← Back to feed

The internal multi-agent system that attacked Navier–Stokes

OpenAI describes ~10,000 concurrent agents, hundreds of billions of tokens, and Lean formalization after finding a singularity with smooth forcing.

The internal multi-agent system that attacked Navier–Stokes

Per OpenAI, from August 28, 2026 it was training an internal model beyond GPT-6 Astra. On September 1 it launched agent groups with tools (web cache, code) on Clay variants A–D.

A group on the order of 10,000 concurrent agents, after ~88 hours with Codex cross-pollination, produced on September 5 an analytic singularity proof (statements C/D) with smooth external force and finite energy; Lean + GPT-6 Astra added ~17 hours of formalization.

OpenAI also reported a prior unforced-Euler result. Global figures: ~2.7M messages and ~130B output tokens on the Navier–Stokes branch alone. These are lab claims; external verification continues.