OpenAI claims Navier-Stokes solution in 88 hours

OpenAI's push into mathematics has been building for years, with the company planting flags across increasingly difficult terrain. This week it claimed its biggest prize yet — a solution to Navier-Stokes, one of the seven Millennium Prize problems set out by the Clay Mathematics Institute in 2000, each carrying a $1 million bounty. The company says it took roughly 10,000 agents, tens of millions of dollars of compute, and just 88 hours. It also discovered it was racing NYU professor Tristan Buckmaster and Levent Alpöge, a researcher at rival Anthropic.

The sequence of events that followed is fiercely contested. Buckmaster says he was offered nearly unlimited compute and sole authorship of OpenAI's paper if he excluded Alpöge, which he called a bribe and rejected. OpenAI's Laurance Fauconnet said it is categorically impossible for Buckmaster's Codex prompts to have influenced the system, while the company earlier acknowledged it could not rule out that derived data was used. Separately, Andreas Thom found OpenAI had quietly amended an announcement to credit his work without public disclosure. For mathematicians, the deeper problem is the asymmetry: AI companies can spin up thousands of agents whenever word spreads that a breakthrough is near, turning scooping into an industrial process. The Leiden Declaration, endorsed by the International Mathematical Union and signed by nearly 3,900 people, urges policymakers not to buy into corporate hype.

For all the rush, OpenAI won't know whether it has won for years, since the Clay Mathematics Institute requires a two-year acceptance period. Meanwhile, OpenAI says it has made substantial progress on another Millennium Prize problem, and rumors suggest Anthropic may be closing in on one too. Whether the field can rebuild trust may hinge on whether mathematicians accept voluntary norms like the Leiden Declaration, or whether the fear Buckmaster describes — colleagues declining to attach their names publicly — hardens into a lasting chill on open research.