OpenAI said to use 10,000-agent swarm on Navier-Stokes problem
A post claims OpenAI ran about 10,000 agents on a model beyond GPT-6 Astra, exchanging 2.7 million messages over four days to argue that fluids can form a singularity in finite time, which Astra then translated into Lean over 17 hours.
A MATH PROBLEM SURVIVED 90 YEARS OF HUMAN GENIUS. IT DID NOT SURVIVE 88 HOURS OF 10,000 AI AGENTS. Since the 1930s nobody could say whether the Navier-Stokes equations - the math behind every flowing liquid on Earth - can break down and produce infinities. -> OpenAI says the question is now closed. They pointed a swarm of roughly 10,000 agents at it, running on an internal model beyond GPT-6 Astra. The swarm exchanged 2.7 million messages and wrote about 130 billion tokens - a century of correspondence between mathematicians, compressed into four days. The result is brutal in its simplicity: a fluid can start perfectly still and, under the right conditions, spin up a singularity. Velocity goes infinite in finite time. Astra then spent 17 more hours translating the argument into Lean, where a machine checked every single step. The run cost millions. They are not claiming the $1M prize, and human mathematicians still get the final word. The headline is not the proof. The headline is the org chart: ten thousand agents just worked as one research lab. Decades of human work now fit into a long weekend ↓
This post quotes another post: x.com/slash1sol/status/2098395296845729871