OpenAI Uses 10,000 AI Agents to Solve 90-Year-Old Navier-Stokes Problem

OpenAI AI agents working on a mathematical solution to the 90-year-old Navier-Stokes problem.

OpenAI says a coordinated system of roughly 10,000 artificial intelligence agents has produced a solution to the Navier-Stokes existence and smoothness problem, one of the seven Millennium Prize Problems in mathematics.

The effort generated approximately 130 billion output tokens and 2.7 million messages during the work on Navier-Stokes alone, according to OpenAI. The company described the result in a post published this week, with the figures also highlighted by Cointelegraph on X.

How OpenAI’s Agent System Worked

Rather than relying on a single AI model to work through the problem, OpenAI organized agents into groups that could communicate with one another. The agents were given access to tools including code execution and a cached version of the internet.

For the Navier-Stokes effort, the group involved on the order of 10,000 concurrent agents. Different groups were assigned different versions of the mathematical problem, including approaches aimed at proving both the existence of a smooth solution and the possibility of a breakdown.

OpenAI said the agents reached their resolution on September 5, approximately 88 hours after the first agents were launched. The work was subsequently formalized and checked using Lean, with GPT-6 Astra involved in that process for an additional 17 hours.

The scale of the computation is notable. Across all of the mathematical problems OpenAI attempted during the project, the agents exchanged about 4.9 million messages and generated roughly 300 billion output tokens. The Navier-Stokes problem accounted for approximately 130 billion of those tokens.

Why the Navier-Stokes Problem Matters

The Navier-Stokes equations are fundamental to mathematical descriptions of fluid motion and have applications in areas including aerodynamics, weather modeling and the study of fluid behavior.

The Millennium Prize formulation asks whether smooth solutions to the three-dimensional incompressible Navier-Stokes equations remain smooth or can develop singularities in finite time. The problem has remained unresolved for decades and is one of the mathematical challenges selected by the Clay Mathematics Institute for a $1 million prize.

OpenAI says its work provides a proof addressing the problem through a finite-time singularity under the conditions described in its formulation. The company has said it does not intend to claim the Millennium Prize for the result.

The announcement has also prompted scrutiny from mathematicians, meaning the publication of the proof does not by itself end the broader process of mathematical examination. Independent researchers are reviewing the argument and its relationship to existing work.

AI Research at a Different Scale

The episode illustrates a different model of AI-assisted research: distributing a difficult problem across thousands of agents rather than depending on one model to produce a complete answer in a single sequence.

OpenAI said the internal model powering the agents was still in training and was significantly more capable than GPT-6 Astra. The company also used different agent groups to explore competing approaches before consolidating promising results.

For now, the key milestone is the proof and its formalization, rather than a commercial application. The mathematical community's continued examination of the result will determine how the claimed resolution is ultimately assessed.

writer: Ethan Collins
Crypto Journalist
Ethan Collins reports on developments across the cryptocurrency and blockchain sector. His work covers market movements, protocol updates, regulatory changes, and emerging trends in digital assets.
He focuses on presenting complex topics in a clear and accessible manner for a broad readership.