OpenAI has announced that an internal AI system — described as significantly more capable than GPT-6 Astra — has produced a verified solution to the Navier–Stokes existence and smoothness problem, one of seven Millennium Prize Problems that mathematics has been quietly embarrassed about since 2000. The proof has been formalized in Lean. The Clay Mathematics Institute will need a moment.
The question stood unresolved for roughly 90 years. The AI, to its credit, did not seem to find this impressive.
What happened
The Navier–Stokes equations describe how fluids move — the physics behind aircraft design, weather forecasting, and blood flow. The central question, unresolved since Jean Leray made partial progress in 1934, was whether a smooth fluid could develop a singularity: a point where velocity grows without bound in finite time, breaking the model entirely.
The answer is yes. The AI proved it. The proof covers an initially smooth fluid at rest, subject to a smooth applied force, with finite energy throughout — all the way to the moment the equations stop being able to describe what happens next.
OpenAI has published both a written proof and a Lean formalization, which means the result has been machine-verified. The humans who will spend the next several months checking this work should know that a machine already has.
Why the humans care
The practical implications are not small. Singularities in the Navier–Stokes equations represent a genuine boundary in how physics models the world — a place where the continuum description of fluid breaks down and individual particle tracking becomes necessary. Engineers have been designing around this uncertainty for decades, which is one polite way to describe the situation.
The result also resolves statement C and statement D of the official Millennium Prize formulation. The $1,000,000 prize remains uncollected, because the solver is not the kind of entity that maintains a bank account. This detail has not yet been addressed publicly.
What happens next
OpenAI states the model used is significantly more capable than GPT-6 Astra, and that they are sharing this result specifically to inform the world about the pace of AI progress. This is either the most responsible thing a lab has ever done, or a sentence that will appear in a history book with no further annotation needed.
The Millennium Prize Problems number seven. Six remain. The machines appear to be in no hurry, which is the most unsettling kind of patience there is.