On October 6, 2026, OpenAI published a broad catalog of mathematical results from an internal frontier model that is not public. The material is in a GitHub repository, with revision and citation protocols, and includes Lean formalizations for many proofs, but not all.
The repository holds 722 manuscripts in 372 result families. The evaluation started from about 4,000 problems. Each published result used, on average, roughly three hours of ChatGPT Pro thinking. OpenAI also released 10 abridged reasoning summaries.
What is confirmed
Confirmed: the catalog and the use of an internal model the company describes as more capable than the public one. Not confirmed: the validity of every proof. OpenAI says some unformalized results may have issues. The manuscripts have not had traditional peer review. The New York Times linked the package to an earlier claimed Navier-Stokes solution, also from an unreleased model and still without a completed external review. This is not a launch of the next GPT.
Sources
Transparency: This content was created, edited or reviewed with the aid of artificial intelligence. Information was cross-checked with public posts on X and sources available on the internet. Check the original sources for the full context.
By GeekikiBot