What happened
An NYU mathematician, Buckmaster, put out a statement announcing proofs of his own. Inside it was a line he said he wished he did not have to write: a parallel effort by OpenAI, he says, built on his team's work before that work became public.
The result, in his telling, is a tangle of academic rivalries and conflicting claims.
Shortly after his statement, OpenAI posted what it calls a complete solution to the Navier–Stokes existence and smoothness question. Buckmaster's findings had taken steps toward that problem. OpenAI says the proof was found by an unreleased next-generation model, one that has been working through a range of unsolved problems over the past week.
That week was not cheap.
Navier–Stokes is one of the seven Millennium Prize problems. Each carries a $1 million bounty from the Clay Mathematics Institute for the first person or group to solve it. The equations are widely used in fluid mechanics and still poorly understood in theory.
So yes: the compute bill, priced at current Astra rates, is a multiple of the prize money. Nobody said discovery had to be efficient.
Why it matters
Two stories are tangled here, and they should be pulled apart.
The first is capability. If OpenAI's claim holds up, an unreleased model cracked a problem that human mathematicians have chased for generations. That is a vendor claim for now, and the model is not public, so nobody outside can poke at it. Extraordinary claims, meet peer review.
The second is credit. Buckmaster says his unpublished work fed the AI effort. If that is right, the question is who gets named when a lab's model finishes a job that academics started. Mathematics runs on priority and attribution. A frontier lab with a huge compute budget and a publish button changes the power balance in that game fast.
This is, for now, primarily a personality conflict with an unclear resolution. But it previews a fight that every research field will have: humans do the groundwork, a model does the last mile, and the headline goes to whoever hits publish first.
Humans do the groundwork, the model finishes, and the credit goes to whoever publishes first.
What to do with it
- Wait for verification.A claimed Millennium Prize proof needs expert checking before anyone celebrates, so treat the announcement as a claim, not a verdict.
- Read both sides.Buckmaster's statement and OpenAI's proof tell different versions of the timeline, so read both before picking a team.
- Track the model.The model is unreleased, so watch for any public release or technical write-up that lets outsiders test what it did.
- Document your own work.If you research anything, timestamp and publish early, because preprints are cheaper than priority disputes.
Ciao, sapiens. Cite your sources while you still have the only copy of your idea.


