OpenAI's math results, explained
Thirty-second narrated explainers, most important first. Press play for the top 20, or pick a field below.
All 372 results, by field
About
OpenAI released a collection of 722 manuscripts in 372 families, produced by an unreleased internal OpenAI model. According to its README, the vast majority came from one fixed procedure, using on average three hours of ChatGPT Pro thinking compute per result.
None of it is peer reviewed. Some results come with a Lean formalization: a version of the proof checked by a computer, which is strong evidence provided the formal statement says what the paper says. Many results are not formalized, and the README says that some of those "could have issues". Every chapter here describes what a manuscript claims.
The ranking is Chad Smith's editorial judgment of impact (how famous the problem is, times how much the result would settle), not OpenAI's. Source: github.com/openai/math.
Made by Chad Smith, 2026 · chadsmith.dev