Altman says OpenAI's Navier-Stokes result came from 10,000 agents running its next GPT, due out soon
Speaking at ADIPEC in Abu Dhabi, he said the agents worked for about four days with no hierarchy and that AI is now mostly smarter than top scientists; he gave no release date, and the claims were not independently verified.
Why it matters
OpenAI had described the September run as using an internal model; Altman's remark ties it to the next public GPT release, which would put the system behind a claimed Millennium Prize result in customers' hands. On a podcast last week he put the run's cost at about $10 million and said the same work could cost a hundredth of that within a year. Watch for the model's release, its system card and whether mathematicians accept the Navier-Stokes construction, which OpenAI has published as Lean files.
Key facts
- Remarks at the ADIPEC opening in Abu Dhabi on November 2
- About 10,000 agents ran for about four days
- Run launched September 1 and took about 88 hours, per R&D World
- Cost put at "order $10 million" on an October 29 podcast
- No release date given for the next GPT