Mira Murati's Thinking Machines Lab ships Inkling, an open-weights MoE model

Open-weights AI just gained a heavyweight entrant. On July 15, 2026, Thinking Machines Lab, the startup founded by former OpenAI CTO Mira Murati, released its first model, Inkling.

What Inkling is

Inkling is a Mixture-of-Experts model with 975 billion total parameters, of which 41 billion are active at any time. It supports a context window of up to 1 million tokens and was pretrained on 45 trillion tokens spanning text, images, audio and video.

The headline feature is not a benchmark score. It is the license posture: the weights are public. Developers and companies can download Inkling, modify it, and fine-tune it for their own workflows. A lighter version, Inkling-Small, with 12 billion active parameters, is coming as a preview.

A deliberately honest pitch

The unusual part of the launch is what the lab did not claim. Thinking Machines openly admits that Inkling is not the strongest model on the market. The bet is different: customizable AI that you can shape for your business will beat one-size-fits-all chatbots.

That framing puts Inkling in a specific lane. Rather than competing head-on with closed frontier models on raw capability, the lab is wagering that control, adaptability and ownership matter more to companies building real products than a few points on a leaderboard.

Why it matters

Murati’s lab has been one of the most closely watched startups in AI since she left OpenAI, and its first release stakes out a clear position in the open-versus-closed debate. With a near-trillion-parameter open-weights model now downloadable, the question for builders becomes concrete: open and tunable, or closed and strongest?

For teams that have wanted a large multimodal foundation they can fully own and adapt, Inkling is now one of the biggest options on the table. Which side of that trade you build on is the choice this release sharpens.