OpenAI says AI resolved the Navier–Stokes Millennium Prize Problem

On 8 September 2026, OpenAI said an internal AI model, run as roughly 10,000 coordinating agents, produced a written proof and a Lean formalisation showing the forced, three-dimensional Navier–Stokes equations can develop a singularity — a point of infinite speed — in finite time. That satisfies options C and D of the Clay Mathematics Institute's 2000 problem statement, one of seven Millennium Prize Problems worth $1 million each. The agents reached the result after 88 hours; a separate model spent a further 17 hours formalising and checking it in Lean, the automated proof assistant.

OpenAI's fluid needed a specifically constructed external force to break down. Most mathematicians have long focused on the harder unforced case, where a fluid breaks itself with no outside push — and that remains open. A proof posted 17 September 2026 goes further, showing OpenAI's forcing method cannot be extended to reach the unforced case at all.

OpenAI's Sébastien Bubeck estimated the computational cost at several million dollars, per Quanta Magazine. OpenAI says it will not claim the Clay Institute's $1 million prize — under Clay's rules, official recognition needs publication, a two-year wait and broad acceptance, a bar only the Poincaré Conjecture has cleared. About 12 hours before OpenAI's announcement, NYU's Tristan Buckmaster and Anthropic's Levent Alpöge published a related, Lean-verified proof for the forced Euler equations, using a mix of AI models; OpenAI has ceded priority to them on that result while claiming the larger Navier–Stokes result itself. Both efforts built on a technique developed by Diego Córdoba and Luis Martínez-Zoroa without AI assistance, starting with Martínez-Zoroa's 2021 doctoral work.

The Clay Institute's own review is a multi-year process that has not begun. Watch for independent mathematicians checking whether the Lean-formalised theorem actually matches what mathematicians meant to prove — a separate step from Lean's own logical check — and whether anyone finds a genuinely new idea for the unforced case the 17 September proof says this method cannot reach.