OpenAI math results release (722 manuscripts, 372 result groups)
OpenAI published 722 math manuscripts in 372 groups of related results on GitHub, produced by an unreleased internal model that attempted about 4,000 research problems.
OpenAI released a large batch of AI-generated research manuscripts, each group holding a main result, supporting arguments or alternative proofs. Many results come with Lean proofs, and OpenAI says more will be added. It also published 10 summaries of the model's reasoning and compute estimates, with the average result needing about three hours of ChatGPT Pro thinking. The results are at different stages of verification and may contain errors.
- Date
- Tuesday, 6 October 2026
- Lab
- OpenAI
- Kind
- paper
- Access
- paper only
Figures
| Measure | Value | Measured by |
|---|---|---|
| Manuscripts released | 722 grouped into 372 sets of related results | company |
| Problems attempted | approximately 4,000 research problems given to the model during testing | company |
| Average compute per result | about 3 hours of ChatGPT Pro thinking equivalent estimate | company |
The model is unreleased; OpenAI says it is working to release it. The IAS advisory group stressed its consultation was not an endorsement of the results.
Sources
- aa.com.tr/en/science-technology/openai-releases-findings-on-hundreds-of-math-problems/4081
- qz.com/openai-math-results-github-millennium-prize-100726
- edtechinnovationhub.com/news/openai-releases-722-ai-generated-mathematics-papers-as-resear
This record was checked against its sources on 7 October 2026. How we check