Anthropic hires OpenAI co-founder Andrej Karpathy for pretraining research

Andrej Karpathy, a founding member of OpenAI and former AI director at Tesla, has joined Anthropic to lead a new pretraining research team. The move reunites one of AI's most influential researchers with frontier model development, this time at the company behind Claude. Karpathy announced his decision on X, stating he expected the next few years at the frontier of LLMs to be especially formative and that he was excited to return to hands-on R&D. His history with OpenAI has been intermittent: he co-founded the company in 2015 but left after two years to work with Elon Musk at Tesla. He returned to OpenAI briefly in 2023 but left again in early 2024 to launch Eureka Labs, an AI education platform focused on a course called LLM101n that covers advanced language modeling topics. The course has not yet been released, and it is unclear if Karpathy remains involved with the startup.

Karpathy's arrival at Anthropic underscores the fierce competition for top research talent among frontier AI companies. His expertise in large language models — from academic research at Stanford to hands-on work at Tesla and OpenAI — makes him a major asset for pretraining, the foundational phase of model development that determines a model's core capabilities. By building a dedicated pretraining team, Anthropic aims to accelerate improvements to Claude's performance, efficiency, and alignment. The hiring comes shortly after Anthropic acquired Stainless, a developer tools startup used by OpenAI, Google, and Meta, further signaling the company's aggressive expansion. For industry observers, Karpathy's move highlights the ongoing talent churn in AI and Anthropic's strategy of securing deep expertise to compete head-to-head with OpenAI.