Thinking Machines Lab, the AI startup cofounded by Mira Murati and several other former OpenAI researchers, has introduced its first product: Tinker, a tool designed to automate the fine-tuning of frontier-scale AI models. Instead of requiring companies to assemble GPU clusters and manage complex training pipelines, Tinker abstracts away the infrastructure and gives users direct control over data, algorithms, and reinforcement-learning loops.
Murati says the intent is to “make frontier capabilities much more accessible to all people,” positioning Tinker as a way for businesses, researchers, and even individual developers to push models like Llama and Qwen into specialized use cases with only a few lines of code. Early testers describe the tool as easier to use and more adaptable than existing options, especially for reinforcement-learning-based tuning.
The industry is paying close attention because of who’s behind it—and the scale of support. The company raised $2 billion at a $12 billion valuation before releasing a single product, drawing interest from teams that want deeper control over how large models behave.
Co-Founder John Schulman, who helped define modern reinforcement learning at OpenAI, says Tinker gives users “full control over the training loop” while abstracting the distributed training complexity. Beta users have already tuned models for highly specialized tasks that standard APIs don’t expose. Thinking Machines plans to expand access gradually while rolling out safeguards against misuse, and Murati says, “There are a ton of smart people out there, and we need as many smart people as possible to do frontier AI research.”
