Altman Previews Next-Generation GPT: 10,000 Agents Collaborate to Complete Navier–Stokes Proof

Sam Altman has revealed an important clue about the next-generation GPT.

On November 2 local time, at the ADIPEC 2026 energy conference held in Abu Dhabi, Altman, speaking about the Navier–Stokes mathematical breakthrough previously announced by OpenAI, revealed that the work was completed by 10,000 Agents driven by the upcoming next-generation GPT.

They worked continuously for about 4 days, running collaboratively the entire time.

Altman particularly emphasized that this system had no traditional hierarchical structure and no unified person in charge. Each Agent could independently explore research directions, share progress, and decide what to do next based on existing results.

"No hierarchy, no boss, completely decentralized."

Altman believes this represents an entirely new way of working and doing research.

This statement also gave the outside world a clearer clue for the first time: the research model that had been used internally at OpenAI and whose capabilities far exceed GPT-6 Astra is very likely the next-generation GPT that will soon be released to the public.

It describes the laws of motion of fluids such as water and air, and is one of the most fundamental mathematical tools in fluid dynamics. From aircraft design and weather forecasting to research on blood flow, this set of equations is indispensable.

But surrounding it, a fundamental mathematical question has troubled researchers for nearly 90 years: in three-dimensional space, can a fluid motion that is initially smooth suddenly develop a singularity in finite time?

Simply put, the fluid velocity becomes infinite in finite time, and the originally smooth state of motion undergoes a loss of control in the mathematical sense.

In 2000, the Clay Mathematics Institute included this problem among the seven Millennium Prize Problems in mathematics, with a prize of $1 million for each problem.

On September 8 this year, OpenAI officially announced the results of this research.

According to OpenAI, the internal model constructed a mathematical proof: under the application of a smooth external force, a fluid that is initially at rest and whose state of motion is smooth can form a singularity in finite time while keeping total energy finite.

The key to this construction is a vortex that continuously contracts inward while being stretched. As the vortex core becomes smaller and smaller, the rotation speed continuously increases, ultimately forming a velocity divergence in finite time. OpenAI stated that this proof resolves cases C and D in the formal statement of the Navier–Stokes problem, and simultaneously released the paper and Lean formal proof files.

What is truly surprising is the research process behind it.

OpenAI disclosed that the entire experiment started on September 1. Researchers had different groups of Agents separately explore different mathematical problems and proof paths. The Agents could read cached internet information and could also run code.

As the research progressed, the team found that a breakthrough on a related Euler equation problem provided new clues for the Navier–Stokes research, so they shifted more computing resources to the latter. Subsequently, OpenAI used Codex to organize the interim results of different Agent groups, feeding valuable findings back to other groups to drive further research.

During the experiment, the internal model also underwent a further training upgrade, and the Agents were updated accordingly. Ultimately, the scale of Agents participating in the Navier–Stokes research reached about 10,000. They completed the core mathematical exploration in about 88 hours, during which they exchanged 2.7 million messages and generated about 130 billion Tokens. After that, GPT-6 Astra spent another 17 hours completing the formalization and verification work through Lean.

Next-Generation GPT

The core change may lie in Agent collaboration

The most noteworthy aspect of the information Altman revealed this time is that the capabilities of the next-generation GPT may no longer be limited to a single model completing a single complex reasoning task.

This experiment demonstrated another way of working: letting a large number of Agents driven by the same foundation model explore in parallel, exchange research progress with one another, and gradually form research results that can be further advanced.

Of course, 10,000 Agents does not mean 10,000 independently trained super models. They are driven by the same internal model and jointly complete tasks through tools, grouped communication, and integration of results. This system is also not without cost.

In a podcast conversation with Stripe co-founder John Collison on October 29, Altman revealed that OpenAI invested about $10 million in this Navier–Stokes research alone, possibly even slightly more than that figure. But he also made a bet: by September 2027, OpenAI is expected to be able to redo the same work at about one percent of the current cost.

The cost reduction will come from smarter models and better research methods. This also means that the release of the next-generation GPT may bring two changes: the capabilities of the underlying model itself will continue to improve, and the efficiency of large-scale Agent collaboration will also improve.

However, Altman has not currently announced the specific model name or official release date of the next-generation GPT, nor has he explained whether ordinary users will be able to directly access the ten-thousand-scale Agent collaboration system used in this experiment.

English text is being prepared.