Mira Murati unveils Inkling, her first open-weight model
Former OpenAI tech chief has released the first model from her Thinking Machines Lab. Inkling is open-weight, so developers can download and modify it, though the company admits it is not the strongest option today.
- Mira Murati, the former OpenAI chief technology officer, has released the first model from her Thinking Machines Lab, named Inkling.
- Inkling is an open-weight model, so developers can download it and adapt it to their own work.
- The company was founded in February 2025 and is valued at around 12 billion dollars.
- The lab openly admits Inkling is not the strongest model today, among neither private nor open options.
What exactly do you get when one of the best-known names in artificial intelligence ships her first model? For now, an honest start. Mira Murati, who ran the technical side of OpenAI until last year, has unveiled Inkling, the first model from her Thinking Machines Lab. And she said the thing you rarely hear at a launch: this is not the strongest model on the market.
Inkling is an open-weight model. In practice that means you do not have to use it only through someone else’s website or app. Developers can download the weights, run the model on their own infrastructure, and tune it for a specific job. For companies that do not want to send every query to a big lab’s server, that is a real difference.
Why does an open model matter more here than raw power?
Murati founded Thinking Machines Lab in February 2025, together with some former OpenAI colleagues. The company is already valued at around 12 billion dollars, even though until recently it had no public product for end users. The money arrived before the model, which says a lot about how much people trust the team behind it.
Inkling fits into what Murati describes as interaction models. The idea is for collaboration between people and machines to feel more natural, for the model to better catch the nuances in how people actually write and ask, instead of returning boilerplate. That is a different emphasis from the race for the highest benchmark score.
That is why the decision to make the model open is more interesting than its raw strength. Open weights mean the community can test it, change it, and build on top of it. If the goal is more natural interaction, then you need a lot of people using it in different ways, not a closed system behind a single button.
What this means for users and companies
For now, Inkling is not a tool you will open instead of ChatGPT or Gemini and feel a difference. The company itself says stronger models exist, both private and open. If you need the best possible answer right away, this is not that model.
The value sits elsewhere. For development teams, including those building their own tools, an open model means control. You can run it locally, keep your data on your side, and change its behavior without waiting for someone else to unlock a feature. Thinking Machines already earns money through Tinker, a tool for customizing models, so it is clear this crew is aimed at people who build, not just those who chat.
Murati chose to start modestly and honestly, with no promise that this is the best in the world. In an industry where every model is announced as a turning point, admitting that something is just getting started feels almost refreshing. The real test comes with the next model.
And the practical question is yours: would you rather use a model you can download and keep on your own hardware, or does it matter more to have the strongest answer even if it runs on someone else’s server?
