New OpenAI Model Solves Navier-Stokes Millennium Prize
An unnamed OpenAI model has solved the Navier-Stokes Millennium Prize problem after just eight days of training, signaling a massive leap in automated scientific discovery.
OpenAI deployed a massive agent swarm to tackle the unsolved Millennium Problems after hearing rumors of a breakthrough. To solve the Navier-Stokes problem, the system ran for 88 hours, sending 2.7 million messages and consuming approximately 130 billion output tokens. The company's Astra model then spent 17 hours performing Lean formalization and verification. Across all attempted problems, the agents sent 4.9 million messages and used 300 billion output tokens, an effort that would cost a commercial customer roughly $22 million.
The unnamed model represents a significant leap over Astra, showing a step change in capabilities just four days into its training run, which began on August 28. This rapid progress followed a brief pause in frontier reinforcement learning from June 22 to August 28. The breakthrough has also accelerated OpenAI's internal research automation. By September, the company met its goal of deploying an automated research intern, with its median human researcher using coding agents daily at a cost of $600 in inference, while top-percentile researchers consume over $7,000 daily.
The achievement has sparked intense controversy within the academic community. Mathematicians Tristan Buckmaster and Levent Alpoge had been working on similar fluid dynamics proofs using large language models. After learning of their work, OpenAI allegedly pressured Buckmaster to publish the Navier-Stokes proof without Alpoge, who works at rival lab Anthropic. Although some suspected OpenAI trained its model on private Codex user logs, OpenAI and Anthropic researchers stated the probability of such data leakage was virtually zero.
The incident has united academic researchers against the rapid pace of corporate AI development. A group of 25 Fields Medal winners, including Terence Tao, signed a joint declaration warning that the rush to solve major mathematical benchmarks is "detrimental to the science of mathematics," as noted by the signees. They argued that automated, unreadable proofs announced in a rush undermine conceptual understanding and create severe attribution issues.
This is our own summary of reporting by Don't Worry About the Vase
