MMSI-Bench: A Benchmark for Multi-Image Spatial Intelligence

Benchmark Model Rank Results
spatial-reasoning-on-mmsi-benchClaude-3.7-Sonnet–Accuracy: 30.2
spatial-reasoning-on-mmsi-benchDeepSeek-VL2–Accuracy: 27.1
spatial-reasoning-on-mmsi-benchDeepSeek-VL2-Small–Accuracy: 28.6
spatial-reasoning-on-mmsi-benchDeepSeek-VL2-Tiny–Accuracy: 24.0
spatial-reasoning-on-mmsi-benchDoubao-1.5-pro–Accuracy: 33.0
spatial-reasoning-on-mmsi-benchGLM-4.5V-thinking–Accuracy: 33.8
spatial-reasoning-on-mmsi-benchGPT-4.1–Accuracy: 30.9
spatial-reasoning-on-mmsi-benchGPT-4.5–Accuracy: 40.3
spatial-reasoning-on-mmsi-benchGPT-4o–Accuracy: 30.3
spatial-reasoning-on-mmsi-benchGPT-5–Accuracy: 41.9
spatial-reasoning-on-mmsi-benchGemini-2.5-Pro–Accuracy: 36.9
spatial-reasoning-on-mmsi-benchInternVL2.5-1B–Accuracy: 26.1
spatial-reasoning-on-mmsi-benchInternVL2.5-26B–Accuracy: 28.0
spatial-reasoning-on-mmsi-benchInternVL2.5-2B–Accuracy: 29.0
spatial-reasoning-on-mmsi-benchInternVL2.5-38B–Accuracy: 27.9
spatial-reasoning-on-mmsi-benchInternVL2.5-4B–Accuracy: 26.3
spatial-reasoning-on-mmsi-benchInternVL2.5-78B–Accuracy: 28.5
spatial-reasoning-on-mmsi-benchInternVL2.5-8B–Accuracy: 28.7
spatial-reasoning-on-mmsi-benchInternVL3-14B–Accuracy: 26.8
spatial-reasoning-on-mmsi-benchInternVL3-1B–Accuracy: 27.0
spatial-reasoning-on-mmsi-benchInternVL3-2B–Accuracy: 25.3
spatial-reasoning-on-mmsi-benchInternVL3-38B–Accuracy: 26.3
spatial-reasoning-on-mmsi-benchInternVL3-78B–Accuracy: 28.5
spatial-reasoning-on-mmsi-benchInternVL3-9B–Accuracy: 26.7
spatial-reasoning-on-mmsi-benchLLaVA-OneVision-72B–Accuracy: 28.4
spatial-reasoning-on-mmsi-benchLLaVA-OneVision-7B–Accuracy: 24.5
spatial-reasoning-on-mmsi-benchLlama-3.2-11B-Vision–Accuracy: 25.4
spatial-reasoning-on-mmsi-benchNVILA-15B–Accuracy: 30.5
spatial-reasoning-on-mmsi-benchNVILA-8B–Accuracy: 28.1
spatial-reasoning-on-mmsi-benchQwen2.5-VL-32B–Accuracy: 27.7
spatial-reasoning-on-mmsi-benchQwen2.5-VL-3B–Accuracy: 26.5
spatial-reasoning-on-mmsi-benchQwen2.5-VL-72B–Accuracy: 30.7
spatial-reasoning-on-mmsi-benchQwen2.5-VL-7B–Accuracy: 25.9
spatial-reasoning-on-mmsi-benchSeed1.5-VL–Accuracy: 29.7
spatial-reasoning-on-mmsi-benchdots.vlm 1–Accuracy: 34.1
spatial-reasoning-on-mmsi-bencho3–Accuracy: 41.0