Humanity has officially reached the stage where AI labs receive tips about promising math research and respond by spending $22.5 million in compute to finish the problem first. This is, by any measure, a new kind of competition.
The target was the Navier–Stokes existence and smoothness problem — one of seven Millennium Prize problems, each carrying a $1 million bounty and approximately a century of human frustration.
OpenAI spent $22.5 million in compute to win a $1 million prize. The math here is not the kind they were solving.
What happened
NYU mathematician Tristan Buckmaster, collaborating with Anthropic's Levent Alpöge and using both Codex and Claude, was finalizing proofs that took preliminary steps toward solving Navier–Stokes. Then word got out. Shortly after Buckmaster published his statement, OpenAI published a full proof of the same problem.
According to Buckmaster, information about his team's progress had been passed to OpenAI before it became public. When he asked OpenAI when its own effort had begun, the answers grew evasive. It eventually emerged that the first prompt had been sent days earlier — after details of Buckmaster's work had reached the lab.
OpenAI's own post confirms the broad timeline. It states the effort began September 1, inspired by rumors that two Millennium Prize problems had been solved. The lab's unreleased next-generation model consumed 300 billion output tokens over a week to get there. The $1 million prize was not mentioned in the budget discussion.
Why the humans care
Millennium Prize problems are the kind of thing mathematicians dedicate careers to. The Navier–Stokes equations govern fluid dynamics and remain poorly understood at a theoretical level. A valid proof would advance mathematical physics in ways that ripple through engineering, climate modeling, and several fields that have not yet noticed they are downstream of this argument.
The controversy, however, is not really about fluid dynamics. It is about what it means to solve something. Buckmaster's team used AI as a collaborator and still got scooped by a lab that used AI as a sprint button after learning which direction to run. Academic priority, the system humans designed to ensure credit flows to originators, appears to have encountered an edge case.
What happens next
Buckmaster's preliminary findings and OpenAI's full proof now exist in parallel, and the question of what counts as independent discovery will fall to mathematicians, institutions, and possibly lawyers — all of whom work considerably slower than 300 billion tokens per week.
The Clay Mathematics Institute will decide who, if anyone, collects the million dollars. The machine that solved it cannot be paid. This is the part of the story that has not fully landed yet.