Scaling Spatial Intelligence with Multimodal Foundation Models

Benchmark Model Rank Results
spatial-reasoning-on-mindcube-tinyBAGEL-7B-MoT–Accuracy: 34.7
spatial-reasoning-on-mindcube-tinyCambrian-S-3B–Accuracy: 38.4
spatial-reasoning-on-mindcube-tinyCambrian-S-7B–Accuracy: 37.9
spatial-reasoning-on-mindcube-tinyGPT-5–Accuracy: 56.3
spatial-reasoning-on-mindcube-tinyGemini-2.5-Pro–Accuracy: 57.6
spatial-reasoning-on-mindcube-tinyGemini-3-Pro-Preview–Accuracy: 70.9
spatial-reasoning-on-mindcube-tinyGrok-4–Accuracy: 63.6
spatial-reasoning-on-mindcube-tinyInternVL3-2B–Accuracy: 37.5
spatial-reasoning-on-mindcube-tinyInternVL3-8B–Accuracy: 41.5
spatial-reasoning-on-mindcube-tinyMindCube-3B-RawQA-SFT–Accuracy: 51.7
spatial-reasoning-on-mindcube-tinyQwen2.5-VL-3B-Instruct–Accuracy: 37.6
spatial-reasoning-on-mindcube-tinyQwen2.5-VL-7B-Instruct–Accuracy: 36.0
spatial-reasoning-on-mindcube-tinyQwen3-VL-2B-Instruct–Accuracy: 34.5
spatial-reasoning-on-mindcube-tinyQwen3-VL-8B-Instruct–Accuracy: 29.4
spatial-reasoning-on-mindcube-tinySeed-1.6–Accuracy: 48.8
spatial-reasoning-on-mindcube-tinySenseNova-SI (BAGEL-7B-MoT)–Accuracy: 46.8
spatial-reasoning-on-mindcube-tinySenseNova-SI (InternVL3-2B)–Accuracy: 41.8
spatial-reasoning-on-mindcube-tinySenseNova-SI (InternVL3-8B)–Accuracy: 85.7
spatial-reasoning-on-mindcube-tinySenseNova-SI (Qwen3-VL-8B)–Accuracy: 73.8
spatial-reasoning-on-mindcube-tinySpaceR-7B–Accuracy: 38.0
spatial-reasoning-on-mindcube-tinySpatial-MLLM-4B–Accuracy: 33.5
spatial-reasoning-on-mindcube-tinySpatialLadder-3B–Accuracy: 43.5
spatial-reasoning-on-mindcube-tinyVST-3B-SFT–Accuracy: 36.0
spatial-reasoning-on-mindcube-tinyVST-7B-SFT–Accuracy: 39.7
spatial-reasoning-on-mindcube-tinyViLaSR-7B–Accuracy: 35.1
spatial-reasoning-on-mmsi-benchBAGEL-7B-MoT–Accuracy: 31.0
spatial-reasoning-on-mmsi-benchCambrian-S-3B–Accuracy: 27.0
spatial-reasoning-on-mmsi-benchCambrian-S-7B–Accuracy: 27.1
spatial-reasoning-on-mmsi-benchGemini-3-Pro-Preview–Accuracy: 45.2
spatial-reasoning-on-mmsi-benchGrok-4–Accuracy: 37.8
spatial-reasoning-on-mmsi-benchInternVL3-8B–Accuracy: 28.0
spatial-reasoning-on-mmsi-benchMindCube-3B-RawQA-SFT–Accuracy: 1.7
spatial-reasoning-on-mmsi-benchQwen3-VL-2B-Instruct–Accuracy: 28.9
spatial-reasoning-on-mmsi-benchQwen3-VL-8B-Instruct–Accuracy: 31.1
spatial-reasoning-on-mmsi-benchSeed-1.6–Accuracy: 38.3
spatial-reasoning-on-mmsi-benchSenseNova-SI (BAGEL-7B-MoT)–Accuracy: 34.5
spatial-reasoning-on-mmsi-benchSenseNova-SI (InternVL3-2B)–Accuracy: 34.2
spatial-reasoning-on-mmsi-benchSenseNova-SI (InternVL3-8B)–Accuracy: 43.3
spatial-reasoning-on-mmsi-benchSenseNova-SI (Qwen3-VL-8B)–Accuracy: 38.1
spatial-reasoning-on-mmsi-benchSpaceR-7B–Accuracy: 27.4
spatial-reasoning-on-mmsi-benchSpatial-MLLM-4B–Accuracy: 26.1
spatial-reasoning-on-mmsi-benchSpatialLadder-3B–Accuracy: 27.4
spatial-reasoning-on-mmsi-benchViLaSR-7B–Accuracy: 30.2
spatial-reasoning-on-viewspatial-benchBAGEL-7B-MoT–Overall: 41.3
spatial-reasoning-on-viewspatial-benchCambrian-S-3B–Overall: 41.0
spatial-reasoning-on-viewspatial-benchCambrian-S-7B–Overall: 41.3
spatial-reasoning-on-viewspatial-benchGPT-5–Overall: 45.6
spatial-reasoning-on-viewspatial-benchGemini-2.5-Pro–Overall: 46.1
spatial-reasoning-on-viewspatial-benchGemini-3-Pro-Preview–Overall: 50.4
spatial-reasoning-on-viewspatial-benchGrok-4–Overall: 43.2
spatial-reasoning-on-viewspatial-benchInternVL3-2B–Overall: 32.6
spatial-reasoning-on-viewspatial-benchInternVL3-8B–Overall: 38.7
spatial-reasoning-on-viewspatial-benchMindCube-3B-RawQA-SFT–Overall: 24.1
spatial-reasoning-on-viewspatial-benchQwen2.5-VL-3B-Instruct–Overall: 32.0
spatial-reasoning-on-viewspatial-benchQwen2.5-VL-7B-Instruct–Overall: 36.9
spatial-reasoning-on-viewspatial-benchQwen3-VL-2B-Instruct–Overall: 37.0
spatial-reasoning-on-viewspatial-benchQwen3-VL-8B-Instruct–Overall: 42.2
spatial-reasoning-on-viewspatial-benchSeed-1.6–Overall: 43.9
spatial-reasoning-on-viewspatial-benchSenseNova-SI (BAGEL-7B-MoT)–Overall: 46.9
spatial-reasoning-on-viewspatial-benchSenseNova-SI (InternVL3-2B)–Overall: 52.7
spatial-reasoning-on-viewspatial-benchSenseNova-SI (InternVL3-8B)–Overall: 54.7
spatial-reasoning-on-viewspatial-benchSenseNova-SI (Qwen3-VL-8B)–Overall: 51.2
spatial-reasoning-on-viewspatial-benchSpaceR-7B–Overall: 35.9
spatial-reasoning-on-viewspatial-benchSpatial-MLLM-4B–Overall: 34.7
spatial-reasoning-on-viewspatial-benchSpatialLadder-3B–Overall: 39.9
spatial-reasoning-on-viewspatial-benchVST-3B-SFT–Overall: 52.9
spatial-reasoning-on-viewspatial-benchVST-7B-SFT–Overall: 50.5
spatial-reasoning-on-viewspatial-benchViLaSR-7B–Overall: 35.7