OpenAI uploads hundreds of model-solved open math problems with Lean proofs to GitHub
AI寒武纪 · wechat · 2026-10-07
Per this Chinese newsletter recap, OpenAI announced that an internal frontier model produced a batch of new mathematical results, now published in a GitHub repo (openai/math) with extensive Lean formal verification code. Each solution reportedly took about three hours of ChatGPT Pro deep-thinking compute; OpenAI released 10 model reasoning summaries, attempt statistics, and compute cost estimates. It will sponsor workshops and academic exchanges to help mathematicians absorb the results, and is running safety evaluations before eventually releasing the underlying model. The Institute for Advanced Study convened an independent advisory group — including Ed Witten, Timothy Gowers, and Martin Hairer — which set release guidelines from 600+ peer surveys and cautioned that mathematicians must retain the freedom to pose their own questions and get equal access to frontier AI tools and compute.
More from Research
- OpenAI proof dump shows first superlinear circuit lower bound, but progress stays slow — thomasahle · 2026-11-22
- Fortnow: Fall's dramatic results show P vs NP still far from solved, as OpenAI proof dump yields first superlinear circuit lower bound — thomasahle · 2026-11-22
- One photo, 8 VLAs, only 3 correct: ERQA embodied-reasoning benchmark questioned — giffmana · 2026-11-22
- Fleuret: treat AI as oracles, not explainers — human-comprehensible math will be negligible — francoisfleuret · 2026-11-22
- Vovk paper shows probabilistically nonrandom objects have vanishing universal measure — _onionesque · 2026-11-22
- AI generates a Riemann proof, AI generates the rebuttal—and it scores 100% AI — zetalyrae · 2026-11-22