--- title: Inkling: Our Open-Weights Model | daily.dev description: Thinking Machines Lab has released Inkling, their first open-weights foundation model trained from scratch. It is a Mixture-of-Experts transformer with 975B... canonical: https://daily.dev/posts/inkling-our-open-weights-model-piggefds2 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Inkling: Our Open-Weights Model | daily.dev og:description: Thinking Machines Lab has released Inkling, their first open-weights foundation model trained from scratch. It is a Mixture-of-Experts transformer with 975B... og:url: https://daily.dev/posts/inkling-our-open-weights-model-piggefds2 og:image: https://api.daily.dev/og/posts/PiGGefDS2.png og:image:alt: Inkling: Our Open-Weights Model og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Inkling: Our Open-Weights Model **[Hacker News](https://daily.dev/sources/hn)** · 17 min read · 1 upvotes · 0 comments ## Summary Thinking Machines Lab has released Inkling, their first open-weights foundation model trained from scratch. It is a Mixture-of-Experts transformer with 975B total parameters (41B active), a 1M token context window, and was pretrained on 45 trillion tokens spanning text, images, audio, and video. Key features include controllable thinking effort (allowing cost/performance tradeoffs), native multimodal reasoning across text, images, and audio, and strong epistemic properties including calibration and censorship resistance. A smaller companion model, Inkling-Small (276B total, 12B active), is also previewed and matches the larger model on many benchmarks. Both models are available for fine-tuning on the Tinker platform, with full weights on Hugging Face and inference support across vLLM, llama.cpp, SGLang, and major cloud providers. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What are the specs of Thinking Machines Lab's Inkling model? Inkling is a Mixture-of-Experts transformer with 975B total parameters and 41B active per token, supporting up to a 1M-token context window. It was pretrained on 45 trillion tokens spanning text, images, audio, and video, and supports controllable thinking effort to trade off cost against performance. Its full weights are published on Hugging Face, including an NVFP4 checkpoint for NVIDIA Blackwell inference. _Developers evaluating open-weights foundation models can track releases like Inkling on daily.dev._ ### How does Inkling-Small compare to the full Inkling model? Inkling-Small is a 276B-parameter Mixture-of-Experts model with 12B active parameters, compared to Inkling's 975B total and 41B active. Despite being far smaller, it matches or exceeds Inkling on many benchmarks such as HLE with tools, MCP-Atlas, and Charxiv RQ with python, thanks to improved pre-training data and recipe, making it suited for cost- and latency-sensitive workloads like coding or synthetic data generation. _Teams weighing model size against latency can follow Inkling-Small's full release on daily.dev._ ### What architecture choices did Thinking Machines make instead of RoPE for Inkling's attention mechanism? Inkling uses a learned, input-dependent relative positional embedding instead of Rotary Positional Embedding (RoPE), finding it performs better and extrapolates better to longer sequences. It also interleaves sliding-window and global attention layers at a 5:1 ratio with 8 KV heads, and applies short convolutions after key/value projections and on residual branch outputs. _Engineers designing long-context transformer architectures can dig into details like these on daily.dev._ ## Similar posts on daily.dev - [Inkling model from Thinking Machines Lab now on Databricks](https://daily.dev/posts/inkling-model-from-thinking-machines-lab-now-on-databricks-1wahujdlp) · databricks · 1 upvotes · 0 comments --- Tags: [#open-source](https://daily.dev/tags/open-source), [#llm](https://daily.dev/tags/llm), [#deep-learning](https://daily.dev/tags/deep-learning), [#multimodal](https://daily.dev/tags/multimodal), [#mixture-of-experts](https://daily.dev/tags/mixture-of-experts) [View this post on daily.dev](https://daily.dev/posts/inkling-our-open-weights-model-piggefds2) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Inkling: Our Open-Weights Model","url":"https://daily.dev/posts/inkling-our-open-weights-model-piggefds2","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/inkling-our-open-weights-model-piggefds2"},"datePublished":"2026-07-15T19:32:45.468Z","dateModified":"2026-09-14T08:56:58.739Z","description":"Thinking Machines Lab has released Inkling, their first open-weights foundation model trained from scratch. It is a Mixture-of-Experts transformer with 975B...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/75afb375e167b2b6dda3debda99be29d?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/75afb375e167b2b6dda3debda99be29d?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Hacker News","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Hacker News","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/hn","url":"https://daily.dev/sources/hn"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/inkling-our-open-weights-model-piggefds2","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"open-source,llm,deep-learning,multimodal,mixture-of-experts","timeRequired":"PT17M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Hacker News","item":"https://daily.dev/sources/hn"},{"@type":"ListItem","position":3,"name":"Inkling: Our Open-Weights Model"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/inkling-our-open-weights-model-piggefds2#faq","mainEntity":[{"@type":"Question","name":"What are the specs of Thinking Machines Lab's Inkling model?","acceptedAnswer":{"@type":"Answer","text":"Inkling is a Mixture-of-Experts transformer with 975B total parameters and 41B active per token, supporting up to a 1M-token context window. It was pretrained on 45 trillion tokens spanning text, images, audio, and video, and supports controllable thinking effort to trade off cost against performance. Its full weights are published on Hugging Face, including an NVFP4 checkpoint for NVIDIA Blackwell inference. Developers evaluating open-weights foundation models can track releases like Inkling on daily.dev."}},{"@type":"Question","name":"How does Inkling-Small compare to the full Inkling model?","acceptedAnswer":{"@type":"Answer","text":"Inkling-Small is a 276B-parameter Mixture-of-Experts model with 12B active parameters, compared to Inkling's 975B total and 41B active. Despite being far smaller, it matches or exceeds Inkling on many benchmarks such as HLE with tools, MCP-Atlas, and Charxiv RQ with python, thanks to improved pre-training data and recipe, making it suited for cost- and latency-sensitive workloads like coding or synthetic data generation. Teams weighing model size against latency can follow Inkling-Small's full release on daily.dev."}},{"@type":"Question","name":"What architecture choices did Thinking Machines make instead of RoPE for Inkling's attention mechanism?","acceptedAnswer":{"@type":"Answer","text":"Inkling uses a learned, input-dependent relative positional embedding instead of Rotary Positional Embedding (RoPE), finding it performs better and extrapolates better to longer sequences. It also interleaves sliding-window and global attention layers at a 5:1 ratio with 8 KV heads, and applies short convolutions after key/value projections and on residual branch outputs. Engineers designing long-context transformer architectures can dig into details like these on daily.dev."}}]} ```