Sam Altman on OpenAI/Anthropic Navier-Stokes millennium-problem dispute

TL;DR

Sam Altman’s public statement on the OpenAI-Anthropic dispute over a claimed model-generated proof of a Navier-Stokes millennium problem. Altman confirms that OpenAI’s team, led by Sébastien Bubeck, reached out to coordinate a joint release after discovering that Anthropic’s team (with Tristan and Levent) had solved the Euler equation but not Navier-Stokes; he says OpenAI offered to let Anthropic go first and offered Tristan lead-author position on a rewrite of the OpenAI proof, but coordination broke down and Anthropic threatened plagiarism accusations. Altman further concedes OpenAI attempted the problem because of internet rumors that Anthropic had solved a millennium problem. The tweet is filed as a datapoint on frontier-lab math-proof capability claims and inter-lab priority disputes, not as a technical artifact.

Key claims

OpenAI’s team reached out first to coordinate a joint release and, on learning that the other team had Euler but not Navier-Stokes, offered to let them go first [tweet].
OpenAI offered Tristan (the non-Anthropic member of the other team) lead-author position on a rewrite of the OpenAI proof; the same offer was not extended to Levent (Anthropic) because he “was not willing to talk or coordinate” [tweet].
The other team responded with what Altman calls “unfounded accusations of plagiarism” and refused to communicate; OpenAI did not rush to publish despite this [tweet].
On inspection, Altman claims the two approaches “appear to be different” [tweet].
Altman explicitly says OpenAI tried the problem because of internet rumors the prior week that Anthropic’s models had solved a millennium problem [tweet].
Altman claims OpenAI’s “latest model can solve many, many other math problems” beyond this one [tweet].

Method

Not applicable — the tweet is a public statement, not a technical artifact. Bubeck’s quoted parent tweet clarifies that his outreach message to Levent was aimed at release coordination and that he never asked for Levent’s removal from authorship of Levent’s own work.

Results

Not applicable. The tweet’s substantive claim about model capability — “our latest model can solve many, many other math problems” — is unsupported by any citation or paper.

Why it’s interesting

Two things worth tracking:

Frontier-lab public claims about open-ended math capability are moving fast. This sits next to Gemini 2.5 Pro Capable of Winning Gold at IMO 2025 (Gemini 2.5 Pro solves 5/6 IMO 2025 problems with verifier-guided scaffolding), DeepSeekMath-V2: Towards Self-Verifiable Mathematical Reasoning (DeepSeekMath-V2’s self-verifiable proof pipeline), and Cursor: agent swarm discovers novel solution to Problem Six of the First Proof challenge (Cursor’s agent swarm on the First Proof Problem Six). The Navier-Stokes-scale claim, if it survives peer review, is a step above IMO-level and beyond anything currently filed here — but Altman’s tweet by itself is not evidence of that claim.
Attribution and inter-lab coordination is becoming a governance surface. The dispute is entirely social — priority, joint release, author list — and echoes classical academic disputes rather than anything specific to ML. Worth noting alongside The Hugging Face incident and the road ahead and Brief independent investigation of agents' behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident as another instance of frontier labs airing coordination failures in public.

Neither paper on the underlying proofs has surfaced in the wiki; when either does, this tweet page should be revisited to check whether Altman’s account of the sequence of events holds up.