Drama swirls around OpenAI’s legendary mathematical milestone

OpenAI announced that an internal AI model it trained starting August 28 produced a solution to the Navier‑Stokes problem, which concerns the equations governing fluid flow. The company said the model — described as more powerful than the recently released GPT‑6 Astra and run with roughly 10,000 concurrent agents — has shown “unprecedented performance” on internal benchmarks, including mathematics.

The Navier‑Stokes question is one of seven Millennium Prize Problems; each carries a $1 million award. OpenAI said it does not plan to accept the monetary prize. The company’s blog post framed the result as a major breakthrough for the mathematics community while acknowledging debate about how the work was produced.

Controversy erupted after NYU mathematics professor Tristan Buckmaster and researcher Levent Alpöge (affiliated with Anthropic) published related findings the day before OpenAI’s announcement. Buckmaster told OpenAI he had been putting drafts of their work into Codex sessions during their project and later questioned whether those sessions or other user data had been accessed or used to train OpenAI’s model. He said he was told the model did not “look up user data,” and that a question about training did not receive an answer.

OpenAI pushed back in its announcement, stating that “no specific user data was accessed in order to solve this problem,” while also allowing that it cannot rule out the possibility that de‑identified data derived from product usage could have helped improve models. Sebastien Bubeck, on OpenAI’s technical staff, said the team did not see Buckmaster and Alpöge’s work until it was publicly released and that the proofs produced by OpenAI differ significantly. Buckmaster later posted on Mastodon asserting that OpenAI’s statements indicated they had used training data from a period after his team had found their result, keeping the dispute unresolved publicly.