Inference
EN
Epoch AI

MirrorCode

Evaluated by Epoch AI on their own harness.

ModelsModels with a result.39
Top resultBest result on this benchmark.97.8%Claude Sonnet 5.5
Top-3 spreadPoints from first to third.4.4 pts
YearYear of release.—

Result and price

38 of 39 results

Results

39 results
#ModelResultThe score from the source.
01Claude Sonnet 5.5max effort97.8% ±1.5
02Claude Opus 5.5max effort95.6% ±2.6
03Claude Fable 5.1max effort93.3% ±3.9
04Claude Opus 5max effort87.8% ±7.2
05Claude Haiku 5.5max effortAbove fitFar above this model's index.78.9% ±8.0
06GPT-6.1 Terramax effort76.7% ±9.3
07Claude Fable 5max effort70.0% ±10.4
08Claude Opus 4.8max effort65.6% ±10.8
09DeepSeek V4.1 Flash 1017max effort64.4% ±11.4
10Claude Sonnet 5max effort63.3% ±10.7
11GPT-6.1 Solmax effort61.1% ±11.3
12GPT-6 Astramax effort60.0% ±11.8
13Gemini 4 Argonhigh effort58.9% ±11.7
14Muse Spark 1.3max effort54.4% ±10.8
14Grok 4.7xhigh effort54.4% ±11.4
14Kimi K3max effort54.4% ±11.4
17Qwen3.8 Maxxhigh effort52.2% ±10.9
18GPT-6 Solmax effort48.9% ±11.2
19Qwen3.8 Max (0902)xhigh effort47.8% ±10.3
20Muse Wondermax effort45.6% ±10.6
21GPT-5.6 Terramax effort41.1% ±12.4
21Gemini 3.7 Flashhigh effort41.1% ±12.2
21Grok 4.6xhigh effort41.1% ±11.0
24Gemini 3.8 Flashhigh effort38.9% ±11.9
25GLM-5.2max effort36.7% ±11.5
26DeepSeek V4 Pro 0813max effort33.3% ±10.0
26Claude Opus 4.6max effort33.3% ±10.0
28Grok 4.5high effort32.2% ±10.7
28GPT-5.2xhigh effort32.2% ±10.0
30Claude Opus 4.7high effort31.1% ±9.0
31Mistral Large 4high effort27.8% ±8.9
32GPT-6 Lunamax effort26.7% ±9.0
33GPT-5.5xhigh effort23.3% ±10.1
33GPT-5.4xhigh effort23.3% ±8.1
33Gemini 3.6 Flashhigh effort23.3% ±9.3
36GPT-5.6 Lunamax effort21.1% ±9.8
37GPT-5.6 Solhigh effort20.0% ±9.2
38Gemini 3.1 Prohigh effort8.9% ±4.6
39GPT-5high effort4.4% ±3.4