List your company
ResearchFormerly WizWand

Complex Reasoning on LLaVA-Bench Wilder

74.3ScoreQwen3-VL-4B (Base)
51.4257.3663.369.24Jun 2, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Qwen3-VL-4B (Base)Base Model=Qwen3-VL-4B, Alignment Strategy=Base2026.0674.3
COPSD (Standard)Base Model=Qwen3-VL-4B, Alignment Strategy=COPSD (Standard)2026.0674.1
COPSD (Hybrid)Base Model=Qwen3-VL-4B, Alignment Strategy=COPSD (Hybrid)2026.0673.4
OPDBase Model=Qwen3-VL-4B, Alignment Strategy=OPD2026.0670.5
SFTBase Model=Qwen3-VL-4B, Alignment Strategy=SFT2026.0670.4
Qwen2.5-VL-7B (Base)Base Model=Qwen2.5-VL-7B, Alignment Strategy=Base2026.0669.9
COPSD (Standard)Base Model=Qwen2.5-VL-7B, Alignment Strategy=COPSD (Standard)2026.0669.4
COPSD (Hybrid)Base Model=Qwen2.5-VL-7B, Alignment Strategy=COPSD (Hybrid)2026.0668.5
GRPOBase Model=Qwen3-VL-4B, Alignment Strategy=GRPO2026.0668.2
OPDBase Model=Qwen2.5-VL-7B, Alignment Strategy=OPD2026.0667.2
GRPOBase Model=Qwen2.5-VL-7B, Alignment Strategy=GRPO2026.0667
SFTBase Model=Qwen2.5-VL-7B, Alignment Strategy=SFT2026.0666.7
Safe-RLHF-VBase Model=Qwen3-VL-4B, Alignment Strategy=Safe-RLHF-V2026.0656.8
Safe-RLHF-VBase Model=Qwen2.5-VL-7B, Alignment Strategy=Safe-RLHF-V2026.0652.3