OpenAI says unreleased model solved Navier-Stokes in 88 hours
OpenAI says an internal model more capable than Astra ran about 10,000 agents for 88 hours to produce a singularity solution to the forced Navier-Stokes equations, though the Clay Institute still lists the problem as unsolved.
SITUATION EXPLAINED: OpenAI says an unreleased model solved the Navier-Stokes Millennium Prize Problem in 88 hours. • The result: an initially smooth fluid at rest, with a smooth external force and finite energy throughout, does develop a singularity in finite time. The solution is a vortex spiraling inward and elongating like spaghetti • OpenAI has been training the internal model since August 28, less than two weeks, and calls it significantly more capable than Astra. Training is still ongoing • On September 1 they heard rumors two Millennium problems had been resolved elsewhere, and turned the model loose on all of them using coordinating agent swarms • Roughly 10,000 concurrent agents on Navier-Stokes, reaching the result 88 hours later. Euler fell first, about 100 agents over 50 hours • 2.7 million messages and 130 billion output tokens on Navier-Stokes alone. Lean formalization took another 17 hours via Astra • Noam Brown says it cost millions of dollars to run • The Clay Institute still lists the problem as unsolved, and its criteria are based on the unforced equations while this result addresses the forced version @theojaffee: "It took seven years to solve the last Millennium Prize problem. It took 88 hours to solve this one. 88 hours is like three and a half days. It's not even four days. Crazy."
This post quotes another post: “OpenAI says it solved the Navier-Stokes Millennium Prize problem”