Max Schwarzer, who led post-training for some of OpenAI's most important recent models, announced Monday that he's leaving the company to join Anthropic. Schwarzer held the title of VP of Research at OpenAI, where his team handled the post-training work behind GPT-5, its 5.1 and 5.2 successors, and the coding-focused 5.3-Codex.
The move adds another name to a growing list of senior OpenAI researchers who've ended up at its crosstown rival. But Schwarzer's departure is notable less for the drama and more for the specifics: he helped build the reasoning paradigm that made OpenAI's o1 models possible, and now he's taking that expertise somewhere else.
What he actually built
Schwarzer joined OpenAI in late 2023, straight out of his PhD at Mila in Montreal, where he studied reinforcement learning under Aaron Courville and Marc Bellemare. That background turned out to matter. He landed on the internal Strawberry team, the group responsible for o1-preview, which shipped in September 2024 as OpenAI's first reasoning model. In his departure post on X, Schwarzer said o1-preview started as one of his "derisking runs," a casual framing for what became a flagship product line.
He later worked on post-training for o1 and o3 alongside Eric Mitchell and Yann Dubois, then took over as head of the post-training team in early 2025. By September 2025, OpenAI had promoted him to VP of Research. For someone who joined as a new grad less than two years prior, that's a steep trajectory, and it suggests his work was central to OpenAI's reasoning push.
The Anthropic pull
Schwarzer framed the move as a return to hands-on research. "After leading post-training for a year, I'm longing to start fresh and return to IC research work," he wrote, using the industry shorthand for individual contributor. He said he wants to get back into RL research specifically.
"Many of people I most trust and respect have joined Anthropic over the last couple of years, and I'm excited to work with them again," he added, which is a polite way of saying the talent pipeline from OpenAI to Anthropic has become a well-worn path. His X bio already reads "RL" and "@AnthropicAI."
The numbers back up the anecdote. A SignalFire report from mid-2025 found that OpenAI engineers were eight times more likely to leave for Anthropic than the reverse. Anthropic's retention rate for recent hires sat at 80%. Those figures predate Schwarzer's move, but they describe the dynamic he's now part of.
Why it matters for OpenAI's model pipeline
Post-training is where raw model capabilities become usable products. It covers reinforcement learning from human feedback, safety tuning, instruction following, and the kinds of behavioral adjustments that make a model feel polished rather than erratic. Losing the person who ran that process for GPT-5 and its successors creates a gap, even if Schwarzer insists his team is "set up to succeed going forward without me." Departing leaders always say that.
OpenAI still has deep bench strength, and the GPT-5.2 and 5.3-Codex models Schwarzer's team shipped are already in production. The GPT-5 launch touted 45% fewer hallucinations than GPT-4o and 80% fewer with reasoning enabled. Whether those gains survive the leadership transition is the question nobody can answer yet.
Schwarzer thanked OpenAI CEO Sam Altman, applications lead Fidji Simo, CTO Mark Chen, and Meret Riera in his post. He called OpenAI's researchers "some of the most talented I have ever met." The praise reads as genuine, not as the kind of performative warmth that usually precedes a messy exit. This looks more like a researcher who got what he needed and moved on.
Anthropic, now valued at roughly $380 billion, has been on a hiring tear across research and engineering. Schwarzer's arrival gives the company someone who has shipped reasoning models at scale, a capability Anthropic has been developing in its own Claude model family but hasn't yet matched OpenAI on in production. His start date hasn't been disclosed.
