OpenAI says 10,000 AI agents solved a $1 million math problem. Now mathematicians are fighting

OpenAI announced that a swarm of 10,000 AI agents generated a proposed solution to one of the seven Millennium Prize Problems, the series of mathematical questions that each carries a $1 million reward. The company says the result was produced by an internal model it considers more capable than GPT-6 Astra. The claim frames the output as a proposed solution rather than an established proof.

The Millennium Prize Problems are among the most famous and difficult open questions in mathematics, and a validated solution to any one of them would be a major event. OpenAI’s statement makes clear that this work came from an in-house model operating at a level beyond publicly described versions of GPT-6 Astra, but details about the methods, the exact problem addressed, and the solution itself have not been presented in the material provided.

Following the announcement, questions have arisen about how independently the system arrived at the proposed solution. Members of the mathematical community are reported to be in dispute over the claim — debating issues around the AI’s role in deriving the result, the transparency of the process, and what constitutes a valid, verifiable proof when advanced models are involved.

Those debates underscore the next likely steps: independent scrutiny and verification by experts in the relevant mathematical field, and closer examination of how the agents collaborated and what prior knowledge or human guidance may have influenced the outcome. How those processes unfold will affect both the reception of the claimed solution and broader discussions about AI’s capabilities and responsibilities in research.