Sharing AI progress in mathematics (722 math manuscripts)
OpenAI published 722 math manuscripts in 372 result families, produced by an unreleased internal frontier model, in a public GitHub repository with Lean formalizations of many proofs.
OpenAI released a large batch of new mathematical results from an internal frontier model, following its earlier Navier-Stokes announcement. It consulted the IAS Advisory Group on Mathematics and AI on how to release them. The repository includes Lean formalizations of many proofs, 10 reasoning summaries, compute estimates and statistics on attempted problems. The average result used about three hours of ChatGPT Pro thinking compute.
- Date
- Tuesday, 6 October 2026
- Lab
- OpenAI
- Kind
- paper
- Access
- paper only
Figures
| Measure | Value | Measured by |
|---|---|---|
| manuscripts | 722 grouped into 372 result families | company |
| average compute per result | about 3 hours of ChatGPT Pro thinking estimate in the release | company |
| problems attempted | about 4,000 reported by Latent Space, not confirmed on the OpenAI page I read | Latent Space |
The model is unreleased, and OpenAI says it is working to release it responsibly. Claims such as solving 90 of the top 500 open problems come from secondary coverage and were not verified.
Sources
- openai.com/index/sharing-ai-progress-in-mathematics/
- latent.space/p/ainews-quasi-riemann-hypothesis-openai
- simonwillison.net/2026/Oct/7/jake-boggan/
This record was checked against its sources on 7 October 2026. How we check