OpenAI's frontier model drops 722 math manuscripts, including 90 of top 500 open problems
TheZvi · x · 2026-10-09
OpenAI published a huge batch of new mathematical results from an internal frontier model on GitHub: 90 of the top 500 open problems in math, 722 manuscripts (719 after withdrawals) in 372 families. One model, mostly single-prompt runs averaging 3 hours of compute per solution, out of 4,000 attempted problems, with prompts instructing it to fully resolve even open problems.
More from AGI Musings
- "Solving it kills the grants": AI in math rekindles academia critique — _AustinCalvert_ · 2026-10-10
- "Math belongs to taxpayers": AI solving math sparks ownership debate — _AustinCalvert_ · 2026-10-10
- OpenAI's Noam Brown Delivers Concluding Remarks on AI Alignment at COLM 2026 — erichorvitz · 2026-10-10
- RCT: AI as tutor raises test scores that persist; letting AI write for you fades within a week — emollick · 2026-10-10
- Freeze AI: Author borrows 1980s nuclear freeze movement playbook to rein in AI industry — GarrisonLovely · 2026-10-10
- Dean Ball on 8 underappreciated things to watch over the next decade, from moon robots to Lean — deanwball · 2026-10-10