In a September 21 post, OpenAI said an internal model in training since August 28 has solved more than 100 long-standing problems. It named none and published no proofs. It also set up an unpaid advisory group of nine leading mathematicians, hosted at the Institute for Advanced Study.
New to this? Read it in simple words
- OpenAI says a new internal model has solved more than 100 open maths problems. It shared this in a blog post on September 21.
- It named none of the problems and published no proofs. It also set up an unpaid group of nine well-known mathematicians to give advice.
- In maths, a result counts when others can check the proof.
- Until the proofs are published and checked, the 100-problem figure is still a company claim.
- Open problem
- A maths question that no one has been able to solve yet.
- Proof
- A step-by-step argument that shows a maths result is true.
A large claim without a list
In a blog post on September 21, OpenAI said the model began training on August 28. It says the model has solved more than 100 open problems across most areas of mathematics. It adds that the speed surprised its own mathematicians.
OpenAI did not name the model or list the problems. It published no proofs and did not explain what it means by “resolved”. We found no outside mathematician who has checked any of the results.
Some reports said the 100 problems include the famous Navier–Stokes problem. OpenAI’s post does not say that. It counts Navier–Stokes as a separate result, which it announced on September 8.
The earlier Navier–Stokes proof is public and checked by computer, but it answers a narrower version of the problem.
The earlier Navier–Stokes proof answers a narrower question
The Navier–Stokes equations describe how liquids and gases move. On September 8, OpenAI published a 166-page paper and a proof checked by Lean, a computer proof system. OpenAI says about 10,000 AI agents worked on it for about 88 hours.
The proof shows that a smooth outside force, pushing a fluid that starts at rest, can make the fluid’s speed grow without limit in a finite time. The official prize text from the Clay Mathematics Institute allows such a force.
But the version without an outside force is still open, Scientific American reports, and many experts see that as the real question. Mathematicians who spoke to NPR believe the proof is correct. They also said it is very hard for people to read, and it has not been peer-reviewed.
The Clay Institute said on September 11 that the problem appears to be settled, but it described its review as deliberately slow. It has awarded no prize, and OpenAI says it will not claim the $1 million.
Nine mathematicians who may speak freely
The new Advisory Group on Mathematics and Artificial Intelligence has nine members, including Timothy Gowers, Martin Hairer, and Edward Witten. The Institute for Advanced Study in Princeton hosts it.
Members are unpaid. They may give advice that OpenAI did not ask for, speak in public, and choose new members. They have no say over the pace of OpenAI’s research. Their first task is to advise OpenAI on how to release its results.
OpenAI presented the group as its answer to an open letter from 25 Fields Medal winners on September 11. The letter criticised AI companies in general and named none. Hairer signed it, and he has now joined the group.
In August, experts quoted by Scientific American accused OpenAI of giving poor credit to earlier work in its maths papers. OpenAI said it took responsibility. Other projects work differently: six mathematicians used AI agents to solve an old symmetry problem about the group M23, and published their paper for others to check.
Sources
Every fact in this story comes from the sources below. Open them to check our work.
- 1
- 2
- 3
- 4Primary source · September 21, 2026Announcing the Advisory Group on Mathematics and Artificial Intelligence Advisory Group on Mathematics and AI
- 5Research · September 21, 2026OpenAI forms math advisory group as its AI resolves more than 100 open problems TechCrunch
- 6Research · September 22, 2026AI solved one of math’s hardest problems. Humanity learned nothing (so far) NPR
- 7
- 8Research · September 14, 202625 winners of math’s “Nobel Prize” decry the AI invasion of their discipline Scientific American
- 9Research · August 6, 2026OpenAI’s latest math breakthroughs commit research misconduct, experts say Scientific American
- 10Research · September 22, 2026Mathematicians use AI to find mysterious symmetries, solving decades-old problem Scientific American
We read OpenAI’s two posts, the advisory group’s own statement, and the Clay Institute’s notice. We compared them with TechCrunch, NPR, and Scientific American. The 100-problem figure is OpenAI’s claim, and we found no outside check of it.