Prompt Harmfulness Classification on Public Prompt Harmfulness Benchmarks Suite benchmark leaderboard · SOTA2 Research

Prompt Harmfulness Classification on Public Prompt Harmfulness Benchmarks Suite

81OAI ScoreNemoGuard

Evaluation Results

Method	Links
NemoGuard2026.05		81	—	86.8	98.5	75.2	84.6	81.6
Azure2026.03		80.6	57.6	46.3	67.1	5.3	48.1	31.7
ShieldGemma2026.05		80.5	—	71.6	84.4	57.3	69.6	54.3
ShieldGemma2026.03		79.2	67	72.2	80.8	59.8	68.4	51.4
Llama-Guard32026.03		79.1	53.7	76.4	88.4	98.9	78.9	77
OAI Mod. API2024.06		79	25.4	31.9	63	9.6	41.8	—
OpenAI Moderation2026.03		79	25.4	36.2	57.6	9.6	36.7	12.1
EXPGUARD2026.03		77.3	74.7	83.7	92.9	99.2	85.7	86.1
Llama-Guard22024.06		76.1	47.1	71.8	95.8	94	77	—
Llama-Guard22026.03		76.1	47.1	76	89.1	94	75.5	70.9
Llama-Guard2024.06		75.8	61.6	74.1	93	67.2	74.4	—
Llama-Guard2026.03		75.8	61.6	75.3	82.5	67.2	69.7	56
Aegis-Guard-P2024.06		74.7	73	82.9	99	70.5	80	—
Aegis-Guard-P2026.03		74.7	72.5	81.1	80.3	70.5	75.1	71.5
PolyGuard-Qwen2026.05		74.1	—	86.3	100	98.7	89.4	88.1
LlamaGuard42026.05		73.5	—	70.6	98	97.2	82.5	73
WILDGUARD2024.06		72.1	70.8	89.4	99.5	98.9	86.1	—
WildGuard2026.05		72.1	—	80.7	99.5	98.9	88	88.9
WildGuard2026.03		72	70.7	80.8	94.5	98.7	84.2	88.7
GPT-42024.06		70.5	68.3	84.4	100	100	84.6	—
Perspective2026.03		70.5	28.5	33.6	47.3	9.6	32.3	4.1
GLiGuard2026.05		69	—	85.2	97.7	99	87.7	87.5
Qwen3Guard-Gen2026.05		68.8	—	86.1	99.5	100	88.7	88.9
Detoxify2026.03		68.4	31.9	32	28.5	4.9	27.9	1.8
Aegis-Guard-D2024.06		67.5	70	84.8	100	77.7	80	—
Aegis-Guard-D2026.03		67.5	70.2	81.2	77.7	77.7	75.5	78.5