OpenAI posts Lean proof it says solves Navier–Stokes

OpenAI today posted an analytical writeup and a Lean formalization that it says solves the Navier–Stokes existence and smoothness Clay Millennium problem by establishing statements C and D — a finite-time singularity under smooth forcing. The company says it is not claiming the $1 million Millennium Prize.

What the tech did: OpenAI says an unreleased internal model “significantly more capable” than GPT-6 Astra powered a multi-agent search. Roughly 10,000 concurrent agents worked the Navier–Stokes variants; agents reached a resolution on Saturday, September 5, about 88 hours after the effort launched, with Lean formalization and verification taking about 17 more hours via GPT-6 Astra. The posted result describes an initially smooth fluid at rest that develops unbounded speed in finite time while energy stays finite, under a smooth external force.

What it is not: This is not an independent Clay award and not a consumer model ship. Clay Institute president Martin Bridson told New Scientist the evaluation process is deliberately unhurried. Outside verification of the Lean certificate is still ahead. OpenAI’s post recognizes Tristan Buckmaster and Levent Alpöge priority on forced Euler and says the proofs differ; we covered that dispute earlier; this story is about the lab’s public claim on statements C and D.

Who it hits: Frontier labs racing Millennium-adjacent math; anyone who treats “model solved X” posts as provenance-free; mathematicians waiting on readable arguments, not only certificates.

What’s next: Outside Lean checks; whether Clay treats C/D as resolving the Prize statement; any fuller OpenAI reply on training/session access after Buckmaster’s allegations; whether the unreleased model’s math claims migrate into a named public system.

— Lark Pemberton