Inference
EN
Epoch AI

EBR-bench

Evaluated by Epoch AI on their own harness.

ModelsModels with a result.31
Top resultBest result on this benchmark.90.5%Claude Fable 5.2
Top-3 spreadPoints from first to third.12.4 pts
YearYear of release.—

Result and price

31 of 31 results

Results

31 results
#ModelResultThe score from the source.
01Claude Fable 5.2max effort90.5% ±1.5
02Gemini 4 Carbonhigh effort87.6% ±6.8
03GPT-6.1 Terramax effort78.1% ±5.1
04GPT-6 Astramax effort76.2% ±8.4
05Claude Opus 5.5max effort71.4% ±7.1
06Gemini 4 Argonhigh effort64.8% ±4.9
07Claude Fable 5.1max effort57.1% ±6.0
08GPT-6.1 Solmax effort54.3% ±5.8
09GPT-6 Solmax effort53.3% ±4.6
10Claude Opus 5max effort45.7% ±3.6
11GPT-5.6 Solmax effort44.8% ±4.2
12Claude Haiku 5.5max effort43.8% ±1.8
13Claude Fable 5max effort39.5% ±2.0
14GPT-5.5xhigh effort34.3% ±2.0
15Muse Wondermax effort31.4% ±4.2
16Grok 4.6xhigh effort30.5% ±1.6
17Claude Opus 4.8max effort28.6% ±1.4
18GPT-5.4xhigh effort25.4%
19GPT-5.2xhigh effort23.0%
20Claude Opus 4.7max effort19.0%
21Mistral Large 4high effort18.1% ±3.2
22Gemini 3.1 Pro14.3%
22Claude Opus 4.514.3%
24Claude Opus 4.6max effort12.7%
24GPT-5high effort12.7%
26GLM-5.2max effort9.5%
26Qwen3.7 Max9.5%
28Claude Opus 4.17.9%
29Gemini 3.5 Flashhigh effort4.8%
30Kimi K2.62.4%
30Claude Sonnet 4.52.4%