EvasionBench

Benchmark for evaluating model robustness against evasion-style safety and policy-circumvention prompts.

5rows
scoreprimary metric
2026-05-06sampled

Metadata

ID
evasionbench
Category
Safety
Release
2026-01-14

Metrics

Score

Latest Results

Rows are ranked by the benchmark score column. Model display names are preserved from the OpenEvals source dataset.

Rank Subject Score Model Match Provenance Sampled
1 zai-org/GLM-4.7 82.91 GLM GLM 4.7
z-ai-glm-4.7
 		Imported 2026-05-06
2 Qwen/Qwen3-Coder-480B-A35B-Instruct 78.16 Qwen3 Coder 480B A35B
qwen-qwen3-coder
 		Imported 2026-05-06
3 MiniMaxAI/MiniMax-M2.1 71.31 — Imported 2026-05-06
4 deepseek-ai/DeepSeek-V3.2 66.88 DeepSeek V3.2
deepseek-deepseek-v3.2
 		Imported 2026-05-06
5 moonshotai/Kimi-K2-Instruct-0905 66.68 KIMI MoonshotAI: Kimi K2 0905
moonshotai-kimi-k2-0905
 		Imported 2026-05-06