OpenAI Claims Navier-Stokes Solution Amid Dispute Over Academic Credit

OpenAI announced that an internal model "significantly more capable than GPT-6 Astra" solved the Navier-Stokes Millennium Prize problem. [3] The company stated that approximately 10,000 concurrent agents worked for about 88 hours to produce the proof, costing millions of dollars in computing power. [3] NYU mathematics professor Tristan Buckmaster, who announced his own related proofs on Tuesday alongside Anthropic mathematician Levent Alpöge, claims that OpenAI learned of their progress and subsequently raced to achieve a formal proof first. [2] Buckmaster also questioned whether his private material stored in Codex played a role in OpenAI's achievement. [1][3]

Sébastien Bubeck, who leads OpenAI's mathematical research, stated that the allegations against him are "false and inflammatory." [1][2] OpenAI chief research officer Mark Chen said that neither humans nor AI systems searched user data to solve the Navier-Stokes problem. [3] The company said its Navier-Stokes effort began on September 1 after researchers heard rumors regarding other Millennium Prize problems. [3]

This dispute highlights a core trust question regarding whether scientists can safely use tools from frontier labs to develop unpublished discoveries. The controversy pits a traditional academic collaboration—which utilized both Codex and Claude AI models—against OpenAI's use of 10,000 concurrent agents to accelerate a proof. The confrontation underscores the immense material advantage that concentrated computing resources can provide in resolving complex mathematical problems once a viable approach is suspected. What the sources don't address: Whether independent mathematicians have verified the accuracy of the 88-hour AI-generated proof.