Did AI Just Solve Navier-Stokes? What OpenAI's Claim Actually Proves

OpenAI is claiming that one of its internal systems has cracked one of the Clay Institute's Millennium Prize Problems. This is a remarkable story, and it would represent the biggest AI-assisted mathematical discovery yet. But there's also a second story going on: a fight over who really gets credit.

Here's what's actually happened: OpenAI heard a rumor that a rival lab's model had made progress on one of mathematics' most famous unsolved problems. That rumor was enough to trigger a sprint: To work the problem, OpenAI used 10,000 parallel AI agents, 88 hours of compute, and roughly $22.5M in accumulated costs. Now, they are claiming a real proof. And about that $1M prize? OpenAI says it doesn't intend to claim it. (Maybe they don't need the money.)

Quick answer: The $1M Clay Millennium Prize for Navier-Stokes has not been won. What OpenAI has claimed to prove targets a specific variant of the problem, blow-up under smooth forcing, that remains unverified by independent mathematicians. Whether it satisfies the full problem as most experts understand it is genuinely contested.

What Is the Navier-Stokes Equation, and Why Is It Worth $1 Million?

To understand what OpenAI is claiming, you first need to understand what makes Navier-Stokes so hard.

The Navier-Stokes equations describe how fluids move: water, air, blood. They underpin weather forecasting, aircraft design, and circulatory medicine. Formulated in the 19th century, they work extraordinarily well in practice. The mathematical question is whether they always work: do smooth, well-behaved solutions always exist, or can a fluid governed by these equations develop a singularity, a point where velocity becomes infinite, in finite time?

That question, known as the "existence and smoothness" problem, is one of the Clay Mathematics Institute's seven Millennium Prize Problems, each carrying a $1M reward. As of today, only one of the Millennium Prize Problems has ever been solved: the Poincaré Conjecture, in 2003.

Why we care about this particular problem: Sometimes, math problems are more or less strictly academic. But this one isn't just an abstract puzzle. Every weather model, airplane simulation, and engineering tool that relies on these equations is implicitly assuming they behave well, but no one has proven that assumption always holds. If it turns out solutions can blow up in finite time, that would expose a real gap between the math we rely on and the physical systems it's supposed to describe.

What OpenAI Is Actually Claiming to Have Proven

The story starts not with a mathematical insight but with a rumor. Sam Altman admitted OpenAI launched the effort after hearing that Anthropic's models had solved a major math problem, writing on X: "We were curious if ours could do it too."

The week-long sprint covered multiple Millennium Prize problems and consumed 300 billion output tokens in total, roughly $22.5 million in compute across everything attempted. The Navier-Stokes problem alone used approximately 130 billion of those tokens and 2.7 million messages. Ten thousand agents worked in parallel from September 1st to September 6th. And then - this is the part that should stop you - OpenAI said it doesn't intend to claim the $1M prize, framing the result as evidence of how fast its models are advancing.

Before arriving at Navier-Stokes, OpenAI's agents first resolved the unforced Euler regularity problem, using nearly 100 agents in roughly 50 hours. By establishing blow-up behavior in the simpler Euler setting first, the agents built the conceptual foundation that made the Navier-Stokes approach tractable.

Here's where the fine print matters. The official Clay formulation for Navier-Stokes includes four options, labeled A, B, C, and D. Options A and B ask for global smooth solutions with no external forcing, in ordinary three-dimensional space or a periodic domain. Options C and D allow blow-up examples with a smooth external forcing term. OpenAI's proof targets options C and D, meaning the forced variant is genuinely part of the official Clay problem.

That said, many mathematicians consider options A and B to be the deeper question, because forcing is externally imposed: you're choosing the force to cause the blow-up, rather than asking whether the equations can break down on their own. Whether the approach can be extended to remove the forcing and address A and B is not yet clear, and that question is very much still open. The gap between what's been shown and what many experts consider the heart of the problem is the actual takeaway here.

Inside the Fight Over Who Deserves Credit

Now that we've mapped what was actually proven, the human story has a clear throughline, if you step back from the he-said/she-said.

On August 15, Tristan Buckmaster and Anthropic's Levent Alpöge proved that the Euler equations, the frictionless cousin of Navier-Stokes, can blow up under smooth forcing, extending an approach developed by Diego Córdoba and Luis Martínez-Zoroa. Their results also covered the Boussinesq and incompressible porous media equations. That work took roughly a year. It was unpublished, in progress, quietly built. Then word got out.

OpenAI's sprint began on September 1st after hearing a rumor that they later connected to Alpöge and Buckmaster. Buckmaster had emailed OpenAI privately on September 3rd after hearing his work had reached the company. After completing their project and Lean verification on September 6th, OpenAI reached out to offer a concurrent release. Buckmaster's version of what came next is considerably sharper: his account describes calls in which he says he was pressed over publication and authorship, including suggestions to exclude Alpöge because of his employment at Anthropic.

Buckmaster alleges that when he signaled his intent to go public with these concerns, he was told, "Why would you ruin your career?" and "If you don't want me to be nice, then I don't have to be nice." Bubeck has called these allegations "false and inflammatory." OpenAI denies accessing any private prompts or unpublished work.

I'm not going to adjudicate who's right. These are contested accounts from a private conversation, and the specifics are genuinely unclear. But the pattern doesn't require resolution to be worth naming: an independent researcher's year of unpublished work became the resource a well-funded lab ran a multi-million-dollar sprint around, and now the independent researcher is the one fielding questions about his career.

Can You Actually Trust an AI-Written Math Proof?

With the credit dispute on the table, the harder technical question remains: does the proof hold up?

Navier-Stokes is a Millennium Prize problem, and the standard of evidence is a released, verifiable proof. OpenAI published a 166-page manuscript and a Lean formalization on September 8th, both publicly available for scrutiny. Independent peer review hasn't happened yet.

If you aren't familiar with Lean, know it's a formal proof verification system that checks whether each logical step follows from the previous one. A Lean-verified proof can't be "talked into" looking correct: if it passes, the logical chain is sound. What Lean doesn't do is tell you whether the approach is conceptually meaningful, whether it addresses the problem as experts understand it, or whether a human mathematician would recognize it as a genuine solution.

Terence Tao, arguably the most prominent living mathematician, offered a warning about this dynamic on September 5th, before OpenAI's announcement. Writing on Mathstodon in response to rumors circulating at the time, he noted it was a hypothetical concern: "there is a substantial opportunity cost in converting a historically productive and motivating problem such as Navier-Stokes regularity into a mere viral social media post advertising some benchmark progress, rather than actually advancing the field and developing the next generation of both problems to ask, and people to work on them." After the announcement, Tao praised the Buckmaster-Alpöge work as "a remarkable achievement" and noted their arguments had been formalized in Lean. He hasn't publicly endorsed OpenAI's specific claimed proof.

Why This Fight Is Bigger Than One Math Problem

A rumor, not a published result, not peer review - a rumor - was enough to mobilize one of the world's best-resourced AI labs into a multi-million-dollar sprint. OpenAI framed the result as evidence of how quickly its most advanced systems are improving, which is to say, as a marketing asset. This is a Millennium Prize problem being used as a benchmark demonstration ahead of a public offering.

Axios put the uncomfortable question plainly: what happens when the company providing scientists with AI research tools can also mobilize vastly more resources to compete with them? Buckmaster and Alpöge were using OpenAI's own Codex throughout their year of work. The tools a researcher uses to build toward a result can be owned by the same organization that can outpace them with those same tools at 10,000x scale. That's not a conspiracy. It's a structural feature of the current moment, and it's the scenario people have been worried about, independent of whether OpenAI did anything wrong here.

There's also an irony worth noting: OpenAI paused certain frontier RL training over safety concerns in August, then in September ran an unreleased model from the same development program at full capacity for 88 hours to chase a math headline. That juxtaposition isn't proof of bad faith, but it's not nothing, either.

So, Is Navier-Stokes Solved or Not?

Here's where things actually stand:

The sprint model produced a result. Whether it produced a result, in the sense mathematicians mean, is a different question, and one that'll take considerably more time to answer.