Key Takeaways :
- 10,000 AI agents worked together to crack part of a 90-year-old math problem in 88 hours.
- The effort resolved two of four required proof statements, at a cost of $10 million.
- Independent mathematicians have not yet verified the result.
OpenAI says a swarm of 10,000 AI agents solved part of a 90-year-old math problem in 88 hours, marking a major leap for artificial intelligence research.
A decades-old problem meets modern AI
The Navier-Stokes equations describe how fluids move, from ocean currents to airflow over a wing. Mathematicians have studied the Navier-Stokes equations for 90 years, but key parts remain unproven. The unsolved pieces sit at the center of turbulence, a phenomenon scientists still don’t fully understand.
OpenAI built a new internal AI model in late August. The model showed unusual skill with mathematics early on. Researchers then decided to test it against some of the toughest known math challenges.
Word spread on September 1 that two Millennium Prize problems might have been solved elsewhere. OpenAI responded by assigning 10,000 AI agents to work on the remaining problems. By September 5, roughly 88 hours later, the agents had produced a solution addressing part of the Navier-Stokes puzzle.
Massive computing power behind the breakthrough
The scale of the effort was enormous. The AI agents exchanged nearly 3 million messages and generated 130 billion output tokens, the units of text and code an AI model produces. OpenAI estimates the computing effort would cost around $10 million at standard pricing.
The solution resolved two of the four statements required for the full Millennium Prize proof of the Navier-Stokes equations. That prize, offered by the Clay Mathematics Institute, comes with a $1 million reward for a complete solution. OpenAI says it does not plan to claim the prize with this result.
“Our goal in releasing this result is to report on the substantial progress of our AI models,” the company said. Independent mathematicians have not yet verified the proof, and the Clay Mathematics Institute has not commented publicly on it.
Other researchers were working on the same problem
OpenAI wasn’t alone in tackling this problem. A New York University mathematics professor and a researcher at a rival AI company were separately working toward a similar solution using different tools.
OpenAI said it had not seen the other researchers’ work before publishing its own results and confirmed no user data was used in its process. The company noted its proof differs from the other effort, and the specific results reached are not the same.
The episode highlights how fast AI systems are advancing in complex reasoning tasks that once took human experts years to attempt. Whether the community fully accepts OpenAI’s proof will depend on independent review in the months ahead.
For now, the new internal model stays a research tool inside OpenAI rather than a public product. The company describes it as notably more capable than its previous publicly released systems, suggesting further advances may follow as researchers test it on other unsolved problems, the BBC reports.
