10,000 AI Agents: The Navier-Stokes Proof in Just 88 Hours

OpenAI claims to have solved one of the world's most difficult mathematical problems, which has remained unanswered for nearly ninety years. The Navier-Stokes proof presented by the ChatGPT company pertains to one of the seven Millennium Prize Problems cataloged by the Clay Mathematics Institute and was produced by an internal AI system more powerful than the GPT-6 Astra model. Approximately 10,000 artificial agents worked in parallel, arriving at a solution in just 88 hours.

Summary

• Key points
• The Navier-Stokes problem and why it matters
  • The result: a singularity in fluid motion
• How the multi-agent system worked
• Buckmaster's accusations and OpenAI's response
• Why it matters and what changes now
• FAQ
  • What is the Navier-Stokes existence and regularity problem?
  • How did OpenAI's AI system solve the problem?
  • What does the solution demonstrate about fluid motion?

Key points {#Key_points}

• OpenAI has produced a mathematical proof that resolves the existence and regularity problem of the Navier-Stokes equations.
• A multi-agent system of about 10,000 competing AI agents reached the solution in 88 hours, processing 130 billion tokens and exchanging 2.7 million messages.
• The proof was formalized in Lean and verified with GPT-6 Astra in another 17 hours.
• The solution shows that the motion of a fluid can generate a singularity in finite time, despite viscosity.
• Mathematician Tristan Buckmaster (NYU) accused OpenAI of accelerating the work after learning of a possible parallel solution from him and a researcher at Anthropic; OpenAI denied using their data.

The Navier-Stokes problem and why it matters {#The_Navier-Stokes_problem_and_why_it_matters}

The Navier-Stokes equations describe how fluids move, treating them as a continuous medium rather than as a collection of individual molecules. They are fundamental to aeronautical design, weather forecasting, and studies of blood flow. The open question for nearly a century has been whether this continuous model could "break": that is, whether a fluid initially in regular motion could develop a velocity that grows without bound in finite time, despite viscosity tending to smooth out the motion.

The equations date back to the work of Claude-Louis Navier and George Gabriel Stokes in the 19th century. In 1934, Jean Leray demonstrated that generalized solutions exist, but the question of their regularity remained open. In 2000, the Clay Mathematics Institute included the problem among the seven Millennium Prize Problems, each with a one million dollar prize for anyone who solves it.

The result: a singularity in fluid motion {#The_result_a_singularity_in_fluid_motion}

OpenAI's system produced an analytical proof showing how a fluid initially at rest and regular, subjected to an equally regular force, can develop a singularity in finite time while maintaining finite energy throughout the dynamics. The identified Millennium problem solution is a vortex that wraps inward, progressively elongating like a spiral of spaghetti: the central region narrows as it accelerates, ensuring that energy remains finite, as required by the laws of physics.

The technical challenge was non-trivial: the equations had to generate the breakage on their own, without researchers artificially inserting an infinite force. The terms describing acceleration, pressure gradient, momentum transfer, and viscosity had to grow and precisely cancel each other out, leaving a regular external force even as the fluid's velocity grew without limits.

How the Multi-Agent System Worked

Behind the multi-agent demonstration is a system of coordinated agents powered by an internal model trained since August 28, which is still in the training phase. The agents could access a cached version of the internet and execute code, divided into groups capable of communicating internally. The group that solved Navier-Stokes consisted of about 10,000 concurrent agents.

OpenAI launched the initiative on September 1, after hearing rumors that two Millennium Prize Problems would be solved elsewhere. It then tested its model on all remaining open problems, proposing different variants of the statement to different groups of agents: versions "A" and "B" would lead to a proof, while versions "C" and "D" would lead to a refutation. The agents also solved, almost by chance, a simpler version related to Euler's equations, which then redirected resources towards the main problem.

The agents arrived at the solution on Saturday, September 5, about 88 hours after the first agents were launched. The AI mathematical proof was then formalized in Lean language and verified via GPT-6 Astra in another 17 hours. Just for Navier-Stokes, the system exchanged 2.7 million messages and generated about 130 billion tokens in output.

Buckmaster's Accusations and OpenAI's Response

The announcement quickly became intertwined with controversy. Tristan Buckmaster, a professor at New York University, stated that OpenAI had accelerated its work after learning that he and a researcher from Anthropic were close to announcing their own discovery. Buckmaster added a delicate detail: the couple's work was stored in OpenAI's Codex model, used for writing code, and thus potentially visible to the internal team. On his website, he wrote that he did not know what OpenAI's model had done or whether his data had been used, without directly accusing anyone.

During a press briefing, OpenAI researcher Sebastien Bubeck denied that the company had used the couple's work or had access to the material shared on OpenAI's servers. In its announcement, OpenAI did admit that it could not entirely rule out that de-identified data derived from the use of its products by Alpöge and Buckmaster may have influenced the improvement of the models. However, the company emphasized that the two proofs are substantially different, as they concern distinct variants of the problem: Buckmaster and Alpöge worked on the forced version of Euler's equations, while OpenAI's agents solved the non-forced version.

Why It Matters and What Changes Now

It is a statement that goes beyond pure mathematics: it signals how AI models, powered by enormous amounts of computing power, are entering scientific research territories that were considered exclusive to human work until recently.

For the academic world and investors in the AI sector, the incident raises two distinct but connected questions. On one hand, there is the actual scientific significance of the result, which requires independent verification by the mathematical community. On the other hand, there is the issue of data provenance and competition between laboratories, complicating the interpretation of who really contributed to what.

What is the Navier-Stokes existence and regularity problem?

It is a fundamental question still open: whether the Navier-Stokes equations for a three-dimensional incompressible fluid can develop singularities in finite time, despite the presence of viscosity.

How did OpenAI's AI system solve the problem? {#How_did_OpenAIs_AI_system_solve_the_problem}

OpenAI used a multi-agent system of about 10,000 competing AI agents, coordinated for about 88 hours, arriving at a proof that was then formalized in Lean and verified with GPT-6 Astra.

What does the solution demonstrate about fluid motion? {#What_does_the_solution_demonstrate_about_fluid_motion}

The solution shows that fluid velocities can grow unbounded in finite time, creating a singularity, while the overall energy remains finite: a signal of the breakdown of the continuous approximation of the fluid.

Content created with the assistance of artificial intelligence and human editorial review.

This content is provided for general informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online promotions, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets. Crypto assets are highly volatile and may result in loss. The availability of WEEX services, products, and related events may vary by region. You are responsible for ensuring that your participation is in accordance with applicable local laws and regulations.