Thinking Machines launches Inkling, its first open-weight AI model

Thinking Machines Lab, the AI startup founded by former OpenAI chief technology officer Mira Murati, released its first proprietary AI model on Wednesday morning, called Inkling. Unlike the flagship models from OpenAI, Anthropic, and Google, Inkling is open-weight, meaning outside developers and companies can download and modify it directly. It is a mixture-of-experts system — an architecture that activates only a fraction of its total parameters for any given task — with 975 billion total parameters, though it draws on only about 41 billion for any single task. Thinking Machines trained the model on 45 trillion tokens of text, image, audio, and video, and says it reasons natively across all three, according to the company’s own release materials.

Rather than compete as a general-purpose chatbot the way ChatGPT, Claude, and Gemini do, Inkling is positioned as a starting point for organizations to customize through Tinker, the company’s model-tuning platform. Thinking Machines is candid that the model isn’t best-in-class: its briefing materials state it is “not the strongest model available today, closed or open.” Instead, the company points to efficiency — on one benchmark, it says Inkling uses a third as many tokens as Nvidia’s Nemotron 3 Ultra to match its coding performance.

That framing echoes a broader argument gaining traction across the industry. Microsoft CEO Satya Nadella, whose company has invested in both OpenAI and Anthropic, argued that enterprises using proprietary AI models effectively pay twice: once in subscription costs, and again by handing over business knowledge that can be absorbed into future model versions. Hugging Face CEO Clem Delangue made a similar prediction, saying frontier models will increasingly be reserved for experimentation and high-value tasks while most production AI work shifts to private or open-source alternatives. The clearest evidence so far comes from a joint project with Bridgewater Associates: a customized open model scored 84.7% on financial reasoning tests, beating top proprietary systems, while costing roughly a fourteenth as much to run, according to results the two companies published jointly in late June.

Thinking Machines has emphasized how fast it moved, saying it brought its technology to market and showed revenue in about nine months, compared with:

• OpenAI: roughly five years
• Anthropic: roughly three years

A reported $50 billion fundraising round was said to be coming together last November, but multiple outlets reported it stalled by January, when two co-founders left for OpenAI. Nvidia made a significant investment in Thinking Machines Lab when the two companies announced their March partnership, according to Nvidia. Headcount has since stabilized: the startup now employs roughly 200 people, up from levels reported after a wave of departures earlier this year.

Why it matters

Thinking Machines is betting that enterprises will get more value from open-weight models they can customize themselves than from the closed, centralized systems that OpenAI, Anthropic, and Google currently sell.