On September 8, 2026, OpenAI announced that an unreleased internal AI model generated a proof regarding the Navier-Stokes equations, a super-hard problem in mathematics. According to Mark Chen, Chief Research Officer (CRO) of OpenAI, the unreleased model established the claim regarding the equations by running approximately 10,000 agents in parallel for 88 hours, followed by 17 hours of formal verification. The computational cost is reported to have reached several million dollars.

In response to this announcement, Tristan Buckmaster, a mathematician at New York University, issued a statement questioning OpenAI's approach. Buckmaster has independently published a related paper in collaboration with Levent Alpaygue, a researcher at Anthropic, and discrepancies have been pointed out between the two parties regarding the originality of the achievements and the publication process. OpenAI stated that while they did not directly access the data of specific users, they cannot completely rule out the possibility that anonymized usage data influenced the model's training.

Source: