FrontierMath Tier 3, predicted by Terence Tao to resist AI for years, now saturated
Eliv_nurotic · reddit · 2026-10-05
Epoch AI's FrontierMath dashboard shows Tier 3 has been fully saturated by frontier models less than two years after the benchmark's upper tier was characterized. In December 2024, Fields medalists Terence Tao, Timothy Gowers, and Richard Borcherds called Tier 3 "exceptionally challenging," with Tao predicting it would "resist AIs for several years." That prediction did not hold; the linked Epoch AI chart tracks scores across Tiers 1-4 by model release date.
Related event: AI Saturates FrontierMath Benchmarks Once Thought Years Away(3 posts)→
More from Models
- GPT-6 works for 37 hours straight on a single task, wowing developers — tom_doerr · 2026-10-05
- Fan-Made Timeline Maps All 44 Major Qwen Releases, From 7B to 2.4T Open Weights — ai-lover · 2026-10-05
- GPT Astra 6 (Ultra) Called 'Trash' on Codex: Ignored Instructions, 'Chose Convenience Over Scientific Correctness' — raskingballs · 2026-10-05
- Ethan Mollick slams OpenAI's GPTs shutdown: enterprises and educators left behind — emollick · 2026-10-05
- Stealthy German lab Aleph Alpha drops open-weight Kolibri: 78B params, 3.46B active, 1M context — skdh · 2026-10-05
- Midjourney CEO shares model "inner world" visualizations: Mistral imagines cookie heists, OpenAI draws ASCII islands — DavidSHolz · 2026-10-05