OpenAI takes the Navier-Stokes medal after it learned that two mathematicians were ahead
On September 8 OpenAI announced that it had solved Navier-Stokes, one of the seven Millennium Prize problems. Carlos Santana called it official and historic. According to Simon Willison, the result came from about 10,000 GPT-6 Astra agents over 88 hours and 300 billion output tokens. At the public API price, that is about 15 million dollars.
Tristan Buckmaster and Levent Alpöge had spent a year on the same problem with private Codex sessions paid from their own pocket. On September 3 Buckmaster wrote to OpenAI to avoid a race. They offered him free compute and, on Sunday, an urgent video call. Alpöge was not invited because he works at Anthropic.
In the call, OpenAI showed a prompt and said the model had needed little human intervention. Then it admitted that a whole team started to work on the problem in the days after it learned about their progress. Buckmaster asked if the model was trained on his Codex sessions. OpenAI answered that it did not look at user data and did not answer the second question.
OpenAI gave him two options: publish only the Euler part and give up the Navier-Stokes credit, or sign the paper without Alpöge and credit the model. He refused both. Javier Gamarra sums it up as the third strike for OpenAI after the Musk trial and the military contracts. Javi and Pablo agree: even if they copied nothing, spending millions to arrive before two people is not fair play. Watch in the video.

All the news we covered