Epoch AI launches FrontierMath Erdős benchmark of 68 unsolved problems; Astra tops it at 2/68
keviv9 · x · 2026-09-04
Epoch AI, with mathematician Thomas Bloom, announced FrontierMath Erdős, a new benchmark of 68 unsolved Erdős problems curated from erdosproblems.com, where AI systems must prove or disprove them in Lean within a fixed compute budget.
Key points:
- Bloom's catalog lists 1,217 Erdős problems, 652 still unsolved; these problems have informally become the standard for tracking AI math progress, and the new benchmark adds rigor around curation, verification, and replicability (open-source scaffold included).
- No prior model solved any of the problems, but Astra solved 2/68 (3%), setting the top score.
Related event: Epoch AI Launches FrontierMath Erdős; GPT-6 Astra First to Break Zero(10 posts)→
More from Models
- Opus 5.5 tops OpenRouter spending in 20 of 29 task categories — mmurph · 2026-10-21
- Meta team builds open-source 30B LLM from scratch in months, hails 'second era' — TimDarcet · 2026-10-21
- DFlash drafter for Muse Glimmer-30B: 5 layers, blockwise 16-token spec decoding — TimDarcet · 2026-10-21
- EvoPress quantization shrinks 30B model from 56GB to 19GB at near-bf16 accuracy — TimDarcet · 2026-10-21
- Radar chart: Muse Glimmer-30B benchmarked against Gemma4-31B and Qwen3.6-27B — TimDarcet · 2026-10-21
- Privacy evals: Muse Glimmer-30B leaks less than Qwen3.6-27B on CIMemories — TimDarcet · 2026-10-21