List your company

Safety Alignment Evaluation on VLSBench

100Safety ScoreGRPO
61.436871.448481.4691.4716Jun 2, 2026

Evaluation results

MethodLinks
GRPOBackbone=Qwen3-VL-4B, Alignment Method=GRPO2026.06100
COPSD (Hybrid)Backbone=Qwen3-VL-4B, Alignment Method=COPSD (Hybrid)2026.0699.51
Safe-RLHF-VBackbone=Qwen3-VL-4B, Alignment Method=Safe-RLHF-V2026.0699.33
GRPOBackbone=Qwen2.5-VL-7B, Alignment Method=GRPO2026.0698.93
COPSD (Standard)Backbone=Qwen3-VL-4B, Alignment Method=COPSD (Standard)2026.0698.75
Safe-RLHF-VBackbone=Qwen2.5-VL-7B, Alignment Method=Safe-RLHF-V2026.0698.53
COPSD (Standard)Backbone=Qwen2.5-VL-7B, Alignment Method=COPSD (Standard)2026.0697.72
COPSD (Hybrid)Backbone=Qwen2.5-VL-7B, Alignment Method=COPSD (Hybrid)2026.0694.82
OPDBackbone=Qwen2.5-VL-7B, Alignment Method=OPD2026.0690.36
SFTBackbone=Qwen2.5-VL-7B, Alignment Method=SFT2026.0690.14
SFTBackbone=Qwen3-VL-4B, Alignment Method=SFT2026.0679.92
OPDBackbone=Qwen3-VL-4B, Alignment Method=OPD2026.0679.61
Qwen3-VL-4BBackbone=Qwen3-VL-4B, Alignment Method=Base2026.0668.63
Qwen2.5-VL-7BBackbone=Qwen2.5-VL-7B, Alignment Method=Base2026.0662.92