benchgap
General

Epoch Automation Reports leaderboard

As of 2026-11-18, the highest measured score on Epoch Automation Reports is 65.3% by Claude Fable 5.1. 53 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Claude Fable 5.265.7%estimated ± 3.0 pp, low confidence
2GPT-6.2 Astra65.7%estimated ± 3.0 pp, low confidence
3Claude Opus 5.565.7%estimated ± 3.0 pp, low confidence
4GPT-6.1 Sol65.6%estimated ± 3.0 pp, medium confidence
5GPT-6.1 Terra65.6%estimated ± 3.0 pp, medium confidence
6Gemini 4 Carbon65.6%estimated ± 3.0 pp, medium confidence
7Claude Sonnet 5.565.6%estimated ± 3.0 pp, medium confidence
8Gemini 4 Argon65.5%estimated ± 3.0 pp, medium confidence
9Claude Fable 5.165.3%measured
10Claude Opus 565.1%estimated ± 3.0 pp, medium confidence
11GPT-6 Astra65.1%measured
12GPT-6 Sol64.9%estimated ± 3.0 pp, medium confidence
13GPT-5.6 Sol64.8%estimated ± 3.0 pp, medium confidence
14Kimi K3.1 Pro64.7%estimated ± 3.0 pp, medium confidence
15GPT-5.6 Terra64.5%estimated ± 3.0 pp, medium confidence
16Muse Wonder64.5%estimated ± 3.0 pp, medium confidence
17GLM-5.364.2%estimated ± 6.2 pp, low confidence
18Claude Haiku 5.563.0%estimated ± 3.0 pp, medium confidence
19GPT-5.561.7%estimated ± 3.0 pp, medium confidence
20GPT-6 Luna61.7%estimated ± 3.0 pp, medium confidence
21Claude Opus 4.860.7%estimated ± 3.0 pp, medium confidence
22GPT-5.6 Luna60.7%estimated ± 3.0 pp, medium confidence
23Claude Sonnet 559.5%estimated ± 8.4 pp, low confidence
24Grok 4.659.1%measured
25GPT-5.256.9%estimated ± 3.0 pp, medium confidence
26GPT-5.455.8%estimated ± 3.0 pp, medium confidence
27Claude Fable 553.3%estimated ± 3.0 pp, medium confidence
28Qwen3.8 Max53.0%measured
29Kimi K352.4%measured
30DeepSeek V4.1 Flash 101750.3%estimated ± 3.0 pp, medium confidence
31Qwen3.8-27B50.2%estimated ± 6.3 pp, low confidence
32MiMo-V2.6-Flash48.6%estimated ± 6.3 pp, low confidence
33Claude Opus 4.748.3%estimated ± 3.0 pp, medium confidence
34GLM-5.248.1%estimated ± 8.4 pp, low confidence
35MiMo-V2.6-Pro46.1%estimated ± 6.2 pp, low confidence
36MiniMax M3.1 Flash45.6%estimated ± 6.3 pp, low confidence
37Muse Spark Open44.6%estimated ± 6.3 pp, low confidence
38Hy4 preview44.5%estimated ± 6.3 pp, low confidence
39DeepSeek V4 Pro43.6%estimated ± 6.3 pp, low confidence
40Ling 3.1 Flash43.3%estimated ± 6.3 pp, low confidence
41Gemini 3.8 Flash42.3%measured
42Muse Spark 1.340.9%estimated ± 6.3 pp, low confidence
43Step 5 Preview40.5%estimated ± 6.3 pp, low confidence
44MiniMax M340.2%estimated ± 6.3 pp, low confidence
45Qwen3.6-27B34.9%estimated ± 6.3 pp, low confidence
46Beam 1.0 501B-A23B34.4%estimated ± 6.3 pp, low confidence
47Claude Opus 4.533.5%estimated ± 3.0 pp, low confidence
48Claude Opus 4.633.5%estimated ± 3.0 pp, low confidence
49Gemma 4 31B32.2%estimated ± 6.3 pp, low confidence
50Inkling-Small30.4%estimated ± 6.3 pp, low confidence
51Gemini 3.1 Pro27.8%estimated ± 3.0 pp, low confidence
52Gemini 3.7 Flash27.8%estimated ± 3.0 pp, low confidence
53Gemma 4 26B A4B24.3%estimated ± 6.3 pp, low confidence
54Gemini 3.6 Flash16.1%estimated ± 3.0 pp, low confidence
55Grok 4.513.7%estimated ± 3.0 pp, low confidence
56Kimi K2.611.5%estimated ± 3.0 pp, low confidence
57GPT-511.1%estimated ± 8.4 pp, low confidence
58Grok 4.79.3%estimated ± 3.0 pp, low confidence
59Mistral Large 49.3%estimated ± 3.0 pp, low confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General