OpenAI says an internal AI system, more advanced than its public GPT-6 Astra model, produced a solution to the Navier-Stokes Millennium Prize Problem — one of the most famous unsolved questions in mathematics, with a $1 million prize attached to it since 2000.
According to OpenAI, around 10,000 AI agents worked in coordination for 88 hours before producing an analytical proof showing that a three-dimensional fluid governed by the Navier-Stokes equations can develop a singularity — a point where the math breaks down — in finite time. The company also released a formalized version of the proof in Lean, a programming language mathematicians use to verify proofs line by line.
But the announcement quickly ran into controversy. NYU mathematician Tristan Buckmaster and an Anthropic researcher had been independently working on related research, and Buckmaster has raised questions about whether OpenAI’s effort was influenced by seeing their progress. OpenAI denies any improper access. Independent mathematicians still need to verify whether the proof actually holds.
Why it matters: Regardless of how the credit dispute plays out, the bigger shift is hard to ignore — AI systems are moving from helping mathematicians with calculations to coordinating thousands of agents to attack unsolved problems directly. If verified, this would be one of the strongest examples yet of AI making an original contribution at the edge of pure mathematics, not just summarizing or assisting with existing work.