Mira Murati's Thinking Machines Releases Inkling — a 975B Open-Weight, Multimodal Model Built to Be Customized

Thinking Machines Lab, the startup founded by former OpenAI CTO Mira Murati, has released its first broadly available model — and it chose to make it open. Inkling is an open-weights, multimodal foundation model whose full weights can be downloaded from Hugging Face under a permissive Apache 2.0 license. For a lab that raised billions and has said little publicly, leading with an open model rather than a closed API is a statement of intent.

The specs are serious. Inkling is a mixture-of-experts transformer with 975 billion total parameters and 41 billion active on any given forward pass, pretrained on 45 trillion tokens spanning text, images, audio, and video, and it supports a context window of up to one million tokens. It is, in other words, a genuine frontier-scale system — not a small open model released as a marketing gesture — and it lands as one of the largest open-weight multimodal models available to download.

The benchmarks back up the ambition. Inkling scored 97.1% on the 2026 AIME math competition and 87.2% on GPQA Diamond, a test of graduate-level scientific reasoning, alongside 77.6% on SWE-bench Verified and 54.3% on SWE-bench Pro Public for real-world coding. Those numbers put it firmly among the strongest open models and within reach of leading closed systems on reasoning and math, even if Thinking Machines is careful not to claim outright leaderboard dominance.

That restraint is the strategy. Thinking Machines is positioning Inkling less as "the most capable model in the world" and more as a customizable foundation for companies, researchers, and developers who want real control over how their AI behaves. To that end, the model is designed to be fine-tuned through the lab's own Tinker training platform, turning the open release into an on-ramp for a paid customization business — open weights as the top of the funnel, tooling and support as the product.

The timing sharpens the industry's dividing line. In the same stretch that Microsoft is pushing enterprises toward its own closed MAI models and Chinese labs like DeepSeek and Moonshot keep shipping strong open weights, one of the most watched American startups has planted its flag firmly on the open side of the fence. If Inkling holds up in the wild, Murati's lab won't just be another frontier competitor — it will be a bet that the future of serious AI is something you can download, own, and reshape.