Skip to content

Benchmark

WildClawBench · Total Cost

Lower scores are better.

Models and systems measured
33
Scores recorded
33
Sources
1
IQ weight
No weight published

Score distribution

27.5
Mean
18
Median
26.4
Std dev
6.1
10th percentile
77.4
90th percentile
0
Min
096
The band covers the middle 80% of scores. The line marks the median.

Results

The source marks these results as descriptive, not capability scores, so we list them by name instead of ranking them.

Model or systemScore
Claude Fable 5OpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs87.7
Claude Opus 4.6OpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs81
Claude Opus 4.7OpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs77.4
Claude Opus 4.8OpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs96
DeepSeek V3.2OpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs11.4
DeepSeek V4 Pro 0813OpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs12
Gemini 3.1 Pro PreviewOpenClaw · full 60-task suite · low-effort· Reported by WildClawBench / InternLM· via BenchBoard docs18
Gemma 4 31B ITOpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs3.5
GLM 5OpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs11.4
GLM 5 TurboOpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs15
GLM 5.1OpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs34.8
GLM 5.2OpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs17.1
GPT 5.4OpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs19.8
GPT 5.5OpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs37.8
GPT-5.6 SolOpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs56.8
Grok 4.20 BetaOpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs9.6
Grok 4.5OpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs28.4
Hy3 (Deep Infra)OpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs2.1
Intern-S2 Preview 397BOpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs0
Kimi K2.5OpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs6.6
Kimi K2.7 CodeOpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs72.3
Kimi K3OpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs40.1
MiMo V2 FlashOpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs10.2
MiMo V2 ProOpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs26.4
MiMo V2.5 ProOpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs12.6
MiniMax M2.5OpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs9.6
MiniMax M2.7OpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs7.2
Muse Glimmer 30BOpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs6.1
Muse Spark 1.1OpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs23.6
Qwen3.5 397B A17BOpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs22.2
Qwen3.6 27BOpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs20.9
Qwen3.8 MaxOpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs24.7
Step 3.5 FlashOpenClaw · full 60-task suite· Reported by WildClawBench / InternLM· via BenchBoard docs6.6

Data credits: BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links.