Inference
Epoch AI

EBR-bench

Evaluated by Epoch AI on their own harness.

ModelsModels with a result.20
Top resultBest result on this test.76.2%GPT-6 Astra
Top-3 spreadPoints from first to third.30.5 pts
YearYear of release.N/A

Result and price

020406080100$1$3$10$30$100RESULTOUTPUT PRICE PER 1M TOKENS · LOG SCALE

Results

20 results
#ModelResultThe score from the source.
01GPT-6 Astramax effort76.2% ±8.4
02Claude Fable 5.1max effort57.1% ±6.0
03Claude Opus 5max effort45.7% ±3.6
04GPT-5.6 Solmax effort44.8% ±4.2
05Claude Fable 5max effort39.5% ±2.0
06GPT-5.5xhigh effort34.3% ±2.0
07Grok 4.6xhigh effort30.5% ±1.6
08Claude Opus 4.8max effort28.6% ±1.4
09GPT-5.4xhigh effort25.4%
10GPT-5.2xhigh effort23.0%
11Claude Opus 4.7max effort19.0%
12Gemini 3.1 Pro14.3%
12Claude Opus 4.514.3%
14Claude Opus 4.6max effort12.7%
14GPT-5high effort12.7%
16Qwen3.7 Max9.5%
16GLM-5.2max effort9.5%
18Claude Opus 4.17.9%
19Gemini 3.5 Flashhigh effort4.8%
20Claude Sonnet 4.52.4%