OpenAI Claims New AI Solved Over 100 Open Math Problems

OpenAI has announced that its latest internal AI model, which began training on August 28, has successfully resolved more than 100 long-standing open problems in mathematics. The company, which shared the milestone on September 21, noted that this same system was responsible for its recently publicized solution to the Navier-Stokes Millennium Prize problem. To achieve the Navier-Stokes breakthrough, the system deployed approximately 10,000 concurrent agents equipped with code execution capabilities and cached web data, generating a proposed proof in about 88 hours and completing 17 hours of formalization in the Lean theorem-proving language.

Because OpenAI has not yet made the majority of these 100-plus solutions public, independent researchers cannot yet evaluate their validity or significance. To manage the impending release of these findings, OpenAI is collaborating with the Advisory Group on Mathematics and Artificial Intelligence. This independent, unpaid panel is hosted at the Institute for Advanced Study and is tasked with advising the company on how to coordinate the release of this massive influx of mathematical results.

The rapid generation of proofs has sparked debate within the scientific community. On September 11, a group of prominent mathematicians, including Fields Medal recipients, published a declaration titled "A Severe Misalignment of AI in Mathematics." They argued that using famous open problems as benchmarks is insufficient, emphasizing that true mathematical progress requires understanding why a proof works so its concepts can be applied elsewhere. If AI continues to output solutions at this pace, the primary bottleneck for mathematical research will shift from discovering new proofs to verifying, translating, and absorbing them.