Epoch AI Launches FrontierMath Erdős: GPT-6 Astra Scores 3%, All Other Models Zero
haider1 · x · 2026-09-05
Epoch AI announced FrontierMath Erdős, a benchmark of 68 significant unsolved Erdős problems curated by Thomas Bloom and formalized in Lean, where AI systems must prove or disprove each within a fixed compute budget. It addresses curation, verification, and replicability issues informal Erdős-benchmark usage has had (erdosproblems.com lists 1,217 problems, 652 unsolved). GPT-6 Astra scored 3% — the only nonzero result: Fable 5.1, Fable 5, Sol 5.6, and 5.5 all scored 0%. Across additional attempts Astra eventually solved 5 of 68 open problems, but at over $220,000 in compute cost.
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