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

Multi-image Spatial Reasoning on MindCube-Tiny (test)

69.2Aro Accuracy

GPT-5.2

Updated 1mo ago

Evaluation Results

Method	Links
GPT-5.2 2026.02		69.2	95.5	41.5	58.4
HATCH 2026.02		65.6	31	50.2	50.2
CAMCUE 2026.02		63.2	75	31.67	47.43
GPT-4.1 2026.02		59.6	49.5	47.2	50.6
InternVL-2.5-4B 2026.02		57.6	33.5	45.2	45.9
SpatialLadder-3B 2026.02		56.4	31.5	47.8	46.8
Spatial-MLLM-4B 2026.02		52	29	34.5	37.6
Gemini-3-Pro 2026.02		51.6	54.5	46	49
LLaVA-OneVision-4B 2026.02		50	33.5	39.2	40.7
Qwen2.5-VL-3B 2026.02		46.4	34.5	35.3	37.8
Qwen2.5-VL-72B 2026.02		44.8	42	41.2	42.2
Video-R1 2026.02		44	30	38.5	38.2
Qwen2.5-VL-32B 2026.02		43.6	41.5	38.2	40
Qwen2.5-VL-7B 2026.02		38	33.5	32.7	34.1
SpaceR-7B 2026.02		31.6	31.5	28.7	29.9
InternVL-2.5-8B 2026.02		25.6	37	20.3	24.8
Qwen2.5-VL-7B 2026.02		21.4	38.7	29.5	29.3