Thinking Machines Lab, the artificial intelligence research company founded by former OpenAI CTO Mira Murati, has launched its first product called Tinker, a flexible API designed to simplify fine-tuning of large language models. The company announced the launch on October 1, 2025, marking a significant milestone for the startup that secured $2 billion in seed funding earlier this year.
Tinker provides researchers and developers with granular control over model training while abstracting away the complexity of distributed training infrastructure. According to the official announcement, the API enables users to “control every aspect of model training and fine-tuning while we handle the infrastructure.”
The API supports fine-tuning of open-weight models including Meta’s Llama and Alibaba’s Qwen series, including the massive Qwen-235B-A22B model. Fine-tuning is performed using LoRA (Low-Rank Adaptation) technology, which trains small add-ons to model weights rather than modifying the entire model, making the process more efficient and accessible.
Tinker’s core API functions include forward_backward for gradient calculation, optim_step for weight updates, sample for token generation, and save_state for preserving training progress. This allows researchers to focus on algorithms and datasets rather than engineering overhead.
Mira Murati stated on social media: “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.”
Early users have praised the platform’s capabilities. Tyler Griggs noted that “Tinker lets us focus on the research, rather than spending time on engineering overhead,” while researchers Ziran Yang, Yong Lin, and Chi Jin highlighted that “the training infrastructure has been abstracted away, which makes focusing on our data and evals far easier.”
Currently available in private beta, Tinker is free to start with usage-based pricing planned for the coming weeks. The company has also released an open-source Tinker Cookbook library with state-of-the-art post-training methods to assist developers in building custom AI applications.
Thinking Machines Lab, founded in February 2025, positions Tinker as a research-focused alternative to mainstream AI products, targeting developers and researchers who need fine-grained control over model customization rather than end-user applications.