Epoch AI's FrontierMath Erdos: GPT-6 Astra solves 3%, all others 0% on 68 open problems
chaumian · x · 2026-09-23
Epoch AI and the University of Manchester introduce FrontierMath Erdős (FME), a benchmark of 68 open Erdős conjectures (unsolved as of August 2026), selected from 652 open problems for their mathematical interest and difficulty. AI systems must prove or disprove each conjecture in the Lean proof assistant.
Every model faces the same fixed problems, runs autonomously with a $300 budget per problem. Of five AIs evaluated, only GPT-6 Astra scored 3%; all others scored 0%. The authors argue that prior one-off demonstrations of AI solving open math problems lack systematic study, which FME provides.
More from Models
- Claude Opus 5.5 shown handling a code review, 'taking Theo's job' for the day — 0xkarasy · 2026-09-23
- Sarvam's Saaras V4 adds keyterm prompting to boost speech transcription accuracy — cneuralnetwork · 2026-09-23
- Xiaomi's MiMo V2.6 Pro tops open-source leaderboard at ~$0.13 per task — heyshrutimishra · 2026-09-23
- Xiaomi MiMo V2.6 Pro tops open-source leaderboard at 46, costs ~$0.13 per task — heyshrutimishra · 2026-09-23
- Deep-scan reveals Muse agent's hidden powers: Instagram DMs, HomeKit control, its own email identity — flaneur451 · 2026-09-23
- Yacine on Chinese RL SOTA models: it's just on-policy synthetic data — yacineMTB · 2026-09-23