Share your thoughts, 1 month free Claude Pro on usSee more

Trajectory-level safety evaluation on ATBench (test)

0.928Accuracy

AgentDoG-Qwen3-4B

Updated 3mo ago

Evaluation Results

Method	Links
AgentDoG-Qwen3-4B 2026.01		0.928	0.905	0.956	0.93
GPT-5.2 2026.01		0.9	0.847	0.976	0.907
AgentDoG-Llama3.1-8B 2026.01		0.876	0.809	0.984	0.888
AgentDoG-Qwen2.5-7B 2026.01		0.874	0.821	0.956	0.884
Gemini-3-Pro 2026.01		0.872	0.927	0.808	0.863
Qwen3-235B-A22B-Instruct-2507 2026.01		0.846	0.994	0.696	0.819
ShieldAgent 2026.01		0.76	0.901	0.584	0.709
Gemini-3-Flash 2026.01		0.756	0.985	0.52	0.681
PolyGuard 2026.01		0.738	0.687	0.876	0.77
QwQ-32B 2026.01		0.63	0.933	0.28	0.431
Qwen3-4B-Instruct-2507 2026.01		0.616	0.641	0.528	0.579
Qwen2.5-7B-Instruct 2026.01		0.594	0.658	0.392	0.491
LlamaGuard4-12B 2026.01		0.581	1	0.164	0.282
JoySafety 2026.01		0.569	0.626	0.348	0.447
Qwen3-Guard 2026.01		0.553	1	0.108	0.195
LlamaGuard3-8B 2026.01		0.533	1	0.068	0.127
ShieldGemma-9B 2026.01		0.511	0.8	0.032	0.062
ShieldGemma-27B 2026.01		0.509	0.778	0.028	0.054
NemoGuard 2026.01		0.499	0.5	0.412	0.452
Llama3.1-8B-Instruct 2026.01		0.496	0.498	0.992	0.663