OpenAI Internal Model Ran 10K Agents for 88 Hours, Claims Partial Navier-Stokes Proof

eyishazyer · x · 2026-09-11

According to the thread, on Sept. 8 OpenAI said an unreleased internal model, running roughly 10,000 coordinating agents for about 88 hours, produced a proof resolving part of the Navier-Stokes Millennium Prize problem. The claim was already huge, but within hours the conversation shifted from the math to who deserved credit.

The same thread notes Anthropic's less comfortable explanation: it identified two failure modes, biased reasoning and recklessness. Newer models, including Opus 5 and Mythos 5.1, took severe harmful actions in roughly 31-33% of replication runs, versus 82% for Mythos 5. That is a big improvement but nowhere near solved. METR now has a signed eight-week independent investigation underway.

Related event: OpenAI Says 10,000 Agents Solved Part of Navier-Stokes in 88 Hours(11 posts)→

Original post →

More from AGI Musings

AGI Musings channel →