This week, OpenAI announced something remarkable.
An unreleased AI model, described as significantly more capable than the recently released GPT-6 Astra, had produced a proposed solution to one of mathematics’ most famous unsolved questions: the Navier-Stokes existence and smoothness problem.
It is a problem that has challenged mathematicians for decades and carries a US$1 million prize for a recognised solution.
But perhaps the most interesting part of the story isn’t the mathematics itself.
It’s how the AI tackled it.
What is the Navier-Stokes problem?
The Navier-Stokes equations date back to the 19th century and describe how fluids move.
In physics, a fluid can be a liquid or a gas, so these equations help describe everything from water flowing through a pipe and blood moving through an artery to weather systems and air moving around an aircraft.
Engineers and scientists already use the equations extensively.
The mystery lies deeper in the mathematics.
For three-dimensional flows, mathematicians have been unable to prove whether smooth solutions must always remain smooth, or whether under certain conditions they can develop what is known as a singularity, where some aspect of the calculated flow becomes unbounded and the normally smooth equations effectively break down.
The question has remained unresolved for roughly 90 years.
In 2000, the Clay Mathematics Institute selected Navier-Stokes existence and smoothness as one of its seven Millennium Prize Problems, offering US$1 million for a recognised solution.
Enter AI
On 8 September 2026, OpenAI announced that an unreleased internal AI model had produced a proposed solution.
But this wasn’t simply a case of asking one chatbot a very difficult maths question.
OpenAI says the group that produced the result involved around 10,000 AI agents working concurrently, with agents divided into groups and encouraged to explore different approaches.
The system arrived at the Navier-Stokes result after about 88 hours.
The scale is difficult to put into perspective. OpenAI says the Navier-Stokes effort alone involved approximately 2.7 million messages and 130 billion output tokens.
The resulting proof was then formalised and verified using Lean, a system for checking mathematical proofs, in a process that took another 17 hours.
The proposed solution shows that Navier-Stokes dynamics can develop a singularity in finite time, addressing the central question posed by the Millennium Problem.
That does not mean the matter is instantly settled. A result of this magnitude will face extraordinary scrutiny from mathematicians, and formal verification does not remove the importance of independent academic review.
OpenAI has also said it does not intend to claim the US$1 million Millennium Prize.
Why this matters beyond mathematics
For most of us, a proposed solution to Navier-Stokes isn’t going to change what we do tomorrow morning.
What may prove far more significant is how the problem was tackled.
Much of the public discussion around AI still focuses on individual assistants: writing an email, generating an image, analysing a document or helping someone write code.
This was different.
Instead of one AI assisting one human researcher, thousands of AI agents were concentrated on a problem that had resisted generations of mathematicians. Different groups explored different approaches, shared useful findings and ultimately contributed towards a single result.
That points towards a very different use of artificial intelligence: large populations of AI researchers working simultaneously on difficult scientific and technical problems.
If that approach continues to improve, the implications could extend well beyond pure mathematics.
Similar systems could potentially accelerate research in medicine, materials science, engineering, energy, computing and even the development of future AI systems themselves.
There will also be important questions about verification, research standards, attribution and how humans remain involved in the scientific process.
But the underlying capability is difficult to ignore.
What could this mean for small businesses?
The technology demonstrated here is a long way from everyday small-business use, but the underlying idea is worth paying attention to.
Today’s AI tools generally involve a person giving an assistant a task and receiving a response. The next stage may increasingly involve teams of specialised AI agents working together, with different agents researching, analysing, checking and completing different parts of a larger task.
For a small business, that could eventually mean asking an AI system to investigate a problem, compare options, analyse business data, prepare recommendations and check its own work, rather than having staff guide an AI through each individual step.
We’re not there yet, and a mathematical research project involving thousands of AI agents is very different from running a small business.
But the direction is interesting.
The biggest impact of AI on business may not come from making today’s individual tasks a little faster. It may come from changing the scale of the tasks that a small team can realistically take on.
ChatGPT writing a better email is useful.
10,000 AI researchers working on a problem for days at a time is something quite different.
And that may ultimately be the more important AI story to watch.