Inference
EN
Epoch AI

FrontierMath-Tier-4-2025-07-01-Public

Evaluated by Epoch AI on their own harness.

ModelsModels with a result.36
Top resultBest result on this benchmark.100.0%Claude Fable 5
Top-3 spreadPoints from first to third.100.0 pts
YearYear of release.—

Result and price

28 of 36 results

Results

36 results
#ModelResultThe score from the source.
01Claude Fable 5max effort100.0%
02GPT-5.4xhigh effort50.0% ±50.0
03Gemini 3.5 Flashhigh effort0.0%
03Gemini 3.1 Pro0.0%
03Claude Opus 4.7xhigh effort0.0%
03GPT-5.2high effort0.0%
03Claude Opus 4.6max effort0.0%
03Gemini 3 Pro0.0%
03Qwen 3.6 Max (preview)0.0%
03Claude Sonnet 4.60.0%
03GLM-5.10.0%
03GPT-5high effort0.0%
03Gemini 3 Flash0.0%
03Qwen3.6 Plus0.0%
03o3high effort0.0%
03GLM-50.0%
03Grok 40.0%
03GPT-5 minihigh effort0.0%
03Kimi K2 Thinking0.0%
03Claude Sonnet 4.50.0%
03GLM-4.50.0%
03o4-minihigh effort0.0%
03Grok 3 Minihigh effort0.0%
03Claude Opus 40.0%
03Claude Sonnet 40.0%
03GPT-5 nanohigh effort0.0%
03GLM-4.60.0%
03o3-minihigh effort0.0%
03Grok 3 [Beta]0.0%
03GPT-4.10.0%
03Claude 3.7 Sonnet0.0%
03Claude 3.5 Sonnet0.0%
03Claude 3.5 Sonnet (October 2024)0.0%
03DeepSeek-R1 (May 2025)0.0%
03Gemini 2.5 Pro (Jun 2025)0.0%
03Qwen3-235B-A22B (Jul 2025)0.0%