{"version":1,"type":"story","url":"https://digestai.news/story/inkling-2-small-released-with-594b-parameters-and-17b-active","json":"https://digestai.news/story/inkling-2-small-released-with-594b-parameters-and-17b-active.json","markdown":"https://digestai.news/story/inkling-2-small-released-with-594b-parameters-and-17b-active.md","slug":"inkling-2-small-released-with-594b-parameters-and-17b-active","headline":"Thinking Machines Lab releases Inkling 2 Small with 594B parameters and 17B active parameters","summary":"Thinking Machines Lab introduced Inkling 2 Small, a Mixture-of-Experts model with 594 billion total parameters and 17 billion active parameters, supporting image inputs and variable effort. The model scores 31 on the Artificial Analysis Intelligence Index, surpassing previous models and matching Gemini 3.8 Flash (high). It is initially available via Tinker with closed weights, with plans to release full weights under Apache 2.0 after external safety testing and red-teaming.\n\nWhile all sources agree on the model's specifications, release timeline, and safety conditions, only Artificial Analysis provides comparative benchmark scores against Gemini 3.8 Flash (high), Claude Sonnet 5.5 (medium), and others, including Elo scores on AA-Briefcase and cost per task. Thinking Machines Lab and unite.ai focus on training details, alignment controls, and the model's agentic capabilities, such as autonomously training Qwen3.5-4B on Tinker.","keyPoints":["Inkling 2 Small has 594B total parameters and 17B active parameters.","It scores 31 on the Artificial Analysis Intelligence Index, matching Gemini 3.8 Flash (high).","Full weights will be released under Apache 2.0 after external safety testing and red-teaming."],"whyItMatters":"Inkling 2 Small advances open-weight model performance in reasoning, vision, and agentic tasks, setting a new benchmark for transparency and capability in AI development outside major Chinese labs.","category":{"slug":"models","name":"Generative AI & Models","url":"https://digestai.news/category/models"},"entities":{"companies":["Thinking Machines Lab","Tinker","Scale AI","Google DeepMind","Xiaomi","Meta"],"models":["Inkling 2 Small","Inkling Small","Gemini 3.8 Flash (high)","Claude Sonnet 5.5 (medium)","Qwen3.5-4B","Muse Spark 1.3"],"people":[]},"firstPublishedAt":"2026-11-06T00:28:53Z","updatedAt":"2026-11-06T04:25:07Z","sourceCount":5,"hasPrimarySource":false,"sources":[{"outlet":"thinkingmachines.ai","title":"Inkling 2 Small released","url":"https://thinkingmachines.ai/news/inkling-2-small","publishedAt":"2026-11-06T00:28:53Z","type":"press","primary":false,"lead":true},{"outlet":"MarkTechPost","title":"Thinking Machines Releases Inkling 2 Small, a 594B MoE With 17B Active","url":"https://marktechpost.com/2026/11/05/thinking-machines-inkling-2-small-release","publishedAt":"2026-11-06T04:25:07Z","type":"press","primary":false,"lead":false},{"outlet":"artificialanalysis.ai","title":"Artificial Analysis says Inkling 2 Small matches Gemini 3.8 Flash (high) in its index and would be the top open model outside China once its weights are out","url":"https://artificialanalysis.ai/articles/thinking-machines-releases-inkling-2-small-highest-scoring-open-weights-model-outside-of-chinia-upon-weights-release-it","publishedAt":"2026-11-06T00:55:00Z","type":"press","primary":false,"lead":false},{"outlet":"Unite.AI","title":"Thinking Machines Lab Launches Inkling 2 Small with Open Weights Pending","url":"https://unite.ai/thinking-machines-lab-launches-inkling-2-small-with-open-weights-pending","publishedAt":"2026-11-06T00:44:05Z","type":"press","primary":false,"lead":false},{"outlet":"note.com","title":"[AI News] Huge open-weight models, Claude's personal finance management, GPT-6 Cyber ― 3 stories to catch up on this morning (2026/11/06)","url":"https://note.com/carbon4322/n/n742b3fd0b331?hl=en","publishedAt":"2026-11-05T19:09:00Z","type":"press","primary":false,"lead":false}],"sourceNotes":{"agree":"Thinking Machines Lab released Inkling 2 Small with 594B total and 17B active parameters, image support, and plans to release Apache 2.0 weights after safety testing.","differ":[]},"discussions":[{"site":"Reddit","url":"https://www.reddit.com/r/LocalLLaMA/comments/1xoz01k/inkling_2_small_released/","points":null}],"thread":{"title":"AI Race Accelerates with SWE-2 Flash and Inkling 2","url":"https://digestai.news/thread/cognition-says-swe-2-flash-runs-three-times-faster-and-costs-one-fifth","storyCount":2},"cite":{"text":"Digest AI, \"Thinking Machines Lab releases Inkling 2 Small with 594B parameters and 17B active parameters\", 6 November 2026, https://digestai.news/story/inkling-2-small-released-with-594b-parameters-and-17b-active","publisher":"Digest AI","title":"Thinking Machines Lab releases Inkling 2 Small with 594B parameters and 17B active parameters","datePublished":"2026-11-06T00:28:53Z","url":"https://digestai.news/story/inkling-2-small-released-with-594b-parameters-and-17b-active"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}