" }, { "label": "Original Chinese announcement", "url": "https://www.tencent.com/zh-cn/tencent-cloud-accelerates-international-growth-advances-enterprise-ai-and-expands-global-partnerships/", "type": "primary" } ], "aliases": [], "first_seen": "2026-10-30T07:13:45.595Z", "last_seen": "2026-10-30T07:13:45.595Z", "upstream_present": false, "announcement_date": "2026-10-29", "availability": "announced", "notes": "Tencent’s October 29 announcement describes the general Hy4 and Hy4 Flash as forthcoming. External access, parameters, context and weight license remain unverified. Do not inherit the August Hy4 preview checkpoint specifications. Image input records the explicitly announced visual understanding; video support is not established." }, { "id": "tiiuae/falcon-40b-instruct", "slug": "tiiuae--falcon-40b-instruct", "name": "Falcon-40B-Instruct", "developer": "Technology Innovation Institute", "developer_id": "tiiuae", "family": "falcon", "description": null, "release_date": null, "date_precision": null, "open_weights": true, "license": "Apache-2.0", "parameters_b": 40, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/tiiuae/falcon-40b-instruct", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/tiiuae/falcon-40b-instruct/blob/ecb78d97ac356d098e79f0db222c9ce7c5d9ee5f/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/tiiuae/falcon-40b-instruct/blob/ecb78d97ac356d098e79f0db222c9ce7c5d9ee5f/config.json", "type": "config" }, { "label": "Contemporaneous hosting-team availability guide", "url": "https://huggingface.co/blog/falcon", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-26T09:05:47.987Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": false, "availability": "public", "notes": "Public weights and specifications are verified in the official model card. The first public release date remains unverified; the June 5, 2023 Hugging Face launch guide already lists this model. Repository upload dates are not treated as release dates." }, { "id": "tiiuae/falcon-7b", "slug": "tiiuae--falcon-7b", "name": "Falcon-7B", "developer": "Technology Innovation Institute", "developer_id": "tiiuae", "family": "falcon", "description": null, "release_date": null, "date_precision": null, "open_weights": true, "license": "Apache-2.0", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/tiiuae/falcon-7b", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/tiiuae/falcon-7b/blob/ec89142b67d748a1865ea4451372db8313ada0d8/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/tiiuae/falcon-7b/blob/ec89142b67d748a1865ea4451372db8313ada0d8/config.json", "type": "config" }, { "label": "Contemporaneous hosting-team availability guide", "url": "https://huggingface.co/blog/falcon", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-26T09:05:47.987Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": false, "availability": "public", "notes": "Public weights and specifications are verified in the official model card. The first public release date remains unverified; the June 5, 2023 Hugging Face launch guide already lists this model. Repository upload dates are not treated as release dates." }, { "id": "tiiuae/falcon-7b-instruct", "slug": "tiiuae--falcon-7b-instruct", "name": "Falcon-7B-Instruct", "developer": "Technology Innovation Institute", "developer_id": "tiiuae", "family": "falcon", "description": null, "release_date": null, "date_precision": null, "open_weights": true, "license": "Apache-2.0", "parameters_b": 7, "active_parameters_b": null, "parameters_source": "primary_source", "architecture": "dense", "context_window": 2048, "modalities": { "input": [ "text" ], "output": [ "text" ] }, "reasoning": null, "tool_call": null, "source_url": "https://huggingface.co/tiiuae/falcon-7b-instruct", "source_type": "primary", "sources": [ { "label": "Official model card", "url": "https://huggingface.co/tiiuae/falcon-7b-instruct/blob/8782b5c5d8c9290412416618f36a133653e85285/README.md", "type": "model_card" }, { "label": "Official model configuration", "url": "https://huggingface.co/tiiuae/falcon-7b-instruct/blob/8782b5c5d8c9290412416618f36a133653e85285/config.json", "type": "config" }, { "label": "Contemporaneous hosting-team availability guide", "url": "https://huggingface.co/blog/falcon", "type": "documentation" } ], "aliases": [], "first_seen": "2026-10-26T09:05:47.987Z", "last_seen": "2026-10-26T09:05:47.987Z", "upstream_present": false, "availability": "public", "notes": "Public weights and specifications are verified in the official model card. The first public release date remains unverified; the June 5, 2023 Hugging Face launch guide already lists this model. Repository upload dates are not treated as release dates." } ] }