What Thinking Machines Lab does
Thinking Machines Lab is an AI research and product company founded in February 2025 by former OpenAI CTO Mira Murati. The lab builds customizable, multimodal frontier models and ships research-grade tools that lower the cost of adapting open-weight LLMs to specialized tasks. Its debut product, Tinker, is a managed API for distributed fine-tuning of open-source models like Llama and Qwen using LoRA, letting researchers and engineers run post-training jobs without managing GPUs, schedulers, or sharding.
Who it's for
Tinker targets ML researchers, post-training teams at AI companies, and engineering groups inside enterprises that want to specialize an open-weight base model on their own data. Thinking Machines also publishes research on LLM determinism, RL, and evaluation through its Connectionism blog, making it a reference point for practitioners who want to understand how frontier models actually behave.
Pricing
Tinker is a usage-based API priced per training token and per hour of distributed compute. Pricing is enterprise-negotiated for production volumes. There is no free plan today; access is gated through a research preview waitlist.
Team & funding
Thinking Machines Lab has assembled an extraordinary founding team: CEO Mira Murati (ex-OpenAI CTO), co-founders John Schulman (OpenAI co-founder, RLHF pioneer), Barret Zoph (former OpenAI VP of Research, Post-Training), Lilian Weng (former OpenAI VP), and CTO Soumith Chintala (creator of PyTorch). The company closed a record $2 billion seed round in July 2025 led by Andreessen Horowitz at a $12B post-money valuation, with participation from NVIDIA, Accel, ServiceNow, Cisco, AMD, and Jane Street. By late 2025, Bloomberg reported the company was in talks to raise additional capital at roughly a $50B valuation. Headcount sits around 30 elite researchers and engineers, with a partnership announced with NVIDIA in March 2026 to deploy one gigawatt of Vera Rubin compute.