Ten thousand AI agents just solved one of the hardest math problems in history
OpenAI's fleet of 10,000 AI agents solved a 90-year-old math problem in 88 hours, proving that AI can now do frontier research on its own.
- 10,000
- AI agents deployed
- 88 hours
- time to solve the proof
- ~$15M
- estimated compute cost
- $1M
- prize offered since 2000
Last week, OpenAI sent 10,000 AI agents to work on a math problem that had stumped every human mathematician for 90 years. They worked 88 straight hours and came back with an answer. The estimated compute cost was around $15 million. AI can now do frontier science, not just assist with it.
The problem is called Navier-Stokes. The equations describe how fluids move — think ocean currents or the air around a plane wing. For 90 years, mathematicians could not answer one question: can those equations ever produce runaway, impossible results? OpenAI's agents showed they can. The proof was verified step by step by a computer. The Clay Mathematics Institute has offered $1 million for this answer since 2000. OpenAI says it will not claim the prize.
The scale is unlike anything before. Those 10,000 agents worked in parallel, exchanging millions of messages with each other and processing hundreds of billions of tokens — the small chunks of text that AI models think in. This was not a calculator running faster. It was a coordinated research team made of software, working at a speed no human group could match.
The win came with controversy. Two mathematicians — one at NYU, one at Anthropic — had been quietly working on the same problem for a year. Their August breakthrough apparently reached OpenAI, which then launched a major compute push to solve it first. OpenAI says it never accessed their work, but admitted it 'cannot rule out' that its models learned from traces of it. Knowing a problem is solvable is worth millions in compute. That is a new and uncomfortable dynamic in research.
The practical takeaway is not that AI will replace scientists this year. It is that the gap between AI helper and AI research team is closing fast. If your business solves hard, open-ended problems — in law, medicine, materials, or software — now is a good time to ask what a fleet of software agents could do for you.