Thinking Machines Lab released Inkling on July 15, its first model built from scratch, under Apache 2.0.
The lab openly admits Inkling is not the strongest model available, and that is the entire point.
The bet: companies want AI they can own and reshape, not rent from a frontier lab.
Most AI launches open with a claim of dominance. This one opened with an admission of weakness, and that is why people are paying attention.
What Inkling actually is
Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, released its first in-house model on Wednesday.
Inkling is a mixture-of-experts system with 975 billion total parameters, though it only draws on a fraction of that, about 41 billion, for any given task, a design that keeps very large models faster and cheaper to run.
It was trained on 45 trillion tokens of text, image, audio, and video, and reasons natively across all four, with a context window of up to one million tokens.
The licence is the story. It ships under Apache 2.0, meaning developers can download, modify, and deploy it with virtually no strings attached. Full weights sit on Hugging Face. Fine-tuning runs through Tinker, the company’s own customisation platform, and customers own whatever they build.
Here is the launch post:
https://x.com/thinkymachines/status/2077454609551921208
Why “not the best” is the pitch
Thinking Machines is clear that Inkling is not the strongest model available, instead focusing on how the model is customizable, which could help users get better performance with lower costs. That sounds like a weak launch until you read the numbers behind it.
One differentiator is controllable thinking effort. Developers tune how many tokens the model burns on a problem. In testing, Inkling matched Nemotron 3 Ultra on Terminal Bench 2.1 at roughly one third the token cost.
The proof point the lab keeps pointing at is Bridgewater. The hedge fund and the lab trained an open model on Bridgewater’s financial expertise and scored 84.7% on financial reasoning tests, beating top proprietary models at a fraction of the cost.
Read that with one eye open. That figure comes from the two companies’ own evaluation, not an independent one.
The wider argument is gathering weight. Microsoft CEO Satya Nadella warned that enterprises using proprietary AI models effectively pay twice, once in subscription costs and again by handing over business knowledge embedded in their prompts and corrections, which can be absorbed into future model versions.
The risk sits with the lab, not the user. Giving away your best technology under a permissive licence is bold when you have raised $2 billion at a $12 billion valuation. Thinking Machines now has to prove that fine-tuning, support, and platform lock-in pay the bills. For everyone else, a frontier-scale model just became free to download.
Quick Links:
Comments
Be the first to leave a comment.