List your company
ResearchFormerly WizWand

Compositional Reasoning on VisMin Easy (test)

94.88Text ScoreQwen3-VL
18.02437.97757.9377.883Aug 19, 2025
Updated 4mo ago

Evaluation results

MethodLinks
Qwen3-VLModel Size=≥7B, Evaluation protocol=Frozen state2025.0894.8893.2490.34
Qwen3-VL + FOCUSModel Size=≥7B, Evaluation protocol=Frozen state2025.0894.7993.590.51
InternVL3 + FOCUSModel Size=≥7B, Evaluation protocol=Frozen state2025.0894.783.0580.83
InternVL3Model Size=≥7B, Evaluation protocol=Frozen state2025.0894.6284.8481.6
Qwen3-VL + FOCUSModel Size=<7B, Evaluation protocol=Frozen state2025.0894.2892.0688.8
Qwen3-VLModel Size=<7B, Evaluation protocol=Frozen state2025.0894.290.6887.35
Qwen2.5-VLModel Size=≥7B, Evaluation protocol=Frozen state2025.0893.8592.6588.55
Qwen2.5-VL + FOCUSModel Size=≥7B, Evaluation protocol=Frozen state2025.0893.8592.5788.72
LLaVA-OVModel Size=≥7B, Evaluation protocol=Frozen state2025.0893.4362.9960.68
LLaVA-OV + FOCUSModel Size=≥7B, Evaluation protocol=Frozen state2025.0893.4364.1462
InternVL3 + FOCUSModel Size=<7B, Evaluation protocol=Frozen state2025.0893.3474.3371.26
InternVL3Model Size=<7B, Evaluation protocol=Frozen state2025.0893.2529.9629.36
Qwen2.5-VLModel Size=<7B, Evaluation protocol=Frozen state2025.0892.1447.7446.2
Qwen2.5-VL + FOCUSModel Size=<7B, Evaluation protocol=Frozen state2025.0892.0656.9255.82
LLaVA-OVModel Size=<7B, Evaluation protocol=Frozen state2025.0820.989.261.55
LLaVA-OV + FOCUSModel Size=<7B, Evaluation protocol=Frozen state2025.0820.9827.018.42