OpenAI Says Internal AI System Solved the Navier–Stokes Millennium Prize Problem

OpenAI says an internal AI system has produced a proof that the three-dimensional Navier–Stokes equations can develop a singularity in finite time — a resolution of the Navier–Stokes existence and smoothness problem, one of the seven Millennium Prize Problems established by the Clay Mathematics Institute in 2000. The company released both a written proof and a formalization in the Lean proof assistant, and says the result was machine-verified.

The stakes are hard to overstate. The question of whether smooth three-dimensional fluid motion can break down has remained unresolved for roughly 90 years, since Jean Leray proved in 1934 that solutions exist in a generalized sense. A resolution touches the mathematical foundations of models used in aircraft design, weather forecasting, and the study of blood flow, and it arrives as AI labs race to demonstrate that their systems can do frontier research, not just chat.

OpenAI states plainly why it published: "We believe it is important to inform the world about the pace of AI progress and what to expect from upcoming models." The company says it does not intend to claim the Millennium Prize, and frames the result as "not a culmination, but rather a snapshot in time, of progress on AI development."

What was actually proved

The Navier–Stokes equations apply Newton's second law of motion — "F=ma" — to fluids, treating a fluid as a continuous medium rather than tracking individual molecules. The core open question was whether that continuum approximation can break down: can the equations for a three-dimensional incompressible fluid with constant density develop a "singularity," even when the motion starts smoothly?

Here, a singularity means the dynamics drive speeds in the fluid to grow without bound within a finite amount of time. Because viscosity tends to smooth out motion, and because a real fluid cannot move infinitely fast, such a breakdown would mark a failure of the equations as a model of the fluid — to continue, one would need to track each particle individually.

According to OpenAI, the system produced an analytical proof and a Lean formalization that an initially smooth fluid at rest can develop a singularity in finite time. The fluid has a smooth force applied to it, and its energy remains finite through the entire dynamics, from rest to the formation of the singularity. OpenAI says this resolves the problem by establishing statement "C" (and also "D") in the official Millennium Prize formulation — that is, the disproof side rather than the positive-existence statements "A" and "B."

The solution itself has a vivid geometric shape. "The solution is a vortex, a spinning swirl of fluid, that spirals inward and gets increasingly elongated, like spaghetti," the company writes. The central region shrinks while speeding up in a way that keeps its energy finite, as physics requires. The technical crux: the breakdown must emerge through the motion of the fluid itself, not from an infinite force inserted by hand. The terms describing acceleration, pressure gradients, momentum transfer, and viscosity "must both become big yet cancel in a precise way," leaving a smooth external force even as the velocity grows without bound.

The equations date to the nineteenth-century work of Claude-Louis Navier and George Gabriel Stokes.

How the proof was found

The story of the discovery is as unusual as the result. OpenAI says it has been training a new internal model since August 28 that "has exhibited unprecedented performance in our benchmarks, including mathematics," and that the model's training is ongoing with performance continuing to improve. The company describes the model as "significantly more capable than GPT‑6 Astra."

On Tuesday, September 1, OpenAI researchers heard rumors that two Millennium Prize problems had been resolved. "Inspired by these rumors and by the step change in performance of our internal model, we launched an effort to evaluate it on all open Millennium Prize problems and a few other high-impact problems," the company writes.

The system was a swarm of coordinating agents with access to tools including a cached version of the internet and the ability to run code. Agents were subdivided into groups that could communicate internally, and the group that produced the Navier–Stokes resolution ran on the order of 10,000 concurrent agents. OpenAI says it maintained the same safeguards it applies to all frontier model evaluations, including monitoring and isolation, throughout.

Different groups received different variants of each problem statement. For Navier–Stokes, some groups worked on versions "A" and "B" — particular forms that would result in a proof of smoothness — while others worked on versions "C" and "D," which would result in a disproof.

The first breakthrough came on an easier problem. OpenAI asked the multi-agent system to try a set of warm-up questions, including a blowup question for the limit of Navier–Stokes with the viscosity term removed: the regularity problem for the Euler equations. "Our agents surprised us by resolving this question," the company writes. The specific variant resolved was the unforced version, with no external force applied to the fluid. Nearly 100 agents worked together for approximately 50 hours to produce the Euler regularity disproof.

That result redirected the entire effort. OpenAI shifted agents away from the other Millennium problems, prompted them with the Euler resolution, and updated the agents to a further-trained version of the internal model when it became available mid-effort. To cross-pollinate the groups, researchers used Codex to consolidate the most useful insights from each group into follow-up prompts drawing on the agents' own intermediate results. The group that ultimately found the Navier–Stokes solution was guided this way.

The agents arrived at their resolution on Saturday, September 5 — about 88 hours after the first agents were launched. Lean formalization and verification took an additional 17 hours via GPT‑6 Astra.

The scale of the compute is concrete. Across all attempted problems, the agents sent 4.9 million messages and used about 300 billion output tokens. The Navier–Stokes effort alone accounted for 2.7 million messages and roughly 130 billion output tokens.

The concurrent-work controversy

The September 1 rumor, OpenAI later realized, related to Levent Alpöge, an Anthropic employee, and Tristan Buckmaster, a math professor at NYU. After completing the project and Lean verification on September 6, OpenAI reached out, believing from the rumor that the pair also had a Navier–Stokes solution, to offer a concurrent release and recognize their priority in a joint announcement.

It turned out that Alpöge and Buckmaster, using an internal Anthropic model, had produced a resolution of the forced Euler problem — not Navier–Stokes. "We recognize the priority of their work on forced Euler and congratulate them on their remarkable mathematical achievement," OpenAI writes. The company says it offered them visibility into all of the prompts used and, later, the proof itself.

The overlap raises obvious questions about information flow between a frontier AI lab and its competitors' internal usage. OpenAI addresses this directly: "We (the researchers and the agents) did not see any of their work through any means until they released it publicly — in particular, no specific user data was accessed in order to solve this problem." The company adds that an investigation confirmed Buckmaster's Codex prompts over the two months preceding the September 8, 2026 announcement and paper "could not have influenced the system in any way, including through training." It also notes the internal model was developed through large-scale reinforcement learning on top of a previously pretrained model, and that the mathematical results differ: Alpöge and Buckmaster proved a result with external forcing, while OpenAI's system proved one without.

The fact that a competing lab's employee was using Codex to attack the same problem — and that OpenAI felt compelled to run a formal investigation into possible prompt leakage — is itself a signal of how tightly coupled the frontier AI ecosystem has become, and how much scrutiny data boundaries at AI labs will now face.

What it means

OpenAI is explicit that the release is a progress report, not a prize claim. "Our goal in releasing this result is to report on the substantial progress of our AI models," the company writes, adding that the milestone "represents substantial work by mathematicians and AI researchers." A major stated goal is "to empower scientists to advance research and technology that benefits all of humanity."

The company links the result to its broader thesis that a new phase of AI capability has begun, and says it is focusing on understanding this model and using what it learns "to help us guide and pace how we pursue further advances in capability." One stated key goal is building AI systems "which are steerable, accountable, and connected to people, which may require more deliberate choices about the pace of progress."

Whether the mathematical community accepts the proof — even one formalized in Lean — will depend on scrutiny of the writeup itself, and on whether Clay Institute procedures treat an AI-generated solution as prize-eligible given OpenAI's decision not to claim it. What is already clear: two rival labs, using internal models, reached blowup results on centuries-old fluid equations within days of each other in September 2026, and both chose to announce them nearly simultaneously.