The $1M Navier-Stokes Puzzle Just Fell to OpenAI's Agents
All blog articles
Water can break. That three-word answer took mathematicians 90 years, and OpenAI's machines about four days.
On September 8, the company announced that an internal system proved the Navier-Stokes equations can develop a singularity in finite time. In plain terms: a perfectly smooth fluid can, under the right push, reach infinite velocity.
The indiscriminate use of AI is turning the subject into a meaningless production quota 'game'.
10,000 AI agents took on Navier-Stokes for 88 hours
OpenAI set roughly 10,000 autonomous agents loose on the problem, powered by an unreleased model it calls stronger than GPT-6 Astra. After 88 hours they had a proof, and 17 more hours yielded a formal verification in Lean.
That Lean check is the real headline, because a machine-verified proof leaves almost no room for hand-waving. The warm-up was telling too: about 100 agents spent 50 hours on a disproof of Euler regularity first.
What the Navier-Stokes proof really shows
These equations describe how water, air and blood move. Engineers rely on them for aircraft design and weather forecasts, so their limits matter far beyond math departments.
The system built a vortex that spirals inward and stretches like spaghetti until velocity blows up. One caveat: the fluid gets pushed by a smooth external force, so the unforced version of the puzzle stays open.
A $1 million prize, a rumor and a credit dispute
Only one of the seven Millennium Problems had fallen before, the Poincaré conjecture, back in 2003. OpenAI says it won't claim the Clay Mathematics Institute's check, despite a compute bill in the millions of dollars.
A race against human mathematicians?
The run started September 1, right after rumors that Levent Alpöge and Tristan Buckmaster neared their own solution. Fields medalist Terence Tao fears a frantic race, while OpenAI insists it never saw their unpublished work.
Our take: the result looks historic, the timing looks rushed, and both can be true. After the AGI claims around GPT-6 Astra, OpenAI clearly wants September 2026 stamped with its name.
Did OpenAI win the $1 million Navier-Stokes prize?
No. OpenAI stated it does not intend to claim the Clay Mathematics Institute's $1 million reward, and the prize process requires publication plus a lengthy community review anyway.
Is the Navier-Stokes problem now fully solved?
The official Millennium Prize formulation is resolved, assuming the proof survives scrutiny. The existence and smoothness of unforced fluids, with no external push, remains an open question.