On 15 July the lab led by Mira Murati - OpenAI's former CTO - showed Inkling, its first public model, and released it with open weights. It's not a small demo model: a large mixture-of-experts class, multimodal, with a long context window, that anyone can download and run themselves.
- Thinking Machines Lab (founded by Mira Murati, former OpenAI CTO) released Inkling - its first model, with open weights.
- Mixture-of-experts architecture: about 975 billion total and ~41 billion active parameters, multimodal, with context up to 1 million tokens.
- Open weights = the model gets downloaded and run on your own infrastructure, without going through someone else's API.
Mira Murati took much of OpenAI's product team with her and raised one of the most heavily funded new labs. Until now we talked about Thinking Machines Lab more for who leads it than for what it does. Their first public move isn't a closed API behind a key. Open weights - a model anyone can download and run themselves.
What matters, what's noise? Here the substance isn't another benchmark score, it's one word: open. A model of this caliber that you download and run yourself means control stays with you. Your data doesn't travel to someone else's server. The price doesn't jump on every request. And nobody can pull your access mid-work. Murati could have gone the OpenAI way - closed, behind a key. She chose the opposite. That's a position, not a technical detail.
For those of us building on these models, open weights are the difference between a tenant and an owner. A tenant lives by the landlord's rules - raise the price or change the terms, and you have no move. One more serious open model on the market means one more option not to be a tenant.