Wire flash
ScienceOpenAI deploys 10,000 AI agents to solve 90-year-old math problem
Editorial responsibility
- No named human review is recorded for this page.
- Source reporting is collected, normalized, translated or condensed automatically when needed.
- Automatically published source-backed update
A post on X claims that OpenAI deployed 10,000 AI agents, powered by a model more capable than GPT-6 Astra, to solve a long-standing mathematical problem. The problem, which mathematicians had been unable to resolve for approximately 90 years, concerns whether smooth 3D fluid flow can suddenly break down. OpenAI's proposed proof asserts that it can. The agents were divided into groups exploring different approaches, with the best findings shared among them. The search took 88 hours and utilized approximately 130 billion output tokens across 2.7 million messages. GPT-6 Astra then spent another 17 hours verifying and converting the proof into Lean, a system for verifying mathematical proofs. The Lean version is publicly available for mathematicians to inspect. The post suggests that scaling agent count, rather than model size, may be the key to solving hard scientific problems.
Source report
OpenAI has achieved a major breakthrough: a system of 10,000 AI agents, powered by a model significantly more capable than GPT-6 Astra, has solved a math problem that had remained unresolved for approximately 90 years.
A New Scaling Paradigm
This development suggests that in this new era, the scaling law may be defined by agent count rather than model size. With enough capable agents, the system can explore far more scientific ideas than any human group ever could.
When tens of thousands of strong agents can explore, critique, and combine ideas, many problems currently considered "too hard" may become primarily a matter of computational resources.
The Problem Solved
For 90 years, mathematicians have been unable to determine whether smooth 3D fluid flow can suddenly break down. OpenAI's proposed proof asserts that it can.
How the Solution Was Found
- OpenAI divided thousands of agents into groups, each exploring different approaches.
- The best findings were then shared among all groups.
- The search took 88 hours and utilized approximately 130 billion output tokens across 2.7 million messages.
Verification and Publication
After the initial search, GPT-6 Astra spent an additional 17 hours verifying and converting the proof into Lean, a system used to verify mathematical proofs.
The Lean version of the proof is now publicly available, allowing mathematicians to inspect the argument themselves.
Source
rohanpaul_aiNeutral / independent
Part of this Story
OpenAI AI agents solve 90-year-old Navier-Stokes Millennium Problem in 88 hours