OpenAI claims a solution to the Navier–Stokes problem

What this is about

OpenAI published a claimed solution to the Navier–Stokes existence and smoothness problem on September 9, 2026, one of the seven Millennium Prize Problems defined by the Clay Mathematics Institute. The company says its construction shows that an initially smooth, stationary three-dimensional fluid under a smooth external force can develop a singularity in finite time while its energy remains finite.

According to OpenAI, the proof is available both as a mathematical write-up and as a Lean formalization. This is an exceptionally large claim, but not yet a generally accepted resolution. On September 10, the Clay Mathematics Institute still listed Navier–Stokes as “Active.” OpenAI explicitly says it does not intend to claim the Millennium Prize.

What the claimed proof actually does

The Navier–Stokes equations describe the flow of water, air, and other fluids. The open question can be simplified as follows: Does a smooth three-dimensional flow always remain mathematically controlled, or can its velocity or derivatives grow without bound in finite time? Such a point is called a singularity and marks a breakdown of the smooth solution.

OpenAI says its system constructed the second outcome: a vortex that spirals inward, stretches along its axis, and accelerates. The critical step is balancing acceleration, pressure, momentum transport, and viscosity so that the external force remains smooth. The company says the result establishes statements C and D in the official problem formulation. This is a forced flow; the result does not say that every real fluid will spontaneously move infinitely fast.

Why it matters

Navier–Stokes underpins models for aircraft, weather, and blood flow. A resolution of the Millennium problem would clarify a fundamental limit of this mathematical description. More immediately, the release illustrates how AI systems can organize mathematical research. OpenAI reports about 10,000 concurrently coordinated agents in the successful group, roughly 2.7 million messages, and approximately 130 billion output tokens for the Navier–Stokes effort.

The company says the agents reached the result about 88 hours after launch, followed by another 17 hours for Lean formalization and verification using GPT‑6 Astra. Those numbers document the reported compute and coordination effort, not the theorem’s correctness. CNN, Axios, The New York Times, and The Wall Street Journal covered the claim, while the substantive scientific assessment must come from independent experts.

In plain language

Imagine cake batter in a bowl. Viscosity normally slows every motion down. The claimed proof instead describes a carefully shaped vortex that contracts and spins faster without an infinitely powerful mixer. The mathematics is meant to show that the batter’s flow loses its smooth description at one point, not that a real kitchen bowl will explode.

A practical example

An independent review could divide a proof with 300 central lemmas among 30 specialists. Each person checks ten lemmas, searches for hidden assumptions, and compares the equations with the official problem statement. In parallel, reviewers compile the Lean file in a clean environment and inspect whether every assumption is visible.

If three groups identify the same unsupported transition, that passage must be repaired. A successful Lean check would be strong evidence for the internal logical consistency of the formalization. It would not by itself prove that the informal theorem was translated correctly into Lean or that it exactly matches the Clay problem. The numbers in this review example are hypothetical; they do not describe a completed review.

Scope and limits

First, the central evidence currently comes from OpenAI itself. News coverage confirms the publication, not the mathematical truth. Until independent specialists reproduce every step, it should be described as a claimed proof.

Second, a Lean formalization is not a magical seal of approval. The proof assistant checks deductions from formalized assumptions. Errors can still occur in translating the mathematical problem, defining objects, or selecting assumptions.

Third, a mathematical singularity does not automatically make current weather, aviation, or medical software invalid. Numerical models use finite resolution and specific boundary conditions. Recognition and any prize also remain open: the Clay Mathematics Institute continued to list the problem as active after publication, and OpenAI says it will not apply for the prize.

SEO & GEO keywords

OpenAI, Navier–Stokes, Millennium Prize Problem, Clay Mathematics Institute, Lean, formal proof, mathematical AI, fluid dynamics, singularity, GPT‑6 Astra, multi-agent system, AI for science