OpenAI agents claim Navier-Stokes blowup, a $1m Millennium Prize maths problem


For the first time, an artificial-intelligence system is claiming a solution to one of the Clay Mathematics Institute’s Millennium Prize Problems. On 8 September, OpenAI said an internal model running about 10,000 autonomous agents found a finite-time singularity in the three-dimensional Navier-Stokes equations, the mid-19th-century model of how viscous fluids move. The company released both an analytical writeup and a Lean formalization, and said the agents reached a proof after roughly 88 hours, with another 17 hours of Lean checking via GPT-6 Astra. Quanta Magazine and Nature both treated the claim as the largest AI-derived maths result to date if it survives scrutiny, and as a possible turning point in how hard problems get attacked.

The Millennium formulation asks whether smooth solutions to Navier-Stokes in unbounded three-dimensional space can develop infinite velocity in finite time while energy stays finite. OpenAI says its agents established Clay statements “C” and “D”: blowup is possible under a smooth forcing. Hours earlier, NYU’s Tristan Buckmaster and Anthropic’s Levent Alpöge had announced AI-assisted results on related Euler (zero-viscosity) problems and an unverified Navier-Stokes variant, after a summer of Lean-checked LLM proofs. Both camps lean on analytic cascade techniques pioneered by Diego Córdoba and Luis Martínez-Zoroa, who had already produced singularities with less tidy forcing. Charles Fefferman, who wrote Clay’s official problem statement, told Quanta he was thrilled the problem was solved and credited Córdoba and Martínez-Zoroa as the intellectual heroes.

A Lean check is strong evidence, not a blank cheque. Priority fights between the Buckmaster–Alpöge line and OpenAI are already messy, and human mathematicians still have to confirm that the formalized statement matches the Clay problem. Even so, the week’s message is blunt: multi-agent systems with millions of dollars of compute are now competing on century-old open problems that once looked safely beyond machine reach. Mereja will watch the Clay Institute and the peer community for the verdict that actually moves the $1 million prize.

Sources:

OpenAI

Quanta Magazine

Nature

Leave a Reply

Your email address will not be published. Required fields are marked *

+ 11 = 19