benchgap
Agentic · tools

AutomationBench leaderboard

As of 2026-11-21, the highest measured score on AutomationBench is 62.2% by MiniMax M3.1 Flash. 69 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1MiniMax M3.1 Flash62.2%measured
2ERNIE 5.5 Flash Preview61.9%measured
3DeepSeek V4.1 Flash 101760.5%measured
4Hy456.2%measured
5Gemini 4 Carbon53.9%measured
6Atria Dawn Preview53.8%measured
7Gemini 4 Neon53.4%estimated ± 6.1 pp, low confidence
8MiMo-V2.6-Pro53.1%measured
9Ling 3.1 Flash52.5%measured
10MiMo-V2.6-Flash52.3%measured
11GPT-6.2 Astra52.3%estimated ± 6.1 pp, low confidence
12Hy4 flash52.2%measured
13LongCat-2.551.8%measured
14Claude Fable 5.251.4%measured
15Gemini 4 Argon51.3%measured
16Gemini 4 Hydrogen50.6%estimated ± 6.1 pp, low confidence
17Ultramarine 1 Flash50.2%measured
18Mercury 2.549.9%estimated ± 9.1 pp, low confidence
19Mistral Medium 3.5 128B49.7%estimated ± 9.1 pp, low confidence
20Muse Spark 1.349.4%measured
21Fugu Cyber49.2%estimated ± 10.4 pp, low confidence
22GLM-5.3-Flash48.8%measured
23MiMo-V2.5-Pro48.8%estimated ± 9.1 pp, low confidence
24Nemotron 3 Ultra48.6%estimated ± 9.1 pp, low confidence
25Qwen3.6 Plus48.6%estimated ± 9.1 pp, low confidence
26GLM-5.148.6%estimated ± 9.1 pp, low confidence
27Claude Opus 4.548.6%estimated ± 9.1 pp, low confidence
28Grok 4.648.5%estimated ± 7.1 pp, low confidence
29Qwen3.5 397B48.5%estimated ± 9.1 pp, low confidence
30MiniMax M2.748.4%estimated ± 9.1 pp, low confidence
31Qwen3.6-35B-A3B48.4%estimated ± 9.1 pp, low confidence
32Pokee-Isaac 28B48.3%estimated ± 9.1 pp, low confidence
33Gemini 3.8 Flash Cyber48.3%estimated ± 10.4 pp, low confidence
34Kimi K2.548.3%estimated ± 9.1 pp, low confidence
35GLM-548.3%estimated ± 9.1 pp, low confidence
36GLM-5.348.2%measured
37Granite 4.2 30B48.0%estimated ± 9.1 pp, low confidence
38Granite 4.2 8B47.6%estimated ± 9.1 pp, low confidence
39MiniMax M347.6%estimated ± 9.5 pp, low confidence
40Kimi K3.1 Pro47.4%estimated ± 6.1 pp, low confidence
41Qwen4 Max47.3%estimated ± 6.1 pp, low confidence
42Claude Sonnet 5.547.2%estimated ± 6.1 pp, low confidence
43GPT-6.2 Sol47.1%estimated ± 6.1 pp, low confidence
44Muse Ember47.1%estimated ± 6.1 pp, low confidence
45Grok 4.747.1%estimated ± 6.1 pp, low confidence
46GPT-6.1 Terra47.1%estimated ± 6.1 pp, low confidence
47Mistral Large 447.1%estimated ± 6.1 pp, low confidence
48GPT-6 Luna47.1%estimated ± 6.1 pp, low confidence
49Beam 1.0 501B-A23B47.1%estimated ± 6.1 pp, low confidence
50Qwen3.8-27B47.1%estimated ± 6.1 pp, low confidence
51Inkling 2 Small47.1%estimated ± 6.1 pp, low confidence
52Muse Wonder Full47.1%estimated ± 6.1 pp, low confidence
53Granite 4.2 3B46.2%estimated ± 9.1 pp, low confidence
54Claude Mythos 545.1%estimated ± 10.4 pp, low confidence
55Beam 1.0 501B-A23B FP844.7%estimated ± 9.1 pp, low confidence
56Beam 1.0 501B-A23B NVFP444.4%estimated ± 9.1 pp, low confidence
57Spark-X2.6-4B44.2%estimated ± 9.1 pp, low confidence
58Gemini 3.5 Flash Cyber44.2%estimated ± 10.4 pp, low confidence
59Claude Mythos Preview44.1%estimated ± 10.4 pp, low confidence
60Step 5 Preview44.0%measured
61Mercury 343.1%estimated ± 9.1 pp, low confidence
62GPT-6 Astra41.4%measured
63LFM3-1.7B41.3%measured
64Claude Opus 5.540.0%measured
65Gemini 3.8 Flash39.2%estimated ± 9.5 pp, low confidence
66Nemotron 3.5 Super VL 120B A12B38.5%estimated ± 9.1 pp, low confidence
67Nemotron 3.5 Super VL 120B A12B NVFP437.9%estimated ± 9.1 pp, low confidence
68Nemotron 3.5 Bolt 74B-A9B37.4%estimated ± 9.1 pp, low confidence
69Nemotron 3.5 Bolt 74B-A9B BF1637.4%estimated ± 9.1 pp, low confidence
70Apodex 1.137.1%estimated ± 7.4 pp, low confidence
71GPT-5.6 Luna36.5%estimated ± 10.4 pp, low confidence
72Ripple-65B Preview36.4%measured
73GPT-6.1 Sol36.1%measured
74Qwen3.8-Flash-Next36.1%estimated ± 7.1 pp, low confidence
75GPT-5.6 Sol34.0%estimated ± 7.1 pp, low confidence
76GPT-6 Sol33.2%measured
77Hy4 preview32.1%measured
78DeepSeek V4 Pro31.8%measured
79Claude Fable 5.131.4%measured
80dots3-note Preview31.0%estimated ± 7.4 pp, low confidence
81Kimi K330.8%measured
82Gemini 3.7 Flash30.4%measured
83Grok 4.530.2%estimated ± 7.1 pp, low confidence
84Nemotron 3.5 Lightning 30B A3B NVFP430.1%estimated ± 9.1 pp, low confidence
85Claude Opus 4.729.2%estimated ± 10.4 pp, low confidence
86Apodex 1.1 Mini28.4%estimated ± 7.4 pp, low confidence
87LFM2.5-24B-A2B27.7%measured
88GPT-5.6 Terra27.4%estimated ± 7.1 pp, low confidence
89Qwen3.8 Max27.3%measured
90GPT-5.426.5%estimated ± 7.1 pp, low confidence
91Claude Opus 526.0%measured
92GPT-5.525.7%estimated ± 7.1 pp, low confidence
93Ling 3.0 Flash Fin25.2%estimated ± 7.1 pp, low confidence
94DeepSeek V4 Flash 073125.1%measured
95Claude Fable 524.5%estimated ± 7.1 pp, low confidence
96Claude Sonnet 523.5%estimated ± 7.1 pp, low confidence
97LFM2.5-2.6B23.2%estimated ± 9.1 pp, low confidence
98Solar Pro 420.2%estimated ± 7.4 pp, low confidence
99Claude Opus 4.619.8%estimated ± 10.4 pp, low confidence
100Solar Open 218.2%estimated ± 7.4 pp, low confidence
101Claude Sonnet 4.618.0%estimated ± 10.4 pp, low confidence
102Muse Spark 1.111.0%estimated ± 10.4 pp, low confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General