List your company
ResearchFormerly WizWand

Spatial Understanding on MMVP

77AccuracyQwen2.5-VL
41.32850.58959.8569.111Jan 15, 2026
Updated 6mo ago

Evaluation results

MethodLinks
Qwen2.5-VL2026.0177—
LLaVA-OneVision2026.0176.7—
Molmo2026.0175.3—
Gemma3-4b-it2026.0170.8—
SmolVLM22026.0168.7—
LLaVA-NeXT2026.0166.7—
PaliGemma2026.0166.7—
LLaVA v1.52026.0157.7—
LLaVA-AIMv2-2D-RoPEencoder=AIMv2, positional_embedding=2D-RoPE2026.0156—
LLaVA-2D-RoPEpositional_embedding=2D-RoPE2026.0151.3—
LLaVA-AIMv2encoder=AIMv22026.0151.3—
LLaVA-SigLIP-2D-RoPEencoder=SigLIP, positional_embedding=2D-RoPE2026.0150.7—
LLaVA-SigLIP2-2D-RoPEencoder=SigLIP2, positional_embedding=2D-RoPE2026.0148—
LLaVA-SigLIPencoder=SigLIP2026.0143.3—
LLaVA-SigLIP2encoder=SigLIP22026.0142.7—
CoT promptingBackbone=Qwen2.5-VL-7B-Instruct, Model Scale=7B, Reasoning Strategy=CoT prompting2026.04—72.44
CoT promptingBackbone=Qwen2.5-VL-3B-Instruct, Model Scale=3B, Reasoning Strategy=CoT prompting2026.04—63.67
FGRPOBackbone=Qwen2.5-VL-7B-Instruct, Model Scale=7B, Reasoning Strategy=Faithful GRPO2026.04—73.33
FGRPOBackbone=Qwen2.5-VL-3B-Instruct, Model Scale=3B, Reasoning Strategy=Faithful GRPO2026.04—64.3
GPT-4o (CoT)Reasoning Strategy=CoT2026.04—84.33
GPT-5-nano (CoT)Reasoning Strategy=CoT2026.04—75.67
GRPO-TBackbone=Qwen2.5-VL-7B-Instruct, Model Scale=7B, Reasoning Strategy=RL (Task Reward Only)2026.04—74
GRPO-TBackbone=Qwen2.5-VL-3B-Instruct, Model Scale=3B, Reasoning Strategy=RL (Task Reward Only)2026.04—65.22
Non-reasoningBackbone=Qwen2.5-VL-7B-Instruct, Model Scale=7B, Reasoning Strategy=Non-reasoning2026.04—75.78
Non-reasoningBackbone=Qwen2.5-VL-3B-Instruct, Model Scale=3B, Reasoning Strategy=Non-reasoning2026.04—64.67
R1-OneVisionReasoning Strategy=MRM baseline2026.04—56.16
TreeVGRReasoning Strategy=MRM baseline2026.04—71
ViGoRL-SpatialReasoning Strategy=MRM baseline2026.04—73.22
Vision-R1Reasoning Strategy=MRM baseline2026.04—72.22
VL-RethinkerReasoning Strategy=MRM baseline2026.04—75.89