- Sources: OpenAI account, relayed, Buckmaster statement, TechCrunch, discussion, HN thread
- Summary: OpenAI states an unreleased internal model produced a resolution to the Navier-Stokes existence and smoothness problem, one of the seven Millennium Prize Problems, with the agents reaching it on 2026-09-05 about 88 hours after launch and Lean formalization and verification taking a further 17 hours through GPT-6 Astra, using 2.7 million agent messages and approximately 130 billion output tokens on that problem within 4.9 million messages and about 300 billion output tokens across all attempted problems. Tristan Buckmaster published a statement alleging OpenAI took up the approach he developed with his collaborator Levent Alpoge, an Anthropic employee whom OpenAI stated it would not invite as a co-author because of its competitive relationship with his employer, after learning of it from him. OpenAI states that neither its researchers nor its agents saw any of that work before public release and that no specific user data was accessed to solve the problem, states separately that it cannot rule out that de-identified data derived from Buckmaster and Alpoge's product usage helped improve its models, and records that its proofs differ significantly from theirs with the precise results proved differing in the Euler case.
- Why it matters: A vendor's agents are credited with a Millennium Prize resolution that has not been peer reviewed, and the contested question is whether original work run through that vendor's coding agent stays the user's own.
- Follow-up: Track peer review of the resolution, OpenAI's further response to the statement, and any correction to either account.
send feedback on this story