On September 8, 2026, OpenAI published a proposed solution to the Navier-Stokes existence and smoothness problem — one of seven Millennium Prize Problems designated by the Clay Mathematics Institute, each carrying a $1 million award. If the proof holds up to community scrutiny, it would be among the most significant mathematical results in decades, and the first to be produced primarily by an AI system.
The Problem
The Navier-Stokes problem asks a deceptively simple question: given smooth initial conditions, do solutions to the equations that govern fluid motion (liquids, gases, airflow, ocean currents) remain smooth indefinitely, or can they develop singularities — points where the equations break down — in finite time? The problem has remained open for roughly 90 years. It's not just an abstract curiosity; singularity formation in fluid equations has direct implications for turbulence modeling, weather prediction, and aerodynamics.
How OpenAI Ran the Proof
OpenAI deployed approximately 10,000 concurrent AI agents built on GPT-6 Astra, running in parallel for 88 hours. The agents explored different mathematical approaches, communicated intermediate findings to each other, ran tools, consolidated ideas, and eventually handed a proposed solution to a frontier model for formal verification.
The scale of computation is hard to overstate: the run generated 2.7 million messages and approximately 130 billion output tokens. Across all attempted Millennium Prize problems, the total output reached 300 billion tokens.
OpenAI's proposed proof argues that smooth 3D fluid motion can develop a finite-time singularity under smooth external forcing — starting from a fluid at rest, applying a smooth external force while keeping total energy finite, and showing the system blowing up in finite time. The proof was subsequently formalized and checked in Lean 4, a mathematical proof assistant, giving the result machine-verified internal logical consistency — a higher bar than most mathematical proofs ever clear.
The Controversy
The reception from the mathematics community has been cautious. The Clay Mathematics Institute's prize process requires proposed solutions to survive prolonged expert scrutiny and gain broad acceptance before any award is considered. OpenAI explicitly stated it does not intend to claim the prize.
A separate controversy involves data provenance. VentureBeat reported that OpenAI "can't rule out" that the agents benefited from a researcher's private Codex data that was inadvertently accessible to the system during the run — a concern that would not invalidate the math itself, but raises questions about the boundary between AI discovery and AI-assisted retrieval of existing human work.
The Lean formalization, while impressive, also isn't finished. Reviewers have noted that verifying that the formalized statement exactly matches the mathematical requirements of the Millennium Prize formulation is itself a nontrivial task that still requires human expert judgment.
What to Make of It
Even if Navier-Stokes remains officially open for another few years pending community review, what OpenAI has demonstrated is something new: that a large fleet of AI agents can produce a mathematically coherent, formally verified argument for a problem that stumped human mathematicians for nearly a century. The question of whether that constitutes "solving" the problem is partly philosophical. What's not in dispute is that this kind of computational approach — thousands of agents, massive token budgets, formal verification — has now produced results at the frontier of pure mathematics. That's a line that's been crossed, regardless of how the prize committee eventually rules.