Thinking Machines Launches Tinker; Periodic Labs Raises $300M

Thinking Machines Lab, founded by former OpenAI CTO Mira Murati and other ex-OpenAI researchers including John Schulman, has launched Tinker, an API that automates fine-tuning of large language models. Tinker, now in private beta, is initially free before switching to usage-based pricing and abstracts distributed training so users can customize models via simple Python; it supports models including Alibaba’s Qwen-235B-A22B and Meta’s Llama and uses low-rank adaptation (LoRa) to reduce compute costs. The lab also published an open-source Tinker Cookbook with recipes for math reasoning, conversational datasets, retrieval, and prompt distillation, and research groups at Princeton, Stanford, UC Berkeley and Redwood Research have already run projects on the platform. Separately, Periodic Labs — cofounded by ex-OpenAI contributor Liam Fedus and former DeepMind researcher Ekin Doğuş Çubuk — announced $300 million in funding to build an “AI scientist” that pairs large language models with robots to run high-throughput physical experiments in a Menlo Park lab. These developments illustrate a broader trend of well-funded AI alumni teams building tools and automation to accelerate scientific and AI research.