OpenAI system presents solution to Navier-Stokes problem

OpenAI presented a mathematical proof and verification in Lean on September 5 resolving the Navier-Stokes existence and smoothness problem. The result shows that three-dimensional fluid motion governed by the Navier-Stokes equations can develop a singularity within a finite period. An internal artificial intelligence system generated the entire analytical argument.

The Navier-Stokes equations rely on Isaac Newton’s second law of motion to calculate the movement of fluids as continuous matter. Engineers and meteorologists apply these formulas to aircraft aerodynamics, climate predictions, and blood circulation analysis.

Smooth motion breaks down.

Mathematicians Claude-Louis Navier and George Gabriel Stokes established the underlying equations during the nineteenth century. Jean Leray proved in 1934 that generalized solutions exist, but researchers spent the next nine decades trying to determine whether smooth solutions always persist. The Clay Mathematics Institute formally designated the smoothness question as one of seven Millennium Prize Problems in 2000.

On the same topic: Anthropic researcher resigns over artificial intelligence peril

The mathematical mechanism relies on a vortex spiraling inward with severe axial elongation, causing the fluid velocity to grow without bound while maintaining a precise balance among acceleration, pressure gradients, momentum transfer, and viscosity to keep total energy finite. This behavior confirms statements C and D under the official Millennium Prize framework. The breakdown emerges entirely from the fluid motion rather than from infinite external forces.

Coordinated agent architecture behind the mathematical proof

OpenAI started training the underlying internal model on August 28, observing capabilities beyond GPT-6 Astra in mathematical problem solving. The engineering team deployed a network of coordinating agents on September 1 to test all outstanding Millennium Prize formulations. Approximately 10,000 concurrent agents operated together inside isolated sandboxes equipped with web caching and code execution environments. The research group divided agents into separate clusters assigned to opposing proof hypotheses.

Before targeting fluid equations, a team of nearly 100 agents resolved the unforced Euler regularity problem within 50 hours. That outcome prompted OpenAI to reallocate computing resources directly toward Navier-Stokes.

Codex periodically synthesized the most effective techniques generated across different clusters to redirect the agent teams toward viable paths. The agents completed the Navier-Stokes resolution approximately 88 hours after initial deployment, followed by 17 hours of automated formalization through GPT-6 Astra. Across all evaluations, the multiagent system exchanged 4.9 million messages and processed 300 billion output tokens. The Navier-Stokes workflow accounted for 2.7 million messages and 130 billion output tokens.

Read also: AI has over 10% risk to end humanity, Anthropic researcher says

The calculation ran continuously.

Interaction with independent research efforts

OpenAI initiated the effort after tracking rumors concerning independent work by Anthropic researcher Levent Alpöge and New York University mathematics professor Tristan Buckmaster. Following internal verification on September 6, OpenAI contacted both researchers to propose a coordinated announcement, discovering that Levent Alpöge and Tristan Buckmaster had solved the forced Euler equation. OpenAI acknowledged their achievement and shared prompt histories and proof drafts.

OpenAI stated that neither the researchers nor the autonomous agents viewed external research material prior to public release. The company clarified that it does not intend to submit an official claim for the financial award associated with the Millennium Prize.