OpenAI Agents Produce Navier-Stokes Proof in 88 Hours Using 10,000-Agent Push
Updated
Updated · KDnuggets · Sep 30
OpenAI Agents Produce Navier-Stokes Proof in 88 Hours Using 10,000-Agent Push
3 articles · Updated · KDnuggets · Sep 30
Summary
OpenAI said its internal AI system produced a proposed Navier-Stokes solution in 88 hours, then spent another 17 hours formalizing and verifying the proof in Lean.
Roughly 10,000 concurrent agents exchanged 2.7 million messages and generated about 130 billion output tokens after OpenAI redirected compute from other Millennium Prize problems toward Navier-Stokes.
The result addresses one Clay-approved route to the problem by constructing a finite-time singularity for 3D Navier-Stokes with smooth external force, not the better-known unforced smoothness question.
Tristan Buckmaster later questioned whether unpublished work he and Levent Alpöge had entered into Codex could have influenced the model; OpenAI said prompts from the prior two months could not have affected it.
The Clay Mathematics Institute said the problem has apparently been settled, but review and credit attribution remain open as the case tests how AI-generated mathematical discoveries should be recognized.
Did OpenAI’s 10,000-agent system truly solve a Clay-approved Navier–Stokes route, or just formalize a result mathematicians still need to validate?
How much of OpenAI’s Navier–Stokes breakthrough was new machine discovery versus an AI-powered extension of Buckmaster and Alpöge’s earlier blowup program?
If Lean verified the proof, why are mathematicians still cautious—and what does that reveal about truth, understanding, and credit in AI-made math?