OpenAI Says 10,000 AI Agents Solved $1M Maths Problem as Researchers Question Timeline

OpenAI is facing a dispute with researchers after claiming it deployed about 10,000 AI agents to produce a solution to a legendary mathematical puzzle that had baffled human experts for nearly a century.

Artificial intelligence models and thousands of automated agents helped the tech firm produce a claimed solution to a complex mathematics challenge in just hours, a problem that had previously stumped human researchers for decades.

By turning a swarm of around 10,000 autonomous digital agents loose on a notoriously tricky mathematical hurdle, the creators of ChatGPT said they produced a solution in about 88 hours.

OpenAI Claims a Breakthrough in Just 88 Hours

Notably, mathematicians have spent nearly 90 years trying to resolve the Navier-Stokes problem, which concerns whether equations used to describe the movement of fluids such as air and water can develop singularities under certain conditions.

Describing the breakthrough as a major 'milestone', the company pointed to the discovery as clear proof that artificial intelligence capabilities are accelerating at a rapid pace.

The Clay Mathematics Institute — a US-based organisation offering substantial cash rewards for solving specific mathematical mysteries through its Millennium Prize — has yet to independently verify or formally accept OpenAI's claimed solution.

A recently developed internal model trained by the company demonstrated strong mathematical capabilities, showcasing how these data-driven computer systems process vast amounts of information to spot and predict patterns.

Restricted to internal use because it was more capable than the company's latest publicly available model, the company's powerful new model prompted researchers to test its muscle on some notorious mathematical hurdles.

OpenAI acknowledged setting thousands of AI bots trained on its new internal system to work on the uncracked challenges on 1 September, after it had 'heard rumours that two Millennium Prize problems had been resolved'.

Roughly 88 hours after unleashing 10,000 bots on the challenge, OpenAI said it had reached the claimed result on 5 September, cracking the Navier-Stokes existence and smoothness problem — a notorious mathematical hurdle in fluid dynamics.

OpenAI said the agents exchanged nearly three million messages and generated 130 billion output tokens — the individual strings of text and code generated by an AI — solely on the Navier-Stokes challenge. GPT-6 Astra then took about 17 hours to verify the result using Lean, a system designed to formally check mathematical proofs.

The AI Effort Cost Millions

Despite the bots finishing the task in what felt like a flash, the company noted that the agents exchanged nearly three million messages and generated 130 billion output tokens — the individual strings of text and code generated by an AI — solely on the Navier-Stokes challenge. OpenAI has described the effort as consuming millions of dollars in computing resources.

OpenAI's published API pricing lists GPT-6 Astra's output at $50 per million tokens. At that list price, 130 billion output tokens would equate to about $6.5 million. However, this is only a theoretical comparison based on public API pricing and does not establish what OpenAI actually spent on its internal research effort.

The Navier-Stokes problem is one of seven Millennium Prize Problems established by the Clay Mathematics Institute, with each carrying a $1 million (£740,000) reward for an accepted solution.

Explaining their motivation on Tuesday, the company stated, 'Our goal in releasing this result is to report on the substantial progress of our AI models', while clarifying that 'We do not intend to claim the Millennium Prize for this result.'

Researchers Challenge OpenAI's Timeline

Controversy surrounded the announcement after New York University mathematics professor Tristan Buckmaster revealed hours earlier that he and Levent Alpöge — a researcher at OpenAI competitor Anthropic — had also been working on related problems in fluid dynamics.

The pair utilised OpenAI's Codex coding assistant for their research, but Buckmaster noted that on 3 September, he said he had learned that information about his team's progress had been passed to OpenAI.

Dropping his statement just hours before OpenAI went public with its Navier-Stokes breakthrough, Buckmaster questioned whether the timing of OpenAI's work had been influenced by information about his team's progress, citing email threads detailing his direct challenges to the company regarding its timeline and methods.

Although he had not yet examined OpenAI's complete mathematical proof, Buckmaster felt compelled to share 'what I was told, when, and what was proposed to me...because the alternative is to let a sequence of announcements say something I know to be false.'

Buckmaster has also said he does not know whether his team's data was used and has explicitly stated: 'I do not know what their model did, or how. I do not know whether our data was used. I am not accusing anyone of anything.'

OpenAI Denies Using Their Research

Responding on Tuesday, OpenAI publicly congratulated Buckmaster and Alpöge on their 'concurrent work', describing the independent research as 'remarkable'.

Insisting it had never viewed 'any of their work through any means until they released it publicly,' the firm said its researchers and AI agents did not see the pair's work before it was released publicly and said no specific user data was accessed to solve the problem.

'While unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models,' the company added, noting that 'our proofs differ significantly and even the precise results proved are different.'

OpenAI has said its proof was developed independently and that its precise mathematical results differed from the researchers' work. The company also acknowledged the pair's work on a related forced Euler problem.

The Clay Mathematics Institute has not independently validated OpenAI's claimed solution, meaning the result has not yet been formally recognised as a solution to the Millennium Prize Problem.