OpenAI claims AI cracked Navier–Stokes, a $1 million problem

The Navier–Stokes equations describe how liquids and gases move — airflow around planes, weather, blood flow — and the unresolved question is whether a smooth starting state can develop a point where velocity grows without bound in finite time. Physical water does not suddenly reach infinite speed, so a mathematical example of such a blowup would mean the equations themselves fail at that point. The Clay Mathematics Institute selected Navier–Stokes in 2000 as one of seven Millennium Prize Problems, each carrying a $1 million award.

What stands out beyond the proof itself is how OpenAI says the result was reached. Rather than a single chat session, it describes roughly 10,000 AI agents exploring different directions in parallel, passing promising ideas between groups and shifting compute toward the most fruitful paths. The scale — about 130 billion tokens of output for Navier–Stokes alone — suggests this is not a method most labs could reproduce easily, at least for now.

The claim now enters a slower phase. The 166-page proof and Lean code must be read independently by mathematicians, and OpenAI says the internal model used was still in training and not yet publicly available. Because other mathematicians were also using AI on related fluid-equation problems at around the same time, questions about research priority and the independence of ideas appear likely to continue alongside the technical review.