动察 Beating AI Flash: OpenAI has published an AI-completed proof of the Navier-Stokes equations. This is a mathematical problem that remained unsolved for approximately 90 years, and one of the seven "Millennium Prize Problems," carrying a $1 million prize.

OpenAI stated that it used an internal model still in training, significantly stronger than GPT-6 Astra, with roughly 10,000 agents working collaboratively to find a solution in 88 hours, followed by an additional 17 hours using Astra to complete Lean machine verification. The paper and verification code have been made public, but they still await review by the mathematical community.

Behind this achievement lies a controversy over priority of publication. New York University mathematician Tristan Buckmaster and Anthropic employee Levent Alpöge had previously advanced research in the same direction with the help of Claude and Codex. Buckmaster claimed that after their unpublished progress reached OpenAI, the company poured massive computational resources into the same approach within days, and even proposed that Buckmaster rewrite OpenAI's Navier-Stokes proof, but without allowing Alpöge to be credited as an author.

OpenAI now acknowledges that this sprint did begin with rumors that "Anthropic may have solved a Millennium Prize Problem." On September 1, the team began having the new model tackle multiple mathematical challenges, first using nearly 100 agents to produce results on Euler equations, then concentrating resources on Navier-Stokes, ultimately scaling up to about 10,000 agents. OpenAI denies having seen Buckmaster's research before his paper was made public, and denies accessing specific user data, but has not completely ruled out another possibility: that de-identified data generated by the two individuals' use of OpenAI products may have been used to improve the model.

Sébastien Bubeck also responded to the controversy over authorship and the remark about "destroying a career." He stated that he did not ask Alpöge to relinquish authorship of their original work; what was discussed was having Buckmaster take the lead on writing OpenAI's proof, and he believed it was inappropriate to credit an Anthropic employee on OpenAI's work. Regarding the phrase "why would you want to destroy your own career," Bubeck admitted the wording was poor and apologized, but denied it was a threat. Sam Altman also publicly sided with Bubeck and confirmed that OpenAI had initially heard the rumors about Anthropic and wanted to see if its own model could achieve the same feat.

OpenAI stated it will not claim the $1 million prize associated with this challenge.