OpenAI has released what it describes as an AI-generated solution to the Navier-Stokes existence and smoothness problem, one of mathematics’ seven Millennium Prize Problems. The development could mark an important moment in the use of artificial intelligence for advanced mathematical research.
1. This is one of mathematics’ hardest problems
The Clay Mathematics Institute selected seven Millennium Prize Problems in 2000 and allocated $1 million to the solution of each. Until now, only the Poincaré Conjecture had been resolved. The Navier-Stokes problem concerns equations governing the motion of fluids such as air and water.
2. The breakthrough is about mathematical ‘blow-up’
Navier-Stokes equations are widely used to describe fluid motion, but mathematicians have long sought to determine whether suitably smooth three-dimensional solutions must always remain smooth. The new work argues that initially smooth fluid dynamics can develop a singularity, or ‘blow up’, in finite time.
3. AI agents did more than check a human proof
What makes the development particularly significant is the role assigned to AI. OpenAI says an internal model produced the analytical proof rather than merely checking an already completed human argument. The company also produced a formalised version using Lean, a theorem prover that can mechanically verify mathematical reasoning.
4. The breakthrough has triggered a credit dispute
NYU mathematician Tristan Buckmaster and Anthropic mathematician Levent Alpöge had independently been working on a closely related approach with assistance from publicly available AI models. Buckmaster subsequently questioned whether their research influenced OpenAI’s effort. OpenAI has denied that its researchers or agents accessed their private transcripts. The extent of any intellectual influence remains disputed.
5. Mathematics could face a new compute divide
The episode raises a question extending far beyond Navier-Stokes. If frontier AI systems can explore enormous numbers of mathematical approaches simultaneously, access to advanced models and computing resources could become a significant research advantage. OpenAI has said its effort involved running about 10,000 agents concurrently and cost millions of dollars.
Human mathematicians still contribute something difficult to measure in computing power: choosing worthwhile problems, identifying promising approaches, interpreting results and developing ideas that can influence areas far beyond an individual proof.
The Navier-Stokes development therefore raises a bigger question than whether AI can solve extraordinarily difficult mathematics. The emerging frontier is how discovery itself changes when human researchers, powerful AI agents and enormous computing resources begin tackling the same unsolved problems.