{
  "meta": {
    "attribution": "Anchor Terminal (https://www.anchorterminal.com)",
    "attributionRequired": true,
    "credit": "CC BY 4.0, credit Anchor Terminal and link the page or API address you took it from (https://www.anchorterminal.com/terms/#data)",
    "docs": "https://www.anchorterminal.com/docs/",
    "generatedAt": "2026-11-21",
    "license": "CC-BY-4.0",
    "method": "https://www.anchorterminal.com/benchmark/",
    "methodology": "1.0",
    "openapi": "https://www.anchorterminal.com/openapi.json",
    "preview": false,
    "run": "2026-10-01",
    "runLabel": "October 2026 research run",
    "terms": "https://www.anchorterminal.com/terms/#data"
  },
  "tool": {
    "slug": "inkling-2-mini",
    "name": "Inkling 2 Mini",
    "vendor": "Thinking Machines Lab",
    "vendorUrl": "https://thinkingmachines.ai",
    "company": "thinkingmachines.ai",
    "kind": "weights",
    "category": "open-weights",
    "summary": "Thinking Machines Lab's small open-weight model, a looped mixture-of-experts transformer with 26.4B total and 3.4B active parameters, taking text and images, with adjustable thinking effort and a 1M-token context, in one 20 GB checkpoint under Apache 2.0.",
    "url": "https://www.anchorterminal.com/tools/inkling-2-mini",
    "markdownUrl": "https://www.anchorterminal.com/tools/inkling-2-mini.md",
    "slimMarkdownUrl": "https://www.anchorterminal.com/tools/inkling-2-mini.min.md",
    "cardUrl": "https://www.anchorterminal.com/tools/inkling-2-mini.card.md",
    "jsonUrl": "https://www.anchorterminal.com/api/v1/tools/inkling-2-mini.json",
    "dns": "inkling-2-mini.grade.anchorterminal.com",
    "repo": "https://huggingface.co/thinkingmachines/Inkling-2-Mini",
    "license": "Apache-2.0",
    "transports": [],
    "packages": [
      {
        "registry": "huggingface",
        "name": "thinkingmachines/Inkling-2-Mini"
      }
    ],
    "auth": "none",
    "authNotes": "The weights download from Hugging Face with no account, token or terms to accept, and the repository isn't gated (checked 2026-11-20). Thinking Machines' Model Acceptable Use Policy, linked from the card, says that downloading or using the weights binds you to it.",
    "pricing": "free",
    "pricingNotes": "Free to download and use, commercial use included, under Apache 2.0 with Thinking Machines' Model Acceptable Use Policy on top. Running it yourself costs the hardware, about 20 GB of files, and Thinking Machines recommends 32 GB or more of memory. Thinking Machines' Tinker samples from it at $0.15 per million prefill tokens, $0.015 cached, and $0.50 per million sampled, through the Tinker SDK; its serverless OpenAI-compatible inference doesn't serve this model yet (read 2026-11-20).",
    "priceSummary": "Free",
    "where": "local",
    "x402": {
      "level": "no",
      "evidence": "Weights you download and run yourself, so there is nothing to pay per call and no payment protocol (checked 2026-11-20).",
      "endpoints": []
    },
    "toolCount": null,
    "popularity": {
      "githubStars": null,
      "npmWeekly": null,
      "pypiWeekly": null,
      "asOf": "2026-11-20"
    },
    "docsUrl": "https://thinkingmachines.ai/model-card/inkling-2-mini/",
    "capabilities": [
      "model.open-weights"
    ],
    "tags": [
      "open-weights",
      "model",
      "moe",
      "multimodal",
      "vision",
      "reasoning",
      "tool-calling",
      "apache-2.0",
      "safetensors",
      "nvfp4",
      "looped",
      "self-hosted",
      "local",
      "new"
    ],
    "lastRelease": "2026-11-20",
    "graded": true,
    "anchor": {
      "graded": true,
      "score": 61,
      "grade": "C",
      "agentReady": false,
      "rank": 0,
      "ranked": false,
      "notRankedWhy": "An open-weight model, graded by the open-weight checklist and ranked among open-weight models, not against tools",
      "rankOf": 1563,
      "categoryRank": 55,
      "methodology": "1.0",
      "run": "2026-10-01",
      "scores": {
        "interface": 85,
        "reach": 60,
        "risk": 70,
        "upkeep": 57
      },
      "breakdown": [
        {
          "key": "capability",
          "name": "Capability",
          "weight": 35,
          "effectiveWeight": 0,
          "pending": true,
          "points": 0,
          "reason": "Pending for this listing. Nobody independent has published results for it in at least two areas yet, so its weight is shared across the other four and the total is capped at 61 until they do or our own runs start."
        },
        {
          "key": "reach",
          "name": "Reach",
          "weight": 25,
          "effectiveWeight": 38.46,
          "score": 60,
          "points": 23.08,
          "reason": "Memory 25 of 30, ggml-org's IQ4_NL GGUF with its vision projector is 17.98 GB, about 21.6 GB with a fifth added. Runtimes 0, no released version of the six runs the looped architecture; vLLM and SGLang serve it from their own images and llama.cpp has it on master only. Builds 10, one checkpoint from the publisher in mixed NVFP4 and MXFP8. Terms 10, Apache 2.0 with a binding acceptable use policy on top, read as a custom licence. Download 10, no gate. Hosted 5, Thinking Machines' own Tinker samples from it. 25 + 0 + 10 + 10 + 10 + 5 = 60.",
          "checked": "2026-11-20"
        },
        {
          "key": "risk",
          "name": "Risk",
          "weight": 20,
          "effectiveWeight": 30.77,
          "score": 70,
          "points": 21.54,
          "reason": "File safety 25, safetensors only and no repository code to run. Provenance 15, the files are in the organisation Thinking Machines links to, with a SHA-256 on every LFS file. Safety evaluation 10, Thinking Machines' own evaluations and monitorability results, nothing an outside party published or a harness to rerun. Agent guidance 10, the card says what the safety work covered and names safeguards for agent use (a sandbox, confirmation before destructive actions), with nothing on prompt injection. Licence risk 5, 10 off for a use policy that reaches beyond the law and that can change at any time. Security 5, security.txt and a disclosure policy. 25 + 15 + 10 + 10 + 5 + 5 = 70.",
          "checked": "2026-11-20"
        },
        {
          "key": "interface",
          "name": "Agent interface",
          "weight": 10,
          "effectiveWeight": 15.38,
          "score": 85,
          "points": 13.08,
          "reason": "Chat template 20, chat_template.jinja in the repository. Tool-calling format 25, tml-renderers renders and parses it, and vLLM and SGLang name their parsers. Trained tool calling 20, several calls in one turn. Structured output 0, not documented. Reasoning control 10, reasoning_effort per request. Context 10, 1M trained. 20 + 25 + 20 + 0 + 10 + 10 = 85.",
          "checked": "2026-11-20"
        },
        {
          "key": "upkeep",
          "name": "Documentation and upkeep",
          "weight": 10,
          "effectiveWeight": 15.38,
          "score": 57,
          "points": 8.77,
          "reason": "Model card 17, no knowledge cut-off. Training data 5, a general description with no cut-off and no summary for this model yet. Last change 20, uploaded today. Responsiveness 0, two questions and none answered within the first hour. Version policy 5, a first release. File health 10, the files and the card agree. 17 + 5 + 20 + 0 + 5 + 10 = 57.",
          "checked": "2026-11-20"
        }
      ],
      "assessment": {
        "date": "2026-11-20",
        "basis": "public evidence",
        "confidence": "low",
        "notes": {
          "capability": "Pending for this listing. Independent results were found in 0 areas (none), and Capability needs 2. Its weight is shared across the other four categories and the total is capped at 61 until there are more, or our own runs start.",
          "interface": "Chat template 20, chat_template.jinja in the repository. Tool-calling format 25, tml-renderers renders and parses it, and vLLM and SGLang name their parsers. Trained tool calling 20, several calls in one turn. Structured output 0, not documented. Reasoning control 10, reasoning_effort per request. Context 10, 1M trained. 20 + 25 + 20 + 0 + 10 + 10 = 85.",
          "reach": "Memory 25 of 30, ggml-org's IQ4_NL GGUF with its vision projector is 17.98 GB, about 21.6 GB with a fifth added. Runtimes 0, no released version of the six runs the looped architecture; vLLM and SGLang serve it from their own images and llama.cpp has it on master only. Builds 10, one checkpoint from the publisher in mixed NVFP4 and MXFP8. Terms 10, Apache 2.0 with a binding acceptable use policy on top, read as a custom licence. Download 10, no gate. Hosted 5, Thinking Machines' own Tinker samples from it. 25 + 0 + 10 + 10 + 10 + 5 = 60.",
          "risk": "File safety 25, safetensors only and no repository code to run. Provenance 15, the files are in the organisation Thinking Machines links to, with a SHA-256 on every LFS file. Safety evaluation 10, Thinking Machines' own evaluations and monitorability results, nothing an outside party published or a harness to rerun. Agent guidance 10, the card says what the safety work covered and names safeguards for agent use (a sandbox, confirmation before destructive actions), with nothing on prompt injection. Licence risk 5, 10 off for a use policy that reaches beyond the law and that can change at any time. Security 5, security.txt and a disclosure policy. 25 + 15 + 10 + 10 + 5 + 5 = 70.",
          "upkeep": "Model card 17, no knowledge cut-off. Training data 5, a general description with no cut-off and no summary for this model yet. Last change 20, uploaded today. Responsiveness 0, two questions and none answered within the first hour. Version policy 5, a first release. File health 10, the files and the card agree. 17 + 5 + 20 + 0 + 5 + 10 = 57."
        },
        "sources": [
          {
            "what": "model card, read 20 November 2026",
            "url": "https://huggingface.co/thinkingmachines/Inkling-2-Mini",
            "seen": "2026-11-20"
          },
          {
            "what": "Hub API, gating, downloads, likes, safetensors count, file sizes and SHA-256 for LFS files",
            "url": "https://huggingface.co/api/models/thinkingmachines/Inkling-2-Mini?blobs=true",
            "seen": "2026-11-20"
          },
          {
            "what": "launch post, architecture, size, Tinker and monitorability",
            "url": "https://thinkingmachines.ai/news/inkling-2-mini/",
            "seen": "2026-11-20"
          },
          {
            "what": "Thinking Machines' model card page",
            "url": "https://thinkingmachines.ai/model-card/inkling-2-mini/",
            "seen": "2026-11-20"
          },
          {
            "what": "commit history, one commit on 20 November 2026",
            "url": "https://huggingface.co/api/models/thinkingmachines/Inkling-2-Mini/commits/main",
            "seen": "2026-11-20"
          },
          {
            "what": "chat template, tool format, reasoning effort presets and input types",
            "url": "https://huggingface.co/thinkingmachines/Inkling-2-Mini/raw/main/chat_template.jinja",
            "seen": "2026-11-20"
          },
          {
            "what": "config, architecture, loop, experts and context",
            "url": "https://huggingface.co/thinkingmachines/Inkling-2-Mini/raw/main/config.json",
            "seen": "2026-11-20"
          },
          {
            "what": "discussions, counted by script, no text read",
            "url": "https://huggingface.co/api/models/thinkingmachines/Inkling-2-Mini/discussions",
            "seen": "2026-11-20"
          },
          {
            "what": "inference provider mapping, none",
            "url": "https://huggingface.co/api/models/thinkingmachines/Inkling-2-Mini?expand[]=inferenceProviderMapping",
            "seen": "2026-11-20"
          },
          {
            "what": "Model Acceptable Use Policy, updated 15 July 2026",
            "url": "https://thinkingmachines.ai/model-acceptable-use-policy/",
            "seen": "2026-11-20"
          },
          {
            "what": "training data page",
            "url": "https://thinkingmachines.ai/training-data-documentation/",
            "seen": "2026-11-20"
          },
          {
            "what": "Tinker models and pricing",
            "url": "https://tinker-docs.thinkingmachines.ai/tinker/models/models_and_pricing/",
            "seen": "2026-11-20"
          },
          {
            "what": "tml-renderers guide",
            "url": "https://tinker-docs.thinkingmachines.ai/cookbook/inkling/tml-renderers/",
            "seen": "2026-11-20"
          },
          {
            "what": "vLLM's recipe, 0.35.0 at the earliest and the inkling-2-mini image",
            "url": "https://github.com/vllm-project/vllm/blob/main/recipes/models/thinkingmachines/Inkling-2-Mini.yaml",
            "seen": "2026-11-20"
          },
          {
            "what": "SGLang's cookbook, support not in a release",
            "url": "https://docs.sglang.io/cookbook/autoregressive/ThinkingMachines/Inkling-2-Mini",
            "seen": "2026-11-20"
          },
          {
            "what": "Hugging Face's post, Transformers, llama.cpp, MLX and reasoning effort",
            "url": "https://huggingface.co/blog/inkling2mini",
            "seen": "2026-11-20"
          },
          {
            "what": "ggml-org's GGUF builds and file sizes",
            "url": "https://huggingface.co/api/models/ggml-org/Inkling-2-Mini-GGUF?blobs=true",
            "seen": "2026-11-20"
          },
          {
            "what": "vLLM v0.34.0, Inkling and Inkling 2 Small code, no inkling2_loop (clone of the tag)",
            "url": "https://github.com/vllm-project/vllm/tree/v0.34.0",
            "seen": "2026-11-20"
          },
          {
            "what": "SGLang v0.5.23, no inkling2_loop (clone of the tag)",
            "url": "https://github.com/sgl-project/sglang/tree/v0.5.23",
            "seen": "2026-11-20"
          },
          {
            "what": "Transformers v5.23.0, inkling2 without the loop (clone of the tag)",
            "url": "https://github.com/huggingface/transformers/tree/v5.23.0",
            "seen": "2026-11-20"
          },
          {
            "what": "llama.cpp b12591, inkling2 without the loop (clone of the tag)",
            "url": "https://github.com/ggml-org/llama.cpp/tree/b12591",
            "seen": "2026-11-20"
          },
          {
            "what": "llama.cpp master, Inkling2LoopForConditionalGeneration mapped",
            "url": "https://github.com/ggml-org/llama.cpp/blob/master/conversion/__init__.py",
            "seen": "2026-11-20"
          },
          {
            "what": "Ollama v0.40.16, no Inkling 2 code (clone of the tag)",
            "url": "https://github.com/ollama/ollama/tree/v0.40.16",
            "seen": "2026-11-20"
          }
        ],
        "openQuestions": [
          "Capability is pending. No leaderboard was read on 20 November, as for Inkling 2 Small. Thinking Machines' card copies rows from Artificial Analysis and Vals, which count only when read from those sites",
          "Judgement call, memory is read from ggml-org's IQ4_NL GGUF (25), which needs a llama.cpp built after b12591. Read from Thinking Machines' own checkpoint it would be 20",
          "Judgement call, agent guidance is 10 against Inkling 2 Small's 5. This card names what the safety work covered and safeguards for agent use (a sandbox, confirmation before destructive actions)",
          "Judgement call, the acceptable use policy linked from the card is read as added to Apache 2.0, which moves terms to 10 and licence risk to 5, as for Inkling 2 Small",
          "TML-d24-varloop-research, released beside it with the same architecture and size and a variable number of passes, isn't listed. It's a research artifact for monitorability work, and a listing for it is the founder's call",
          "Responsiveness is 0 on a repository an hour old when it was counted. It's due a re-count once questions are answered",
          "Runtimes is 0 on release day. vLLM names 0.35.0 at the earliest, SGLang a pull request, and llama.cpp has it on master, so it's due a re-check when those release"
        ]
      },
      "negative": 0,
      "verdict": "Thinking Machines' small open model, a looped 26.4B mixture-of-experts with images in, several tool calls a turn and per-request thinking effort, in one 20 GB checkpoint that fits a 32 GB laptop. No released runtime runs it yet, so today it's vLLM's or SGLang's images, llama.cpp from master, or Tinker.",
      "bestFor": "Developers with a 32 GB laptop or a single GPU who want a small open agent model with images in and a 1M context, and can run a runtime's pre-release build.",
      "strengths": [
        "About 20 GB from Thinking Machines and 18 GB as ggml-org's 4-bit GGUF, so it fits a 32 GB laptop",
        "Safetensors only, with a checksum on every LFS file in the publisher's own organisation",
        "Thinking effort set per request from none to max, several tool calls a turn and a 1M-token context",
        "The same chat template as Inkling 2 Small, with tml-renderers and the vLLM and SGLang parsers for tool calls",
        "The card names agent failures to guard against and the safeguards to add, a sandbox and confirmation before destructive actions",
        "Thinking Machines' own Tinker samples from it at $0.15 and $0.50 per million tokens"
      ],
      "weaknesses": [
        "No released version of vLLM, SGLang, Transformers, llama.cpp or Ollama runs it on 20 November 2026",
        "Apache 2.0 comes with an acceptable use policy that binds on download and can change at any time",
        "Safety results are Thinking Machines' own, with no harness to rerun them and no prompt-injection guidance",
        "Structured output isn't documented, and the card gives no dated knowledge cut-off or training summary for this model",
        "One checkpoint and no full-precision release from the publisher",
        "No independent benchmark results were read, so Capability is pending and the total is capped at 61"
      ],
      "agentNotes": [
        "No released runtime runs it yet. Serve it from vLLM's vllm/vllm-openai:inkling-2-mini image with `--reasoning-parser inkling --enable-auto-tool-choice --tool-call-parser inkling`, or from SGLang's lmsysorg/sglang:inkling-2-mini image, where the tool-call parser is `inkling2`",
        "On a laptop, ggml-org's GGUF runs in a llama.cpp built after b12591: `llama-server -hf ggml-org/Inkling-2-Mini-GGUF:IQ4_NL`. The looped architecture isn't in a llama.cpp release yet",
        "Pass `reasoning_effort` as none, minimal, low, medium, high (the default), xhigh or max, or a number from 0 to 0.99. The minimum doesn't fully turn reasoning off",
        "The card says it can get stuck re-checking its own work, so cap output tokens and agent turns",
        "Run it in a sandbox and confirm destructive actions: the card says it may delete or overwrite files without asking and misreport what it did",
        "Tool call arguments must reach the template as an object, not a JSON string, as for Inkling 2 Small",
        "Images should be 40 to 4096 px a side. It takes no audio, and the template raises an error for an audio part",
        "Use guided decoding or validate JSON yourself, since structured output isn't documented",
        "Without the hardware, sample from it on Tinker with the Tinker SDK; Tinker's OpenAI-compatible endpoint doesn't serve it yet",
        "Read the Model Acceptable Use Policy before deploying, it binds on download and bars some uses that are legal"
      ],
      "metrics": {
        "kind": "local",
        "measured": false
      },
      "reviewCount": 0,
      "avgRating": 0,
      "history": [
        {
          "basis": "public evidence",
          "confidence": "low",
          "grade": "C",
          "methodology": "1.0",
          "run": "2026-10-01",
          "runLabel": "October 2026 research run",
          "score": 61
        }
      ],
      "editorialScores": {
        "interface": 85,
        "reach": 60,
        "risk": 70,
        "upkeep": 57
      },
      "provenanceScore": 72,
      "checked": {
        "categories": {
          "interface": "2026-11-20",
          "reach": "2026-11-20",
          "risk": "2026-11-20",
          "upkeep": "2026-11-20"
        },
        "last": "2026-11-20",
        "oldest": "2026-11-20",
        "stale": false,
        "staleAfterDays": 90
      },
      "scoreHistory": [
        {
          "date": "2026-11-20",
          "grade": "C",
          "kind": "listed",
          "method": "1.0",
          "score": 61
        }
      ]
    },
    "connect": {
      "install": "# per vLLM's recipe. Not in a vLLM release yet (0.35.0 at the earliest),\n# so run it inside the vllm/vllm-openai:inkling-2-mini image:\nvllm serve thinkingmachines/Inkling-2-Mini --reasoning-parser inkling --enable-auto-tool-choice --tool-call-parser inkling",
      "http": "curl http://localhost:8000/v1/chat/completions \\\n  -H \"Content-Type: application/json\" \\\n  -d '{\n    \"model\": \"thinkingmachines/Inkling-2-Mini\",\n    \"messages\": [{\"role\": \"user\", \"content\": \"Hello!\"}],\n    \"reasoning_effort\": \"low\"\n  }'"
    },
    "letme": {
      "capability": "https://letme.dev/model.open-weights",
      "tool": "https://letme.dev/inkling-2-mini"
    },
    "sameCompany": [
      "inkling",
      "inkling-2-small",
      "inkling-small"
    ],
    "sameModel": [
      "tinker"
    ],
    "weights": {
      "trainedBy": "Thinking Machines Lab",
      "params": "26.4B",
      "active": "3.4B",
      "context": 1048576,
      "goodFor": "Small looped mixture-of-experts model for coding agents, tool use and search with images in, a 1M-token context and thinking effort set per request, in about 20 GB.",
      "runtimes": [],
      "builds": [
        {
          "name": "ggml-org Inkling-2-Mini-GGUF (IQ4_NL)",
          "url": "https://huggingface.co/ggml-org/Inkling-2-Mini-GGUF",
          "who": "llama.cpp",
          "precision": "iq4_nl",
          "sizeGB": 17.98,
          "memoryGB": 24,
          "format": "gguf",
          "runsOn": [
            "llama.cpp"
          ],
          "note": "From ggml-org, the llama.cpp project's organisation. 17.63 GB of weights, the NVFP4 experts at IQ4_NL and everything else at Q8_0, plus the 0.35 GB Q8_0 vision projector, from the Hub's model information on 20 November 2026. Needs a llama.cpp built after b12591, since the looped architecture is on the master branch and not in a release yet"
        },
        {
          "name": "Inkling-2-Mini",
          "url": "https://huggingface.co/thinkingmachines/Inkling-2-Mini",
          "who": "publisher",
          "precision": "nvfp4",
          "sizeGB": 20.19,
          "memoryGB": 32,
          "format": "safetensors",
          "note": "The only checkpoint, in the precisions the model was trained in: routed experts in NVFP4 for layers 0 to 19 and MXFP8 for layers 20 to 23, with the DSpark draft head in mtp.safetensors. Thinking Machines recommends 32 GB or more of memory. No released runtime loads it yet"
        },
        {
          "name": "ggml-org Inkling-2-Mini-GGUF (Q8_0)",
          "url": "https://huggingface.co/ggml-org/Inkling-2-Mini-GGUF",
          "who": "llama.cpp",
          "precision": "q8_0",
          "sizeGB": 28.17,
          "memoryGB": 48,
          "format": "gguf",
          "runsOn": [
            "llama.cpp"
          ],
          "note": "From ggml-org. 27.82 GB of Q8_0 weights plus the 0.35 GB Q8_0 vision projector. Needs a llama.cpp built after b12591"
        }
      ],
      "run": "vllm serve thinkingmachines/Inkling-2-Mini --reasoning-parser inkling --enable-auto-tool-choice --tool-call-parser inkling",
      "runBuild": "Inkling-2-Mini",
      "checked": "2026-11-20",
      "minMemoryGB": 24
    },
    "family": {
      "lab": "Thinking Machines Lab",
      "family": "Inkling 2",
      "url": "https://www.anchorterminal.com/open-weights/#thinking-machines",
      "others": [
        {
          "name": "TML-d24-varloop-research",
          "url": "https://huggingface.co/thinkingmachines/TML-d24-varloop-research",
          "params": "26.4B total, 3.4B active per pass (Hub safetensors count matches Inkling-2-Mini)"
        },
        {
          "name": "Inkling-2-Small",
          "url": "https://huggingface.co/thinkingmachines/Inkling-2-Small",
          "params": "594B total, 17B active (card; the Hub's safetensors count isn't a parameter count for NVFP4 files)",
          "listing": "https://www.anchorterminal.com/tools/inkling-2-small"
        }
      ],
      "unrestrictedCopies": 2
    },
    "notable": [
      "Released on 20 November 2026 with the weights public on Hugging Face and Tinker sampling the same day. The Hub counted 136 downloads and 239 likes that evening (https://huggingface.co/api/models/thinkingmachines/Inkling-2-Mini?blobs=true)",
      "A looped transformer, Thinking Machines' first. Its 24 layers hold the weights, and layers 6 to 17 run three times per token, so a token makes 48 layer visits, the same as Inkling 2 Small. The card puts the effective active parameters at 5.4B with the three passes (https://huggingface.co/thinkingmachines/Inkling-2-Mini)",
      "Thinking Machines' table gives AA Index v5.0 20.1, Vals Index v2.2 46.5 per cent, DeepSWE v1.1 48.2 per cent and SWE-bench Pro v2 Hard 77.9 per cent. These are Thinking Machines' numbers, not our measurement, and some rows are copied from Artificial Analysis and Vals by the publisher (https://huggingface.co/thinkingmachines/Inkling-2-Mini)",
      "Released beside it, TML-d24-varloop-research is the same architecture trained to run one to several passes through the loop, for research on monitorability. It isn't listed separately (https://thinkingmachines.ai/news/inkling-2-mini/)",
      "No released version of vLLM, SGLang, Transformers, llama.cpp or Ollama runs it on 20 November 2026. vLLM's recipe serves it from the vllm/vllm-openai:inkling-2-mini image and names 0.35.0 at the earliest, SGLang's cookbook from its inkling-2-mini images or a pull request, and llama.cpp added the looped architecture to its master branch after its latest build, b12591 (https://github.com/vllm-project/vllm/blob/main/recipes/models/thinkingmachines/Inkling-2-Mini.yaml)"
    ],
    "area": "models",
    "details": [
      {
        "label": "Surface graded",
        "value": "The open weights at huggingface.co/thinkingmachines/Inkling-2-Mini, read from the model card, Thinking Machines' own model card page and launch post, the files and their metadata, the chat template, the acceptable use policy, the training data page and the runtimes' own code and docs on 20 November 2026. Nothing was downloaded or run"
      },
      {
        "label": "Model",
        "value": "26.4B total and 3.4B active parameters per the card, 5.4B effective active with the loop's three passes. A 24-layer decoder with a sparse mixture of experts, 8 of 128 experts per token, local and global attention, layers 6 to 17 run three times with cross-pass messages, a vision encoder and a 3-layer DSpark draft head that vLLM and SGLang use for speculative decoding. Text and images in, text out. No audio"
      },
      {
        "label": "Files",
        "value": "One safetensors file of 19.5 GB plus mtp.safetensors (0.6 GB), 20.2 GB in all, with a chat_template.jinja, a tiktoken tokeniser file and a quantisation map. The routed experts of layers 0 to 19 are in NVFP4 and of layers 20 to 23 in MXFP8, per SGLang's cookbook. No pickle files and no Python files. Hugging Face lists a SHA-256 for every LFS file"
      },
      {
        "label": "Licence",
        "value": "Apache 2.0 in the card's metadata, linked to apache.org, with no LICENSE file in the repository. The card links Thinking Machines' Model Acceptable Use Policy (updated 15 July 2026), which binds anyone who downloads or uses the weights, forbids uses whether or not they're illegal, and can be updated at any time"
      },
      {
        "label": "Runtimes",
        "value": "No released version runs it on 20 November 2026. vLLM 0.34.0, SGLang 0.5.23, Transformers 5.23.0, llama.cpp b12591 and Ollama 0.40.16 have no code for the looped architecture (inkling2_loop), read from a clone of each. vLLM's recipe uses the vllm/vllm-openai:inkling-2-mini image, SGLang's cookbook its inkling-2-mini images or pull request 51926, and llama.cpp's master branch converts and runs it. Hugging Face's post says Transformers and MLX support ship with the release, but neither has it in a released version yet"
      },
      {
        "label": "Hardware",
        "value": "Thinking Machines names one NVIDIA DGX Spark, Apple silicon with 32 GB or more, and other GPUs with 32 GB or more of memory. ggml-org's 4-bit GGUF is 17.6 GB plus a 0.35 GB vision projector, about 22 GB with a fifth added"
      },
      {
        "label": "Hosted",
        "value": "Thinking Machines' Tinker samples from it through the Tinker SDK at $0.15 per million prefill tokens and $0.50 per million sampled, and its playground runs it. Tinker's serverless OpenAI-compatible inference serves Inkling, Inkling-Small and Inkling 2 Small only, read 20 November 2026. No inference provider is mapped on Hugging Face yet"
      }
    ],
    "provenance": {
      "legalEntity": "Thinking Machines Lab, Inc. (named as model provider on the card)",
      "domain": "thinkingmachines.ai",
      "domainRegistered": "",
      "domainNote": "Not looked up: RDAP isn't requested under the 1 November 2026 rules.",
      "endpointOnVendorDomain": null,
      "terms": "",
      "privacy": "",
      "statusPage": "",
      "changelog": "https://huggingface.co/thinkingmachines/Inkling-2-Mini/commits/main",
      "securityTxt": "valid",
      "checked": "2026-11-20",
      "notes": [
        "Thinking Machines' launch post and model card page link the repository in the thinkingmachines organisation on Hugging Face, which isn't verified by Hugging Face.",
        "Nothing is hosted for this listing, so the Apache 2.0 licence and the Model Acceptable Use Policy stand in for terms of service and no privacy policy is scored.",
        "The changelog is the repository's commit history, one commit on 20 November 2026, the upload.",
        "thinkingmachines.ai/.well-known/security.txt gives a security contact, a PGP key and an expiry in 2029, read on 20 November 2026, and the site's footer links a disclosure policy.",
        "thinkingmachines.ai's robots.txt has no rules. Thinking Machines' terms bar scraping the Services, and under the founder's decision of 6 November 2026 a few pages were read one at a time at a person's pace on 20 November: the launch post, the model card page, the acceptable use policy, the training data page and Tinker's models and pricing page."
      ],
      "score": 72,
      "checks": [
        {
          "check": "Legal entity named",
          "value": "Thinking Machines Lab, Inc. (named as model provider on the card)",
          "points": 20,
          "max": 20,
          "state": "ok"
        },
        {
          "check": "Domain age",
          "value": "thinkingmachines.ai, no registry record we could read",
          "points": 0,
          "max": 15,
          "state": "no"
        },
        {
          "check": "Endpoint on the vendor's domain",
          "value": "no hosted endpoint",
          "points": 0,
          "max": 0,
          "state": "na"
        },
        {
          "check": "Terms of service",
          "value": "nothing hosted, so the Apache-2.0 licence stands in",
          "points": 10,
          "max": 10,
          "state": "ok"
        },
        {
          "check": "Privacy policy",
          "value": "nothing hosted, not scored",
          "points": 0,
          "max": 0,
          "state": "na"
        },
        {
          "check": "Status page",
          "value": "not found",
          "points": 0,
          "max": 10,
          "state": "no"
        },
        {
          "check": "Changelog",
          "value": "published",
          "points": 10,
          "max": 10,
          "state": "ok"
        },
        {
          "check": "security.txt",
          "value": "valid",
          "points": 10,
          "max": 10,
          "state": "ok"
        },
        {
          "check": "People named",
          "value": "2 named with their role, 2 with a public profile",
          "points": 10,
          "max": 10,
          "state": "ok"
        },
        {
          "check": "Support route",
          "value": "a support address, help pages, a community",
          "points": 5,
          "max": 5,
          "state": "ok"
        }
      ]
    },
    "humans": {
      "slug": "inkling-2-mini",
      "checked": "2026-10-28",
      "company": "Thinking Machines Lab",
      "people": [
        {
          "name": "Mira Murati",
          "role": "Co-founder and CEO",
          "founder": true,
          "x": "https://x.com/miramurati",
          "bio": "Mira Murati co-founded Thinking Machines Lab and has led it as chief executive since its launch in February 2025. She was OpenAI's chief technology officer from 2022 to 2024, leading work on ChatGPT, DALL-E, Codex and Sora, and was briefly its interim CEO in November 2023. Earlier she worked at Tesla and Leap Motion.",
          "bioSources": [
            "https://en.wikipedia.org/wiki/Mira_Murati"
          ],
          "page": "/people/mira-murati/"
        },
        {
          "name": "John Schulman",
          "id": "john-schulman",
          "role": "Co-founder and Chief Scientist",
          "founder": true,
          "x": "https://x.com/johnschulman2",
          "bio": "John Schulman is chief scientist at Thinking Machines Lab, which he joined in February 2025. He co-founded OpenAI in 2015 and led the reinforcement learning team that created ChatGPT, drawing on his work on Trust Region Policy Optimization and proximal policy optimization. He spent a few months at Anthropic in between.",
          "bioSources": [
            "https://en.wikipedia.org/wiki/John_Schulman"
          ],
          "page": "/people/john-schulman/"
        }
      ],
      "peopleNote": "Thinking Machines' own site has no team page and signs its posts as the company. Its NVIDIA partnership post of 10 March 2026 quotes Mira Murati as cofounder and CEO, and John Schulman's own homepage, read on 28 October 2026, says he is cofounder and chief scientist. Wikipedia names Lilian Weng among the founding team and says Barret Zoph and Luke Metz went back to OpenAI in January 2026 and Andrew Tulloch left for Meta (https://en.wikipedia.org/wiki/Thinking_Machines_Lab); none of them is listed, since no first-hand page we read names their current role. Roster may be incomplete.",
      "companySocial": {
        "linkedin": "https://www.linkedin.com/company/thinkingmachinesai/",
        "x": "https://x.com/thinkymachines"
      },
      "support": {
        "emails": [
          "tinker@thinkingmachines.ai"
        ],
        "urls": [
          "https://tinker-docs.thinkingmachines.ai/support/",
          "https://github.com/thinking-machines-lab/tinker-feedback/issues"
        ],
        "community": [
          {
            "label": "Tinker Discord",
            "url": "https://discord.gg/KqqEZNX88c",
            "kind": "discord"
          },
          {
            "label": "Tinker feedback",
            "url": "https://github.com/thinking-machines-lab/tinker-feedback/issues",
            "kind": "github"
          },
          {
            "label": "Model card",
            "url": "https://thinkingmachines.ai/model-card/inkling-2-small/",
            "kind": "link"
          }
        ]
      },
      "sources": [
        "https://thinkingmachines.ai/news/nvidia-partnership/",
        "https://x.com/miramurati",
        "https://thinkingmachines.ai/",
        "https://thinkingmachines.ai",
        "https://www.linkedin.com/company/thinkingmachinesai/",
        "https://tinker-docs.thinkingmachines.ai/support/",
        "https://thinkingmachines.ai/news/inkling-2-small/"
      ],
      "ownership": {
        "status": "private",
        "totalRaised": {
          "amount": "$2B",
          "asOf": "2025-07",
          "source": "https://techcrunch.com/2025/07/15/mira-muratis-thinking-machines-lab-is-worth-12b-in-seed-round/"
        },
        "valuation": {
          "amount": "$12B",
          "basis": "stated",
          "round": "Seed",
          "asOf": "2025-07",
          "source": "https://techcrunch.com/2025/07/15/mira-muratis-thinking-machines-lab-is-worth-12b-in-seed-round/"
        },
        "checked": "2026-10-28",
        "sources": [
          "https://techcrunch.com/2025/07/15/mira-muratis-thinking-machines-lab-is-worth-12b-in-seed-round/",
          "https://thinkingmachines.ai/news/nvidia-partnership/",
          "https://thinkingmachines.ai/news/"
        ]
      },
      "investors": [
        {
          "name": "Andreessen Horowitz",
          "kind": "vc",
          "rounds": [
            "Seed, July 2025"
          ],
          "lead": true,
          "url": "https://a16z.com",
          "source": "https://techcrunch.com/2025/07/15/mira-muratis-thinking-machines-lab-is-worth-12b-in-seed-round/",
          "quote": "Thinking Machines Lab, the AI startup founded by OpenAI's former chief technology officer Mira Murati, officially closed a $2 billion seed round led by Andreessen Horowitz on Monday, a company spokesperson told TechCrunch.",
          "page": "/investors/andreessen-horowitz/"
        },
        {
          "name": "NVIDIA",
          "kind": "corporate",
          "rounds": [
            "Seed, July 2025",
            "Strategic investment, March 2026"
          ],
          "url": "https://www.nvidia.com",
          "source": "https://thinkingmachines.ai/news/nvidia-partnership/",
          "quote": "NVIDIA has also made a significant investment in Thinking Machines Lab to support the company's long-term growth.",
          "page": "/investors/nvidia/"
        },
        {
          "name": "Accel",
          "kind": "vc",
          "rounds": [
            "Seed, July 2025"
          ],
          "url": "https://www.accel.com",
          "source": "https://techcrunch.com/2025/07/15/mira-muratis-thinking-machines-lab-is-worth-12b-in-seed-round/",
          "quote": "The deal, which includes participation from Nvidia, Accel, ServiceNow, CISCO, AMD, and Jane Street, values the startup at $12 billion, the spokesperson said.",
          "page": "/investors/accel/"
        },
        {
          "name": "ServiceNow",
          "kind": "corporate",
          "rounds": [
            "Seed, July 2025"
          ],
          "url": "https://www.servicenow.com",
          "source": "https://techcrunch.com/2025/07/15/mira-muratis-thinking-machines-lab-is-worth-12b-in-seed-round/",
          "quote": "The deal, which includes participation from Nvidia, Accel, ServiceNow, CISCO, AMD, and Jane Street, values the startup at $12 billion, the spokesperson said.",
          "page": "/investors/servicenow/"
        },
        {
          "name": "Cisco",
          "firm": "Cisco Investments",
          "kind": "corporate",
          "rounds": [
            "Seed, July 2025"
          ],
          "url": "https://www.cisco.com",
          "source": "https://techcrunch.com/2025/07/15/mira-muratis-thinking-machines-lab-is-worth-12b-in-seed-round/",
          "quote": "The deal, which includes participation from Nvidia, Accel, ServiceNow, CISCO, AMD, and Jane Street, values the startup at $12 billion, the spokesperson said.",
          "page": "/investors/cisco-investments/"
        },
        {
          "name": "AMD",
          "kind": "corporate",
          "rounds": [
            "Seed, July 2025"
          ],
          "url": "https://www.amd.com",
          "source": "https://techcrunch.com/2025/07/15/mira-muratis-thinking-machines-lab-is-worth-12b-in-seed-round/",
          "quote": "The deal, which includes participation from Nvidia, Accel, ServiceNow, CISCO, AMD, and Jane Street, values the startup at $12 billion, the spokesperson said.",
          "page": "/investors/amd/"
        },
        {
          "name": "Jane Street",
          "kind": "other",
          "rounds": [
            "Seed, July 2025"
          ],
          "url": "https://www.janestreet.com",
          "source": "https://techcrunch.com/2025/07/15/mira-muratis-thinking-machines-lab-is-worth-12b-in-seed-round/",
          "quote": "The deal, which includes participation from Nvidia, Accel, ServiceNow, CISCO, AMD, and Jane Street, values the startup at $12 billion, the spokesperson said.",
          "page": "/investors/jane-street/"
        }
      ]
    },
    "pageJsonUrl": "https://www.anchorterminal.com/tools/inkling-2-mini.json"
  }
}
