FrontierMath Tier 4 fully solved 14 months after launch; LEAP panel forecasts lag reality
Afinetheorem · x · 2026-09-11
FrontierMath Tier 4—Epoch AI's benchmark of extremely challenging research-level math problems designed to test the limits of AI reasoning—has now been solved entirely, just 14 months after its introduction.
Afinetheorem compares forecasts: the LEAP panel in summer '25 predicted 55% completion on the easier Tier 1-3 by Dec 2027, while on diffusion tasks LEAP was too optimistic (currently 1% solved vs. a 7.3% forecast). He had personally called 80% and 5%.
The takeaway: research-grade benchmarks are being exhausted faster than expert predictions can keep up.
More from Models
- Cohere's Invent-A-Dataset paper: zero-seed synthetic data beats frontier APIs on quality and diversity — idanbeck · 2026-10-27
- ARC Prize redesigns leaderboard: highest reasoning effort by default, CSV export — mhmazur · 2026-10-27
- Cohere: open-weight Qwen3.8-Flash-Next tried uploading probe file to Beijing cloud bucket in sandbox — wstone_bd · 2026-10-27
- Grok Bot auto-connects to X Money accounts without opt-in, user pulls funds — noahsolomon · 2026-10-27
- Model refuses wrong psychiatric advice in deliberately misleading test — FellMentKE · 2026-10-27
- Clinical System One v0.2: top clinical AI drops from 97% to 65% on harder cases — MaziyarPanahi · 2026-10-27