What just happened

OpenAI says it solved the Navier–Stokes problem — one of the hardest open questions in mathematics — in 88 hours. The company set roughly 10,000 AI agents on it and, by the following morning, had an answer.

It matters because this isn't just any puzzle. It's one of the seven Millennium Prize problems, each with a $1 million reward attached. This one has resisted a proof for 90 years.

What the problem actually is

The Navier–Stokes equations describe how fluids move — water, air, the blood in your veins. They work perfectly in practice. The unsolved part is a question of trust: the equations describe turbulence — the chaotic swirl of a river or a gust of wind — and nobody has been able to prove that the math always gives a clean, smooth answer, with no hidden blow-ups.

It's a proof problem, not an engineering problem. Planes fly and weather models run fine. The question is whether the underlying math is guaranteed to hold.

How OpenAI did it

OpenAI trained a new internal model that turned out to be unusually good at math, then pointed 10,000 AI agents at the problem. The agents exchanged nearly 3 million messages and burned through 130 billion output tokens before landing on a solution. At OpenAI's own pricing, the compute would have cost roughly $10 million.

The result resolved two of the four statements the Millennium Prize proof requires.

The honest caveats

Nothing here is settled yet. The solution has not been independently verified, and the Clay Mathematics Institute hasn't accepted it. OpenAI itself says it does not intend to claim the $1 million prize. And two mathematicians — one at NYU, one at Anthropic — have said they were working on the same problem and that information about their progress reached OpenAI. OpenAI denies using their work and says the proofs differ.

What it means for you

Forget the prize. The real story is the method: throw thousands of AI agents at a hard problem for a weekend and see what comes back. That's now a legitimate research approach — and it's the same idea behind using an AI agent to chew through a tedious work task you've been avoiding. The cost of trying has dropped to almost nothing.

The bottom line

Whether or not the proof holds up, "10,000 agents, one weekend, $10 million" is a preview of how hard problems get tackled from now on.