OpenAI says an internal AI system has produced a Lean-formalized solution related to the Navier–Stokes equations, a central unsolved problem in fluid dynamics. The company assigned more than 1,000 agents to the task for over 50 hours and said the computing cost ran into millions of dollars.
The mathematical claim is now entangled in a priority dispute. NYU mathematician Tristan Buckmaster and Anthropic researcher Levent Alpöge had separately made progress in a relevant area using models including Claude and Codex. Buckmaster says OpenAI increased its effort after learning that the pair was close and later discussed proposals about how the work might be announced.
OpenAI denies inspecting their Codex prompts or proof before the pair released it publicly. The company says its solution differs from their work and has acknowledged Buckmaster and Alpöge’s priority on unforced Euler, an associated result. The researchers did not immediately respond to WIRED’s request for comment.
A formalization in Lean makes a proof machine-checkable, but it does not settle questions of originality, scope or academic credit. The report is developing, and both the mathematical result and the competing accounts will require further scrutiny.