OpenAI says one of its internal systems has cracked a problem that has resisted mathematicians for over ninety years: whether the Navier-Stokes equations, which describe how fluids move, can spin off a singularity — a point where flow speed shoots to infinity. It's one of seven Millennium Prize Problems the Clay Mathematics Institute set out in 2000, each carrying a $1 million reward, and only one has ever been solved.
According to the company, roughly 10,000 AI agents running on an unreleased model more capable than GPT-6 Astra spent 88 hours on the problem, exchanging 2.7 million messages and burning through about 130 billion output tokens. GPT-6 Astra itself then took another 17 hours to formally check the proof line by line in Lean, a language mathematicians use to verify logical arguments.
The timing looks awkward. Twelve hours before OpenAI's announcement, NYU mathematician Tristan Buckmaster and Anthropic researcher Levent Alpöge published a related result of their own. Buckmaster dismissed one of the resulting write-ups as "AI slop" and suggested OpenAI's team may have seen his unpublished work before releasing theirs. Neither side agrees on who actually got there first.
Princeton's Charles Fefferman, one of the mathematicians most familiar with this exact problem, called the news "thrilling" but pointed to the real breakthrough elsewhere: a new technique from mathematicians Diego Córdoba and Luis Martínez-Zoroa that made both computer-assisted proofs possible in the first place. No mathematical body has certified OpenAI's result yet, and that kind of scrutiny for a Millennium Prize claim typically takes years, not days.
