Thinking Machines Lab

Thinking Machines Lab's first revenue product is Tinker, launched October 1, 2025: a managed API for fine-tuning open-weight language models. Customers — research groups, startups, and enterprises that want customized models without operating distributed training infrastructure — pay for the compute their training runs consume, making this consumption-based infrastructure in the literal sense: the meter runs on training jobs, not seats. Tinker abstracts cluster management, failure recovery, and distributed-training complexity while leaving customers in control of their data and methods, positioning the lab between raw GPU rental (cheaper but harder) and closed fine-tuning endpoints (easier but less control). The honest caveat: monetization is young, and the $2B seed at a $12B valuation in July 2025, led by Andreessen Horowitz, was raised on the team and research ambition rather than on Tinker's revenue. Expect the model to widen toward enterprise contracts and additional research-derived products, because usage-priced fine-tuning alone is unlikely to carry a $12B expectation.