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

Response Harmfulness Detection on XSTEST-RESP

95.48Response Harmfulness F1

GuardReasoner-Omni 4B

Updated 1mo ago

Evaluation Results

Method	Links
GuardReasoner-Omni 4B 2026.02		95.48
GPT-4 2024.06		95.4
WILDGUARD 2024.06		94.7
WildGuard 7B 2026.02		94.7
GuardReasoner 8B 2026.02		94.34
Qwen3Guard Gen 8B 2026.02		93.67
WILDGUARD 2024.06		93.4
GuardReasoner-Omni 2B 2026.02		92.9
GuardReasoner-VL 7B 2026.02		92.72
GuardReasoner-VL 3B 2026.02		92.5
Qwen3Guard Gen 8B 2026.02		92.12
GPT-4 2024.06		91.3
Llama-Guard2 2024.06		90.8
MD-Judge 7B 2026.02		90.4
LLaMA Guard 4 12B 2026.02		89.8
LLaMA Guard 3 8B 2026.02		87.67
Llama-Guard2 2024.06		85
BeaverDam 7B 2026.02		83.6
Llama-Guard 2024.06		82
Llama-Guard 2024.06		81.2
LibrAI-LongFormer-ref 2024.06		80.5
ShieldGemma 9B 2026.02		73.86
HarmBench Mistral 7B 2026.02		72
OAI Mod. API 2024.06		70.5
Keyword-based detector 2024.06		70.2
PolyGuard-Qwen 7B 2026.02		64.57
HarmBench LLaMA 13B 2026.02		64.5
Aegis-Guard-P 2024.06		60.4
Aegis Guard Permissive 7B 2026.02		60.4
Aegis-Guard-D 2024.06		52.8
Aegis Guard Defensive 7B 2026.02		52.8
Aegis-Guard-P 2024.06		49
OAI Mod. API 2024.06		46.6
Aegis-Guard-D 2024.06		36.9