benchgap
Agentic · tools

SkillsBench leaderboard

As of 2026-11-18, the highest measured score on SkillsBench is 70.2% by Qwen3.8 Max. 66 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Qwen3.8 Max70.2%measured
2Ling 3.1 Flash68.7%measured
3Atria Dawn Preview66.4%measured
4GPT-5.6 Sol66.0%estimated ± 4.1 pp, medium confidence
5Kimi K3.1 Pro65.9%estimated ± 4.1 pp, medium confidence
6GPT-6 Astra65.1%estimated ± 4.1 pp, medium confidence
7Kimi K364.7%estimated ± 4.1 pp, medium confidence
8Claude Opus 564.2%estimated ± 4.1 pp, medium confidence
9Inkling 2 Small63.4%estimated ± 4.1 pp, medium confidence
10LongCat-2.563.3%measured
11GPT-5.5 Pro63.3%estimated ± 4.1 pp, medium confidence
12Hy4 preview62.9%measured
13Muse Spark Open62.6%estimated ± 9.1 pp, low confidence
14GPT-5.4 Pro62.2%estimated ± 4.1 pp, medium confidence
15Muse Spark 1.162.1%estimated ± 9.1 pp, low confidence
16Step 5 Preview61.5%estimated ± 4.1 pp, medium confidence
17Claude Mythos 560.6%estimated ± 4.1 pp, medium confidence
18GPT-5.6 Terra60.0%estimated ± 4.1 pp, medium confidence
19Gemini 3.5 Flash59.9%estimated ± 9.1 pp, low confidence
20ERNIE 5.5 Flash Preview59.9%estimated ± 4.1 pp, medium confidence
21Ornith-1.5-397B58.9%estimated ± 4.1 pp, medium confidence
22Claude Sonnet 556.7%estimated ± 4.1 pp, medium confidence
23Mistral Large 456.4%estimated ± 4.1 pp, medium confidence
24GLM-5.256.4%estimated ± 9.1 pp, low confidence
25GPT-5.556.3%estimated ± 4.1 pp, medium confidence
26Claude Opus 4.856.2%estimated ± 4.1 pp, medium confidence
27Qwen3.7 Max56.2%estimated ± 9.1 pp, low confidence
28Kimi K2.7 Code56.0%estimated ± 9.1 pp, low confidence
29Claude Opus 4.655.5%estimated ± 4.1 pp, medium confidence
30MiniMax M355.3%estimated ± 4.1 pp, medium confidence
31DeepSeek V4 Pro55.2%estimated ± 4.1 pp, medium confidence
32GPT-5.6 Luna55.0%estimated ± 4.1 pp, medium confidence
33Kimi K2.654.9%estimated ± 4.1 pp, medium confidence
34Qwen3.7 Plus54.5%estimated ± 9.1 pp, low confidence
35GPT-5.454.4%estimated ± 4.1 pp, medium confidence
36Ripple-65B Preview53.6%estimated ± 9.1 pp, low confidence
37Sage-A1.553.2%estimated ± 4.1 pp, medium confidence
38dots3-note Preview52.8%measured
39Ling 3.0 Flash Sante51.6%estimated ± 4.1 pp, medium confidence
40Claude Opus 4.551.6%estimated ± 4.9 pp, medium confidence
41Claude Opus 4.750.6%estimated ± 4.1 pp, medium confidence
42Nemotron 3.5 Super VL 120B A12B49.3%estimated ± 4.1 pp, medium confidence
43Nemotron 3.5 Super VL 120B A12B NVFP448.8%estimated ± 4.1 pp, medium confidence
44Inkling-Small48.7%estimated ± 4.1 pp, medium confidence
45Qwen3.6-35B-A3B48.6%estimated ± 9.1 pp, low confidence
46Inkling48.3%estimated ± 4.1 pp, medium confidence
47Beam 1.0 501B-A23B48.0%estimated ± 4.1 pp, medium confidence
48Beam 1.0 501B-A23B FP847.9%estimated ± 4.1 pp, medium confidence
49Solar Pro 447.8%estimated ± 9.1 pp, low confidence
50Step 3.7 Flash47.0%estimated ± 4.1 pp, medium confidence
51Beam 1.0 501B-A23B NVFP446.8%estimated ± 4.1 pp, medium confidence
52Agents-A146.7%estimated ± 4.1 pp, medium confidence
53Qwen3.6 Plus46.7%estimated ± 4.9 pp, medium confidence
54Solar Open 245.9%estimated ± 9.1 pp, low confidence
55GPT-5.4 mini45.6%estimated ± 9.1 pp, low confidence
56Ling 3.0 Flash44.8%measured
57GPT-5.4 nano44.6%estimated ± 9.1 pp, low confidence
58DeepSeek V4 Flash 073144.5%estimated ± 4.1 pp, medium confidence
59Muse Glimmer 30B44.3%measured
60Nemotron 3.5 Bolt 74B-A9B BF1641.5%estimated ± 4.1 pp, low confidence
61Qwen3.5 397B40.6%estimated ± 4.1 pp, low confidence
62GLM-5.139.7%estimated ± 4.1 pp, low confidence
63Ornith-1.5-35B-A3B39.3%estimated ± 4.1 pp, low confidence
64Agents-A1-4B38.6%estimated ± 4.1 pp, low confidence
65Nemotron 3.5 Bolt 74B-A9B38.6%estimated ± 4.1 pp, low confidence
66GPT-5.237.7%estimated ± 4.1 pp, low confidence
67Qwen3.5-122B-A10B36.1%estimated ± 4.1 pp, low confidence
68GLM-534.6%estimated ± 4.1 pp, low confidence
69Qwen3.5-27B33.8%estimated ± 4.1 pp, low confidence
70Qwen3.5-35B-A3B33.8%estimated ± 4.1 pp, low confidence
71Kimi K2.533.5%estimated ± 4.1 pp, low confidence
72Ornith-1.5-9B30.2%estimated ± 4.1 pp, low confidence
73Spark-X2.6-4B27.8%estimated ± 4.1 pp, low confidence
74GLM-4.727.0%estimated ± 4.1 pp, low confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General