OpenAI announced a major breakthrough, claiming that about 10,000 AI Agents working together with artificial intelligence models produced a solution to the Navier–Stokes problem—a challenging mathematical question unsolved for nearly 90 years—in just 88 hours. However, the proof still needs validation by independent mathematicians.
OpenAI, the developer of ChatGPT, stated it solved this nearly 90-year-old advanced math problem within only 88 hours by using new AI models internally and mobilizing approximately 10,000 AI Agents—autonomous AI systems—to collaboratively seek a solution.
The problem involves the Navier–Stokes equations, which describe fluid motion such as water and air, and are critical to understanding turbulence phenomena that scientists have yet to fully explain.
OpenAI said it began this effort on 1 September after rumors surfaced that a Millennium Prize math problem might have been solved. The company employed an internally developed AI model with advanced mathematical capabilities beyond the latest public versions to tackle this significant unsolved problem.
By 5 September, about 88 hours after deploying roughly 10,000 AI Agents, OpenAI claimed to have discovered a way to prove the Navier–Stokes problem. The AI exchanged nearly 3 million messages and used computational resources equivalent to about 130 billion tokens of AI-generated text and code.
OpenAI estimated that, based on its highest-performance model rates, this effort cost around $10 million USD, approximately 330 million Thai baht.
However, OpenAI acknowledged that the work has yet to be reviewed by independent mathematicians and is not officially recognized by the Clay Mathematics Institute, the US-based organization that established the Millennium Prize Problems—seven major unsolved math problems each with a $1 million prize for a valid solution.
The company stated its work proved 2 of 4 key statements related to the Millennium Prize problem’s proof conditions and emphasized it does not intend to claim the $1 million prize. The goal is to demonstrate rapid advances in AI’s mathematical abilities.
OpenAI also noted the models used were internal systems more capable than publicly available versions. After the AI Agents found a proof approach, the GPT-6 Astra model took about 17 hours to verify the solution.
Concerns Raised by Mathematicians Regarding Research Data
Despite OpenAI’s announcement, controversy quickly arose after Tristan Buckmaster, a mathematics professor at New York University, revealed that he and Levant Alperge, a mathematician at Anthropic (an OpenAI competitor), were also working on solving the Navier–Stokes problem.
Buckmaster said some of their research was stored and processed via Codex, an OpenAI programming assistant tool, and on 3 September he learned that information about their research progress had been transmitted to OpenAI.
He observed that OpenAI’s interest in the Navier–Stokes problem seemed to increase after the company received data about his and Alperge’s work, raising questions about the timing and methods OpenAI used in its project.
Nevertheless, Buckmaster stressed he did not accuse OpenAI of wrongdoing and admitted he had not yet reviewed OpenAI’s full proof. He said it was necessary to disclose what he learned and when, to prevent misunderstandings about the sequence of announcements.
OpenAI denied accessing or viewing Buckmaster’s and Alperge’s work before it was publicly released and affirmed it did not access user data related to the Navier–Stokes problem.
However, the company acknowledged it cannot entirely rule out that anonymized usage data from the two researchers’ use of its products may have helped improve its AI models. It insisted that OpenAI’s proof evidence is significantly different from their work, even in the provable results.
OpenAI’s claimed success reflects a recent trend where AI, leveraging vast computational power, increasingly contributes to advanced mathematical research. Previously, OpenAI announced progress on another math problem proposed about 80 years ago, and Google DeepMind also reported AI capabilities solving high-level math problems.
OpenAI researcher Sebastian Bubeck described this achievement as a remarkable peak in AI’s development over the past 12 months, affirming in a press briefing that OpenAI did not use Buckmaster’s or Alperge’s work in solving the problem.
The Navier–Stokes problem is one of the world’s most challenging math questions, requiring proof that equations describing fluid motion, like water and air, yield continuous and non-anomalous solutions under various conditions.
If OpenAI’s proof passes independent review and meets the Clay Mathematics Institute’s criteria, it would be a major breakthrough for mathematics and AI. However, the Navier–Stokes problem is not yet officially solved, and OpenAI’s claim still requires expert validation.
This success comes amid growing concerns over AI safety, particularly about the rapid advancement of AI Agents’ capabilities. Debates continue over the appropriate regulatory oversight for AI systems surpassing human intelligence in cognitive and research tasks.