Mira Murati, former CTO of OpenAI, has introduced the first product from her new venture Thinking Machines Lab—an API called Tinker, designed to make fine-tuning of large and small open-weight AI models more accessible. The tool aims to remove the technical burden of distributed training while giving researchers the flexibility to control algorithms and datasets.
Announcing the launch on X, Murati wrote: “Today we launched Tinker. Tinker brings frontier tools to researchers, offering clean abstractions for writing experiments and training pipelines while handling distributed training complexity. It enables novel research, custom models, and solid baselines. Excited to see what people build.”
Currently in private beta, Tinker lets developers scale seamlessly from lightweight models to massive architectures like Qwen-235B-A22B with as little as a single line of Python code change. While the service is free at launch, Thinking Machines Lab plans to introduce usage-based pricing in the coming weeks.
The API has already drawn interest from a number of leading research groups. Early adopters include Princeton’s Goedel Team, which has used Tinker to train mathematical theorem provers; Stanford’s Rotskoff Chemistry group, applying it to reasoning-heavy tasks; Berkeley’s SkyRL group, leveraging it for multi-agent reinforcement learning; and Redwood Research, experimenting with challenging AI control problems.
To complement the API, the lab has also released the Tinker Cookbook, an open-source library containing modern post-training methods built directly on the Tinker platform. The Cookbook is designed to accelerate experimentation and encourage customization, lowering the barrier for both established researchers and independent developers to test new ideas quickly.
Murati’s vision for Thinking Machines Lab is to democratize AI research by enabling seamless experimentation at scale. By blending ease of use with flexibility, Tinker provides a foundation for advancing both academic and applied AI, opening opportunities for breakthroughs across domains.
With its launch, Tinker positions itself as a potential game-changer in the fine-tuning space, bridging the gap between cutting-edge infrastructure and the creativity of researchers worldwide.
