OpenAI says 10,000 agents ran 88 hours to attack Navier-Stokes
About 10,000 agents ran concurrently for 88 hours and used 130 billion output tokens to produce a proof, which GPT-6 Astra then spent 17 hours formalizing and verifying in Lean.
10,000 OpenAI agents. 88 hours. A claimed solution to a $1 million math problem. For roughly 90 years, mathematicians have asked whether smooth fluid motion can break down under the Navier–Stokes equations. OpenAI says its internal AI system found a proof. The setup is the part worth paying attention to: → around 10,000 agents working concurrently → an internal model more capable than GPT-6 Astra → 2.7 million messages exchanged → roughly 130 billion output tokens Different groups explored different approaches. Useful findings were consolidated through Codex and fed back into the research. After 88 hours, the agents produced their proposed solution. GPT-6 Astra then spent another 17 hours formalizing and verifying it in Lean. The result concerns an initially smooth fluid at rest, with a smooth external force applied. Its velocity can grow without bound in finite time, while its energy stays finite. OpenAI published the paper and formalization, and says it won’t claim the prize. A published proof still needs independent mathematical scrutiny. But the research setup alone is striking. One prompt is a conversation. Thousands of coordinated agents, sharing intermediate results and checking the mathematics, start to look like a research lab.
This post quotes another post: x.com/ggwplabs/status/2106004888312217829