Andrej Karpathy Joins Anthropic's Pre-Training Research Team

OpenAI co-founder Andrej Karpathy has joined Anthropic's pre-training team under Nick Joseph, where he will lead a group using Claude to accelerate pre-training research.

Andrej Karpathy Joins Anthropic's Pre-Training Research Team

Andrej Karpathy, an OpenAI co-founder and one of the most-watched figures in modern AI, announced Tuesday that he has joined Anthropic. He started this week on the pre-training team under team lead Nick Joseph, and will run a group focused on using Claude itself to accelerate pre-training research.

What He's Working On

Pre-training is the large-scale phase that gives Claude its core knowledge — the part of the stack that most directly determines what a model knows and how well it reasons before any fine-tuning or alignment work. Karpathy's mandate is meta: use Claude to make the next Claude better at pre-training, a feedback loop that has quietly become one of the most active internal projects across every frontier lab.

The Resume

Karpathy was a founding member of OpenAI in 2015, departed in 2017 to run Tesla's Autopilot and Full Self-Driving programs, returned to OpenAI in 2023 for a year, then left in 2024 to start Eureka Labs, an AI-for-education startup. He is also the most widely followed educator in deep learning — his Stanford CS231n notes, "Neural Networks: Zero to Hero" YouTube series, and recurring essays on "vibe coding" have shaped how a generation of researchers learned the field.

Why Anthropic, Why Now

The hire fits a broader pattern. Anthropic passed OpenAI in annualized revenue in April, hitting $30B ARR to OpenAI's $24B — the first time the smaller lab has been ahead on the top line. Internal headcount has roughly doubled since the start of 2025, and the company has been among the most aggressive bidders in the industry for senior research talent. Pulling an OpenAI founding member is the loudest version of that strategy yet.

Karpathy framed the move on X as a "personal update," describing the next few years at the frontier of LLMs as "especially formative." That phrasing — "next few years," not "next decade" — is doing some work. The implication is that whoever wins the pre-training race in the back half of the 2020s gets the durable advantage, and he is betting that team is Anthropic's.

What to Watch

Eureka Labs remains his startup, and Tuesday's note did not say he was winding it down — he framed the Anthropic move as an additional engagement rather than a full pivot. For practical purposes, though, having an OpenAI co-founder publicly working pre-training inside a competitor reshapes the talent-market signal: Anthropic is no longer just attracting senior ICs and team leads, it is pulling the people whose names alone move the conversation.

Comments

Get tomorrow's roundup. Free.

One email each morning. Sneakers, sports, culture, tech.

Link copied