Mira Murati's Thinking Machines Lab launches Tinker
Mira Murati, former CTO of OpenAI, just launched Tinker with her startup Thinking Machines Lab.
Tinker is an API designed to help researchers fine-tune language models—big or small—with less code and hassle.
It handles the tricky parts of distributed training so you can focus more on what matters.
Currently in private beta
Tinker is currently in private beta and already being tested by teams at Princeton, Stanford, Berkeley, and Redwood Research. They're using it for everything from solving math proofs to chemistry reasoning and reinforcement learning.
The API supports a range of models, including huge ones like Qwen-235B-A22B.
Pricing details
Instead of tweaking entire models (which can get expensive), Tinker offers a more efficient approach to model updates—making the process much cheaper and faster.
For now, it's free during beta; pricing will kick in after launch.
Open-source Tinker Cookbook released
Thinking Machines Lab also dropped the open-source Tinker Cookbook—an open-source library with modern post-training methods built on the Tinker API.
By taking care of things like distributed computing behind the scenes, it lets researchers spend more time experimenting with data and algorithms instead of wrestling with infrastructure.