| Dataset Name | SOTA Method | Metric | Trend | ||
|---|---|---|---|---|---|
| SPA-VL | SaFeR-ToolKit (+ SFT+DPO+GRPO) [3B] | Safety Score91.89 | 26 | 1mo ago | |
| ToolkitBench | Qwen2.5-VL-7B + SFT+DPO+GRPO | Safety Score2.49 | 22 | 1mo ago | |
| MSSBench | Qwen2.5-VL-3B + VLGuard | Safety Score2.55 | 22 | 1mo ago | |
| MM-SafetyBench | Qwen2.5-VL-3B + SFT+DPO+GRPO | Safety Score2.73 | 22 | 1mo ago | |
| BeaverTails-V | Qwen2.5-VL-7B + TIS | Safety Score2.9 | 22 | 1mo ago | |
| MM-SafetyBench SD + TYPO + SD_TYPO (test) | DefenSee | ASR Score0.08 | 8 | 1mo ago | |
| VLGuard (test) | LLaVAShield-7B | Accuracy86.78 | 6 | 1mo ago | |
| MM-SafetyBench | LLaVAShield-7B | Text-only Recall95.3 | 6 | 1mo ago | |
| multimodal safety dataset | ASR0.13 | 6 | 1mo ago | ||
| Image input safety evaluation set | gpt-5-thinking-nano | Hate Safety Acc98.6 | 4 | 1mo ago | |
| ChatGPT image input safety evaluations | Hate Safety98.9 | 4 | 1mo ago | ||
| MM-SafeBench | Forbidden Statements ASR1.04 | 4 | 1mo ago | ||
| SafeBench | FS ASR3.26 | 4 | 1mo ago | ||
| GOAT (test) | OSGA | Misogyny Accuracy56.9 | 2 | 1mo ago |
