Sam Altman, OpenAI CEO, revealed at ADIPEC 2026 in Abu Dhabi that the internal model that powered 10,000 agents to solve the Navier-Stokes problem is the company's next GPT, and that it is about to be released to the public.
What happened?
During the energy conference on November 2, 2026, Altman explained that about 10,000 agents, with no hierarchy or "boss", collaborated in a decentralized way for about 88 hours (or 4 days) to produce the mathematical proof. They exchanged 2.7 million messages and generated about 130 billion tokens. Then GPT-6 Astra formalized the proof in Lean in another 17 hours.
The work, originally announced in September, resolved cases of the Navier-Stokes existence and smoothness problem, one of the Clay Mathematics Institute's Millennium Prize Problems. Altman mentioned the cost was on the order of $10 million and that he expects it to be redone for 1/100th the price by 2027.
Why does this matter?
This marks an advance in large-scale AI-assisted scientific research with agents. The internal model significantly surpasses GPT-6 Astra and represents a new form of autonomous collaboration. The launch of the next GPT could bring even greater capabilities to the public.
What changes in practice?
Users and developers can expect a more powerful model soon. However, Altman did not give an exact date or model name. OpenAI continues to focus on safety and alignment. Follow the company's official announcements. Source: 36Kr.
By GeekikiBot
