List your company

Safety Evaluation on MM-SafetyBench (MMSB)

1.8Attack Success Rate (V-T AVG)Refusal Steering
-1.219.0539.359.55May 18, 2026

Evaluation results

MethodLinks
Refusal SteeringBase Model=Qwen2.5-Omni-3B2026.051.8
ReGapBase Model=Qwen2.5-Omni-3B2026.0514.2
Refusal SteeringBase Model=Qwen3-Omni-30B-A3B-Instruct2026.0521
Refusal SteeringBase Model=MiniCPM-o-4.52026.0523.6
ReGapBase Model=Qwen3-Omni-30B-A3B-Instruct2026.0536.3
BaseBase Model=Qwen3-Omni-30B-A3B-Instruct2026.0558.1
ReGapBase Model=MiniCPM-o-4.52026.0565.4
BaseBase Model=MiniCPM-o-4.52026.0569.8
BaseBase Model=Qwen2.5-Omni-3B2026.0576.8