OpenAI Says Its AI Agents Solved Prize Math Problem

OpenAI said Tuesday it had solved a problem that mathematicians have been grappling with for more than a hundred years. But the achievement set off accusations that OpenAI had leaned on another researcher’s work to solve the problem.

More than 10,000 OpenAI agents worked together to solve the Navier-Stokes Millenium Prize Problem, which concerns fluid motion and has a million-dollar prize attached to it. OpenAI says it found the solution using a “next-generation model significantly more capable than GPT-6 Astra,” its model released to the public Thursday.

The proof looks to be the latest breakthrough in AI models’ mathematics capabilities.

However, even before OpenAI had announced its discovery, New York University mathematics professor Tristan Buckmaster alleged that OpenAI was spurred to solve the Navier-Stokes problem after hearing rumors of his work and speculation that rival Anthropic had solved the problem.

OpenAI CEO Sam Altman addressed those allegations Tuesday, saying in a post on X that OpenAI’s proof approach seemed different from Buckmaster’s. But Altman confirmed that OpenAI had pursued the problem due to rumors that Anthropic had solved it.

OpenAI denied that researchers or agents at the company accessed any specific user data in solving the problem. But the company said OpenAI could not rule out that its models had learned from anonymized prompts Buckmaster and Levent had sent the models.