OpenAI Claims ~10,000 AI Agents Cracked the Navier–Stokes Millennium Prize Problem in 88 Hours (September 2026)

Background ​

The Navier–Stokes existence and smoothness problem is one of the seven Millennium Prize Problems established by the Clay Mathematics Institute in 2000, each carrying a $1 million prize; only the Poincaré conjecture (proved in 2003) has previously been solved. The problem asks whether solutions to the Navier–Stokes equations of fluid motion are always smooth and globally defined, and has resisted mathematicians for about 90 years.

On September 8, 2026, OpenAI published an announcement titled "On the Navier–Stokes Millennium Prize Problem," stating that an internal AI system (about 10,000 cooperating agents) produced a solution proving that a three-dimensional incompressible fluid can develop a singularity in finite time (finite-time blowup), accompanied by both a written proof and a Lean formalization.

1. Key figures from OpenAI's announcement ​

Media reports (Sina Tech, France 24), citing OpenAI, put the compute cost in the millions of dollars.

2. What the claimed result says ​

The paper's conclusion: for any positive viscosity, one can construct a three-dimensional incompressible fluid starting at rest that, under smooth forcing, develops locally unbounded velocity in finite time while total kinetic energy remains bounded — i.e., under smooth initial data and smooth forcing, solutions of the NS equations form a singularity in finite time, ruling out the "always smooth and globally defined" direction. The proof is built around a vortex construction (Quanta Magazine's accompanying image shows the vortex, with yellow indicating faster rotation and blue slower).

A caveat on the statement: Forkast News reports that the Clay Millennium Prize criteria are based on the unforced version of the NS equations, whereas OpenAI's result addresses the forced version; the correspondence needs to be confirmed by the mathematics community.

3. Reactions and the priority dispute ​

Quanta Magazine reports that, if the result survives further scrutiny, it would be by far the most important mathematical proof produced by an AI model to date, possibly marking a turning point in how mathematicians approach hard problems. Separately, NYU professor Tristan Buckmaster and Anthropic mathematician Levent Alpöge announced solutions to several closely related problems about 12 hours before OpenAI's announcement; both teams relied heavily on analytic approaches by Diego Córdoba (Institute for Mathematical Sciences, Madrid) and Luis Martínez-Zoroa (CUNEF University). Princeton professor Charles Fefferman, who wrote the Clay Institute's official description of the problem, said "the heroes are Córdoba and Martínez-Zoroa"; Buckmaster said Martínez-Zoroa deserves a Fields Medal.

Priority dispute: Buckmaster stated that his team (with Alpöge) had worked on the problem for almost a year, with a breakthrough on August 15. OpenAI says its effort began September 1 after hearing rumors, denies seeing the other team's work by any means and denies accessing specific user data, while acknowledging it "cannot rule out that de-identified data derived from their usage of our products helped improve our models." OpenAI offered to publish concurrently with Buckmaster's team and recognize their priority, but Alpöge was not invited as a co-author (OpenAI cited its competitive relationship with his employer). Per media reports, OpenAI says it will not claim the prize money.

4. Significance and outlook ​

If the result holds and is validated by the Clay Institute, Navier–Stokes would become the second Millennium problem solved, and the first Millennium-level proof produced primarily by an AI system. The episode highlights three emerging issues: (1) formal verification (Lean checking) raises confidence in AI mathematical proofs but does not replace the judgment of whether the proof addresses the original problem; (2) "rumor-driven" research races — merely learning that a problem has been solved may trigger massive compute spending to get there first; and (3) the boundary between AI training data and priority over research results.

Common misconceptions and details ​

1. "OpenAI cracked the Millennium problem" is an announcement, not an established fact. As of September 9, 2026, the result has not been peer-reviewed and the Clay Institute has not validated it; "solved" in headlines should be read as OpenAI's unilateral claim.
2. Lean verification is not the same as solving the problem. Lean checks that the proof text is internally consistent; it does not automatically guarantee the conclusion maps precisely onto the original problem statement (e.g., the unforced version).
3. The result has no direct engineering consequences. The mathematical singularity holds under idealized continuum assumptions; real fluids are made of molecules, so the result does not directly change engineering fluid dynamics practice.

Summary ​

On September 8, 2026, OpenAI announced that its internal model and about 10,000 AI agents produced a solution to the Navier–Stokes existence and smoothness problem in 88 hours, with a 165-page writeup and a Lean 4 formalization. The announcement has triggered broad discussion in the mathematics community over the validity of the result, its correspondence to the original problem statement and priority. As of this writing, the result has not passed peer review or Clay Institute validation, and its final status remains to be established.

Sources and references ​

• OpenAI official research page (entry to the NS announcement): link
• Simon Willison: On the Navier–Stokes Millennium Prize Problem: link
• Quanta Magazine: AI Has Solved One of Math's $1 Million Millennium Prize Problems: link
• NetEase: 10,000 agents, 88 hours — OpenAI cracks a math Millennium problem: link
• Sina Tech: OpenAI announces NS Millennium solution, costing millions of dollars: link
• Forkast News: OpenAI's 10,000-Agent Navier-Stokes Claim: link
• Superintelligence News: OpenAI's Navier-Stokes Claim Triggers a Debate: link