OpenAI’s 10,000 AI Agents Tackle $1M Math Problem, Sparking Mathematician Debate

OpenAI says an unreleased experimental AI system has generated a potential solution to one of mathematics’ seven Millennium Prize Problems, but questions remain about how independently the system arrived at its result.

According to OpenAI, the system coordinated around 10,000 AI agents that worked for approximately 88 hours to tackle the Navier-Stokes existence and smoothness problem, one of the field’s most renowned unsolved challenges. A successful solution to any of the seven Millennium Prize Problems carries a $1 million award.

The Navier-Stokes equations are fundamental to understanding how fluids such as air and water behave. The mathematical challenge centers on whether a smooth fluid flow can eventually develop a singularity in which velocity becomes unbounded.

OpenAI’s proposed proof argues that such a breakdown can occur. The system describes a fluid vortex that becomes progressively compressed and stretched until its velocity becomes infinite within a finite period, while the overall energy of the system remains finite.

Such infinite velocities do not occur in physical fluids under current scientific understanding. At sufficiently small scales and extreme conditions, the assumption that fluids can be modeled as continuous substances would no longer accurately represent physical reality.

Thousands of AI Agents Worked on the Problem

OpenAI said the result was produced by an internal model that is substantially more capable than GPT-6 Astra, which the company describes as its most advanced publicly available model.

Approximately 10,000 AI agents collaborated on the Navier-Stokes problem, exchanging about 2.7 million messages and producing around 130 billion tokens. Astra then spent an additional 17 hours converting the proposed argument into a formal proof and checking its validity.

If the approach proves correct, its significance could extend beyond pure mathematics. Navier-Stokes equations are widely used in fields including aircraft engineering, weather prediction and research into blood circulation. A deeper understanding of when these equations can fail could eventually help scientists model highly turbulent or extreme fluid behavior.

The larger significance may lie in the collaborative AI approach itself. If thousands of autonomous agents can make meaningful progress on a mathematical problem that has remained unresolved for nearly a century, similar systems could eventually be applied to difficult challenges in fields such as energy, materials science, aerospace and medicine.

Questions Over the AI’s Independence

Despite the potential breakthrough, the result has already prompted debate over whether OpenAI’s system reached the solution entirely on its own.

NYU mathematician Tristan Buckmaster and Anthropic researcher Levent Alpöge had independently been studying a related problem in fluid dynamics and had incorporated AI tools into their research. Buckmaster subsequently questioned whether unpublished aspects of their work could have somehow influenced OpenAI’s systems.

In posts on Mastodon, Buckmaster said he and Alpöge had spent about a year exploring an unconventional approach involving a smooth external force. Codex was among the AI tools they used extensively during the research.

Buckmaster said the approach was unusual enough that he was not aware of many other researchers pursuing the same direction. That led him to question how OpenAI’s system could have arrived at a similar method only days after the company became aware of the researchers’ progress.

Buckmaster also described conversations with OpenAI in which he was initially shown a prompt and told that the internal research effort had started without human guidance beyond providing the mathematical problem itself.

According to Buckmaster, that explanation changed during subsequent discussions. He said OpenAI eventually acknowledged that the initial prompt had been submitted only days earlier, after information about his and Alpöge’s work had reached the company.

OpenAI disputed the suggestion that its researchers or AI agents had viewed the pair’s work before it was published. The company also said that no specific user data had been accessed.

However, OpenAI acknowledged that it could not completely exclude the possibility that de-identified information derived from users’ interactions with its products had contributed to model improvements. The company maintained that its proposed proof is substantially different from the researchers’ work.

The $1 Million Prize Has Not Been Awarded

For now, OpenAI’s result cannot be considered an officially accepted solution to the Navier-Stokes problem.

The Clay Mathematics Institute, which administers the Millennium Prize Problems, requires proposed solutions to undergo extensive examination and receive broad acceptance within the mathematical community before a prize can be awarded.

That means mathematicians will need to independently scrutinize the proof and determine whether its reasoning holds up.

OpenAI has also said it is not seeking the $1 million prize. Instead, the company is presenting the result as an indication of how rapidly its experimental research systems are advancing and how multi-agent AI could potentially contribute to difficult scientific and mathematical problems.

About The Author