What's this about?
OpenAI says an unreleased model stronger than GPT-6 Astra coordinated roughly 10,000 agents to construct a finite-time singularity for three-dimensional Navier–Stokes with smooth forcing. The effort took 88 hours, about 130 billion output tokens, and another 17 hours for GPT-6 Astra to formalize and verify it in Lean.
What exactly is the Navier–Stokes problem?
Navier–Stokes equations describe fluid flow. The open question asks whether initially smooth three-dimensional flow can become mathematically infinite — a singularity — in finite time. OpenAI addresses the forced versions C and D in Clay's official formulation, where a smooth external force acts on the fluid.
What's the community saying?
HN 1,138 points and 981 comments; OpenAI's post had about 89,000 likes.
- Awe centred on the scale: 10,000 agents, 4.9 million messages across the broader run and 300 billion total output tokens.
- Mathematicians stressed that a Lean certificate is strong evidence but not a substitute for expert review of definitions, assumptions and significance.
- A technical dispute questioned whether the smoothly forced construction captures what people informally imagine as the “real” unforced problem, even though it is explicitly in Clay's statement.
- The academic-credit and data-use controversy dominated as much attention as the proof itself.
- Commenters asked whether this was creative “Move 37” reasoning or brute-force search by an extremely capable model and enormous compute budget.