Mira Murati’s Stealth AI Lab Launches Its First Product (www.wired.com)

Stepping out of stealth, Thinking Machines Lab — the well-funded startup cofounded by ex-OpenAI leaders including Mira Murati and John Schulman — launched Tinker, a tool that automates creation of custom “frontier” AI models. Tinker aims to lower the bar for fine-tuning large models by abstracting away cluster management and distributed training while giving users direct control over data and algorithms. The move is notable both for the pedigree and funding behind the team ($2B seed, $12B valuation) and for its potential to democratize frontier research at a time when many top US models are closed. Technically, Tinker currently supports supervised fine-tuning and reinforcement learning workflows on open-source models (Meta’s Llama and Alibaba’s Qwen) via a simple API; users can run training with a few lines of code, download resulting models, and deploy them anywhere. The system preserves low-level tunability (full control over the training loop) while automating scaling and stability. Beta users report it makes RL-based capability discovery far easier. That democratization accelerates innovation but raises dual-use risks: Thinking Machines vets access now and plans automated safeguards later. The launch signals a push toward more open, tunable frontier models and could reshape who can build state-of-the-art AI — for better research access and for new safety challenges.