OpenAI Solves Navier–Stokes with Secret Model & Agent Swarm

On September 8, OpenAI published a 166-page proof along with a Lean formalization demonstrating that three-dimensional incompressible Navier–Stokes equations can develop a singularity in finite time. Starting from rest under a smooth external force, a smooth fluid's speed grows without bound while total energy remains finite, marking a negative resolution to Clay Mathematics Institute's Millennium Prize problem ninety years after Leray.

The proof originated not from GPT-6 Astra—released five days earlier—but from an internal model in training since August 28, which OpenAI describes as significantly more capable with unprecedented benchmark performance, particularly in mathematics. Astra's sole role was executing the final Lean formalization and verification step within 17 hours.

Stock photo for illustration only, not from the actual event

Stock photo for illustration only, not from the actual event

The agent swarm architecture running atop the model represents a breakthrough for system builders, featuring three key design choices:

• Dividing the swarm into groups and assigning different statements from the official problem rather than prompting them to solve Navier–Stokes directly.
• Tackling an easier sub-problem first—the unforced Euler equations—solved by roughly 100 agents in 50 hours, which was then fed into the Navier–Stokes groups.
• Restricting direct group communication while employing Codex to consolidate insights into follow-up prompts.

The mathematical theorem establishes that for every positive viscosity, a smooth external force creates a fluid whose maximum speed approaches infinity as time nears 1, leaving no smooth finite-energy solution for all time. Despite this monumental milestone, OpenAI has confirmed it will not claim the associated one-million-dollar prize.

Source: Dev.to