Share your thoughts, 1 month free Claude Pro on usSee more

Spatial Reasoning (Multi-Image) on MMSI-Bench

45.2Accuracy

Gemini-3-Pro-Preview

Updated 1mo ago

Evaluation Results

Method	Links
Gemini-3-Pro-Preview 2026.03		45.2
GPT-5 2026.03		41.8
Gemini-2.5-Pro 2026.03		38
ViewFusion 2026.03		35.4
VST-7B-RL 2026.03		34.8
ViewFusion 2026.03		32.4
VST-3B-RL 2026.03		32
ViLAVT-7B 2026.02		31.3
Qwen3-VL-8B-Instruct 2026.03		31.1
VILASR-7B 2026.02		30.2
ViLaSR-7B 2026.03		30.2
Qwen3-VL-4B-Instruct 2026.03		30.1
Qwen3-VL-2B-Instruct 2026.03		28.9
SpaceR-7B 2026.02		28.8
Qwen2.5-VL-3B-Instruct 2026.03		28.6
InternVL3-8B 2026.03		28
Spatial-MLLM-4B 2026.02		27.7
SpatialLadder-3B 2026.03		27.4
SpaceR-7B 2026.03		27.4
Qwen2.5-VL-7B 2026.02		27.1
Qwen2.5-VL-7B 2026.02		26.9
Qwen2.5-VL-7B-Instruct 2026.03		26.8
InternVL3-2B 2026.03		26.5
Spatial-MLLM-4B 2026.03		26.1
Cambrian-S-7B 2026.03		25.8
InternVL3-8B 2026.02		25.7
Cambrian-S-3B 2026.03		25.2
RandomChoice 2026.03		25
LLaVA-OneVision-7B 2026.02		24.5