OpenAI claims 722 papers on mathematics from one AI model

On October 6, 2026, OpenAI published abstracts of 722 papers on mathematics. Abstracts are short notes that tell about a paper. An internal model wrote them, and OpenAI did not give its name. Each paper claims a solution to a hard problem in 17 fields of mathematics. Software verified some of the proofs. Groups of mathematicians examine the other proofs.

How it works

A mathematician who knows some of the problems says that thousands of AI agents work in parallel. They try many paths, go back and try again. They check their work with proof assistants, which are programs that make sure that the logic is sound.

The numbers

The document has 722 abstracts in 17 fields. Software verified some proofs. The number of verified proofs is not given. In September 2026, OpenAI showed a 146-page Navier-Stokes proof. Mathematicians will read it for a long time.

What is not known

OpenAI did not tell the name of the model or its method. It did not use peer inspection, the usual check by other experts before a paper is published. Critics say that AI proofs use notation and loopholes that are hard to read. In September, some persons accused OpenAI of use of the work of mathematicians.

Insight: Checks are now the rare skill

OpenAI does not show the model that wrote the 722 papers. For the September Navier-Stokes proof, some groups of experts say that the method is not the method that OpenAI claims. Noam Brown of OpenAI says that no person knows the part that 10,000 agents added. OpenAI seems to hold the claims and the speed. The value moves to the groups that can check them. Leaders must get external checks.