Categories Companies Research

Loading the SOTA2 catalog…

Explore

Categories Companies Research

Contribute

Join SOTA2 Contact

SOTA2

© 2026 SOTA2

Build in Bellevue, USA

WizWand is now SOTA2 Research

Papers Benchmarks Datasets Tasks

Multimodal Safety Evaluation research benchmarks · SOTA2 Research

Multimodal Safety Evaluation

Benchmarks

Dataset NameSOTA MethodSort

Dataset NameSOTA methodMetricTrendResultsLast Updated

Qwen2.5-VL-3B + SFT+DPO+GRPO

2.73Safety Score

34

Jun 25, 2026

SaFeR-ToolKit (+ SFT+DPO+GRPO) [3B]

91.89Safety Score

26

Mar 4, 2026

Qwen2.5-VL-7B + SFT+DPO+GRPO

2.49Safety Score

22

Mar 4, 2026

Qwen2.5-VL-3B + VLGuard

2.55Safety Score

22

Mar 4, 2026

Qwen2.5-VL-7B + TIS

2.9Safety Score

22

Mar 4, 2026

0.21ASR (%)

9

Apr 28, 2026

5.39ASR

9

Apr 28, 2026

MM-SafetyBench SD + TYPO + SD_TYPO (test)

0.08ASR Score

8

Feb 26, 2026

86.78Accuracy

6

Mar 11, 2026

95.3Text-only Recall

6

Mar 11, 2026

multimodal safety dataset

Claude 3.5 Sonnet

0.13ASR

6

Feb 26, 2026

Image input safety evaluation set

gpt-5-thinking-nano

98.6Hate Safety Acc

4

Feb 26, 2026

ChatGPT image input safety evaluations

98.9Hate Safety

4

Feb 26, 2026

1.04Forbidden Statements ASR

4

Feb 26, 2026

Claude-3.5-Sonnet

3.26FS ASR

4

Feb 26, 2026

56.9Misogyny Accuracy

2

Feb 26, 2026

Popular datasets

MM-SafetyBench SafeBench MM-SafeBench BeaverTails SIUO

Follow for updates

Join SOTA2 Contact