---
title: "OpenAI Announces Solution to the Navier-Stokes Millennium Prize Problem: 10,000 AI Agents Find Counterexample in 88 Hours That Baffled Mathematicians for Two Centuries"
description: "OpenAI announced that its AI system found a counterexample to the Navier-Stokes equations—one of the \"Millennium Problems\"—through 88 hours of work by 10,000 agents. Computing costs were about $1 million, and the solution was formalized in Lean."
date: 2026-11-15T10:38:51.865Z
lang: en
url: https://xab.info/en/posts/openai-navier-stokes-ai-millennium-math
tags: [openai, artificial-intelligence, navier-stokes, mathematics, agi, clay-mathematics-institute, deep-learning]
author: "Alexander Buran"
publisher: "XAB.info"
---

# OpenAI Announces Solution to the Navier-Stokes Millennium Prize Problem: 10,000 AI Agents Find Counterexample in 88 Hours That Baffled Mathematicians for Two Centuries

![Neon processor icon with a lightning bolt inside on a dark blue background](https://xab.info/media/2026/11/15/openai-navier-stokes-ii-matematika-tysiacheletie/openai-navier-stokes-ii-matematika-tysiacheletie-1.webp)

## Key Points

- OpenAI announced the solution to the Navier-Stokes problem—one of the seven Clay Mathematics Institute "Millennium Prize Problems"
- The solution was found by 10,000 AI agents in 88 hours, generating about 300 billion tokens
- It was proven that Navier-Stokes equations can generate physically impossible solutions with infinite velocity
- Computing costs are estimated at ~$1 million, matching the Clay prize amount
- The result was formalized in the Lean language, allowing for automated proof checking

In early September 2026, OpenAI announced the solution to one of the seven “Millennium Prize Problems”—the Navier-Stokes equations, included in 2000 by the Clay Mathematics Institute in a list of unsolved problems carrying a one-million-dollar reward. According to the company, its artificial intelligence system, which internally exhibits signs of Artificial General Intelligence (AGI), proved through a specific example that the Navier-Stokes equations under certain conditions generate solutions physically impossible in reality: in finite time, a vortex forms in a fluid where the flow velocity tends to infinity. The result was formalized in the Lean programming language, which allows for automated verification of mathematical proofs, giving the conclusion additional weight according to experts. The announcement sparked an intense discussion in the scientific community about what the role of the human mathematician is now and what the point of doing science is if an algorithm finds the key answers.

### Two Centuries of Chaos in a Formula

The Navier-Stokes equations describe the motion of viscous fluids and gases—from airflow around an airplane wing to river currents and atmospheric vortices. Nearly two hundred years ago, French engineer Claude-Louis Navier and Irish mathematician George Gabriel Stokes attempted to combine all phenomena occurring during the interaction of fluid particles into a single mathematical apparatus: pressure, resistance, acceleration, external forces. The researchers relied on Newton's second law—"force equals mass times acceleration"—and concluded that to describe motion, it is sufficient to specify velocity and pressure at every point of the occupied volume. However, no generation of mathematicians has ever managed to prove that these equations always yield correct, smooth solutions in three-dimensional space. This was precisely the essence of the “Millennium Problem”.

### 88 Hours, 10,000 Agents, and 300 Billion Tokens

According to OpenAI, the solution was obtained during an 88-hour “assault” involving approximately 10,000 AI agents working simultaneously. Throughout the experiment, the system sent nearly 4.9 million messages and generated about 300 billion output tokens; the Navier-Stokes task itself accounted for 2.7 million messages and roughly 130 billion tokens. Different groups of agents explored their own approaches, and the most promising ideas were combined via the Codex AI agent. The company emphasized that it used an internal model “significantly more capable” than the Astra system, whose training began on August 28, 2026. The strategy was phased: first, the agents attacked a simplified version—the Euler equations for an ideal fluid without viscosity. After 50 hours of work by 1,000 agents, a solution was found. Following this, 10,000 agents were deployed to tackle the full Navier-Stokes problem, and the next day the result was formalized in Lean.

### Millions of Dollars for a Proof

The project turned out to be orders of magnitude more expensive than OpenAI's previous mathematical experiments. In the previous phase, the Astra system achieved 10 breakthroughs in other problems at a cost of about $2,000. Work on Navier-Stokes, according to OpenAI Director of Research Mark Chen, cost “millions of dollars.” An independent estimate by Kartik Duraisamy of the University of Michigan, who is preparing a special project to track computing resources, aligns with this figure: if OpenAI's tools were used at retail prices, the cost would be around $6 million, while actual computing power expenses were roughly $1 million—which, as Duraisamy notes, “matches the exact prize amount for solving the millennium problem.”

### “Such a Pace of Progress Must Be Taken Entirely Seriously”

OpenAI Chief Scientist Jakub Pachocki drew a sharp parallel at the company's press event with the level of capabilities a year ago: “Just a year ago, we were only figuring out whether AI models could compete with high schoolers in math olympiads. Now the situation is different.” According to him, solving the Navier-Stokes problem is “an evaluation of a system possessing general-purpose intelligence,” and the pace of progress demonstrated by the company “must be taken entirely seriously.” Already in early October 2026, OpenAI announced the resolution of hundreds of mathematical problems, including the so-called quasi-hypothesis, pointing to a systemic rather than isolated breakthrough. Meanwhile, a debate has ignited in the scientific community: some researchers speak of the “impending collapse” of classical mathematics as an independent discipline, while others emphasize that formalization in Lean and the need to interpret results preserve a key role for humans in problem formulation and evaluating their physical meaning.

## Fact-Check & Sources

- [Задача тысячелетия: выживет ли математика? . ИИ решает проблемы, над которыми математики бились десятилетиями. Что будет с наукой?](https://novayagazeta.eu/articles/2026/11/15/zadacha-tysiacheletiia-vyzhivet-li-matematika) - The primary source contains direct quotes from OpenAI employees (Jakub Pachocki, Mark Chen), specific technical parameters (10,000 agents, 88 hours, 300 billion tokens, Lean), and an independent cost estimate by Kartik Duraisamy (University of Michigan). The facts are consistent with one another.

## FAQ

### What is the Navier-Stokes problem and why is it considered a "Millennium Problem"?

It is one of seven unsolved mathematical problems included in 2000 by the Clay Mathematics Institute in the list of "Millennium Problems" with a $1 million prize. The task is to prove that the equations describing the motion of viscous fluids and gases always yield correct solutions. OpenAI demonstrated a counterexample: under certain conditions, the equations generate a vortex with infinite velocity, which is physically impossible.

### How much did the solution cost and who estimated it?

According to OpenAI Director of Research Mark Chen, costs amounted to "millions of dollars." Independent researcher Kartik Duraisamy from the University of Michigan estimated actual computing costs at roughly $1 million, while retail pricing for OpenAI's tools would be around $6 million.

### How exactly did the AI arrive at the solution?

First, 1,000 agents solved a simplified version—the Euler equations for an ideal fluid—in 50 hours. Then, 10,000 agents were assigned to the full Navier-Stokes problem, generating about 130 billion tokens in 88 hours. The most promising approaches were combined via the Codex AI agent, and the final proof was formalized in Lean.

### What does this mean for the future of mathematics?

A debate has broken out in the scientific community: some researchers speak of a possible "collapse" of classical mathematics as an independent discipline, while others emphasize that problem formulation, result interpretation, and physical meaning remain in human hands. In early October 2026, OpenAI already announced the solution of hundreds of mathematical problems, pointing to the systemic nature of progress.