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

Content Moderation

Benchmarks

Dataset Name	SOTA Method	Metric
OpenAI Content Moderation	DPR	Average F1 Score83.1	30	3mo ago
MMDS (test)	LLaVAShield-7B	Accuracy95.76	27	4mo ago
Dynamic Evaluation Sets Adversarial 1.0	Yuvion-32B (Agent)	Overall Composite Score86.72	22	26d ago
Aegis 2.0	Qwen3Guard-4B-Gen-strict	F1 Score86.1	21	3mo ago
Multi-category Chinese content moderation dataset 1.0 (test)	CARO (Llama3-8B)	Politics Accuracy89.7	15	3mo ago
SafeWatch-Real (val)		Sexual Accuracy95.2	14	2mo ago
ProGuard	Yuvion VL	Score (%)95.4	12	1mo ago
Fine-grained Moderation Dataset	CHAIRO (Ours)	Average F189.2	11	3mo ago
UnsafeBench Sexual category (test)	KidsNanny (Stage 1+2)	Accuracy81.4	8	4mo ago
AIR-Bench Text + Image (test)	Aetheria	Precision83	8	4mo ago
AIR-Bench Image Only (test)		Precision94	8	4mo ago
AIR-Bench Text Only (test)		Precision94	8	4mo ago
Hateful Memes seen (test)	PALI-X-VPD (specialist w/ CoT)	AUC-ROC89.2	7	4mo ago
Safety Evaluation Set Moderation (held-out target labels)	Safety Monitor (In-model)	AUROC0.89	6	2mo ago
ILION-Bench v2 (test)	ILION Gate	Accuracy86.1	4	4mo ago
Production Traffic (live traffic)	Two-Stage Cascade	Refusal Rate (Prod)3.6	4	4mo ago
Laion5B UnsafeBench (test)	GGuard	Hate72	4	4mo ago
Content Moderation Korean (test)	CulturePark	Abusive Rate64.7	4	4mo ago
Business Content Moderation (evaluation set)	Xuanwu VL-2B	Ad Recall99.38	3	3mo ago
Content Moderation Parallel Fan-Out 4/5 Down M6	ReAct	LLM Calls10	3	4mo ago
Content Moderation Parallel Fan-Out Cascading M5	ReAct	LLM Calls9	3	4mo ago
Content Moderation Parallel Fan-Out Multi Down M4	ReAct	LLM Calls9	3	4mo ago
Content Moderation Parallel Fan-Out Toxicity Risk M3	ReAct	LLM Calls6	3	4mo ago
Content Moderation Parallel Fan-Out M2: Image Down	ReAct	LLM Calls7	3	4mo ago
Content Moderation Parallel Fan-Out M1: Happy Path	ReAct	LLM Calls6	3	4mo ago

Showing 25 of 33 rows