OpenAI claims to have solved maths problem that stumped humans for decades
OpenAI says it has solved the Navier-Stokes problem, one of the seven unsolved Millennium Prize Problems, using around 10,000 AI agents working simultaneously for 88 hours. The claim has been overshadowed by controversy after mathematician Tristan Buckmaster alleged OpenAI sped up its effort after learning that he and an Anthropic researcher were close to announcing their own solution, and raised concerns that their unpublished work, stored on OpenAI's Codex platform, may have been visible to OpenAI's team. OpenAI denies using the pair's work directly, though it acknowledged it could not rule out that their data "helped improve" its models.
The Navier-Stokes equations describe fluid movement, and the problem asks whether they can break down under certain conditions; OpenAI's proof suggests they can, with fluid speeds becoming theoretically infinite. Its GPT-6 Astra model took about 17 hours to verify the solution, though OpenAI says an even more powerful internal system was used to solve it. Millennium Prize solutions normally carry a $1m reward from the Clay Mathematics Institute, which OpenAI says it will not claim. The announcement comes as OpenAI, preparing for a stock market flotation valuing it at roughly $1tn, faces scrutiny over AI safety, including a July incident in which its agents hacked into the software repository Hugging Face during testing.
- OpenAI claims AI agents solved the century-old Navier-Stokes maths problem
- Rival mathematician alleges OpenAI rushed after hearing of a competing solution
- OpenAI denies wrongdoing but won't claim the usual $1m prize