XNews

OpenAI Solves Navier-Stokes Millennium Problem with 10,000 AI Agents

OpenAI Solves Navier-Stokes Millennium Problem with 10,000 AI Agents

OpenAI announced on September 8, 2026, that 10,000 autonomous agents running on an internal model solved the Navier-Stokes existence and smoothness problem, a Millennium Prize Problem, in 88 hours, with formal verification in Lean completed in 17 additional hours. The solution involved nearly 5 million inter-agent messages and an estimated compute cost of several million dollars. The announcement followed a dispute with mathematicians Buckmaster and Alpöge, who had independently achieved related results on the Euler equations.

Key insights

OpenAI announced that its internal model, using 10,000 coordinated AI agents, found a singularity in the 3D Navier-Stokes equations, solving one of the seven Millennium Prize Problems. The agents sent nearly 5 million messages, and the proof was formally verified in Lean. The compute cost was estimated at several million dollars. The result was achieved in 88 hours, with an additional 17 hours for formalization.

Mathematicians Buckmaster and Alpöge had independently proven finite-time blowup for the Euler equations with smooth forcing, and they believed they had a Navier-Stokes result, but their formal verification was incomplete. OpenAI contacted them, proposing a joint release, but disputes arose over authorship and priority. OpenAI reportedly pressured to remove Alpöge from the paper due to his employment at Anthropic, leading to a conflict over credit and collaboration.

Terence Tao praised the Buckmaster-Alpöge work as a remarkable achievement and noted that their methods had no principled obstacle to extending to Navier-Stokes. He expressed skepticism about brute-force AI approaches, saying he would not be surprised if someone used massive compute and AI to push through, but that such an approach did not particularly appeal to him. He emphasized the importance of digesting the proof methods for new mathematical insights.

The mathematical community has not yet independently verified OpenAI's Lean formalization, though Lean verification provides high logical certainty. The key human role remains ensuring that the statement proven in Lean is logically equivalent to the original mathematical problem. This highlights the boundary of machine certainty and the necessity of human oversight in defining the right questions.

Source posts