Thinking Machines Lab, the AI startup founded by former OpenAI CTO Mira Murati, has released its first proprietary model, Inkling, choosing to make it open-weight so outside developers and companies can download and modify it directly. This sets it apart from the closed, general-purpose approach taken by rivals such as OpenAI, Anthropic and Google, and tests the company's core wager that AI organisations can adapt for themselves will outperform one-size-fits-all systems from major labs.

Inkling is a mixture-of-experts model with 975 billion total parameters, drawing on roughly 41 billion for any given task, and was trained on 45 trillion tokens spanning text, image, audio and video, though its outputs are currently limited to text, code and structured data. The company says it is not the strongest model available, open or closed, but is designed to give calibrated, uncertainty-flagging answers and let users adjust "thinking effort" for speed, reportedly using a third as many tokens as Nvidia's Nemotron 3 Ultra for equivalent coding performance. Thinking Machines is positioning Inkling as a starting point for fine-tuning via its Tinker platform rather than a finished product, echoing recent comments from Microsoft's Satya Nadella and Hugging Face's Clem Delangue about the drawbacks of relying on proprietary, closed AI models.