EASI Leaderboard

EASI: Holistic Evaluation and Analysis for Spatial Intelligence Made Easy

EASI conceptualizes a comprehensive taxonomy of spatial tasks that unifies existing benchmarks and a standardized protocol for the fair evaluation of state-of-the-art proprietary and open-source models.


MLL-Lab/MindCube-Qwen2.5VL-Plain-CGMap-FFR-Out-SFT	41.29	55.43	45.83	31.1	29.42	39.37


Qwen/Qwen3-8B	41.29	57.9	45.83	31.1	29.42	42.2
Qwen/Qwen3-4B	38.8	55.43	45.43	29.9	23.85	39.37
OpenGVLab/InternVL3-8B	38.3	42.14	41.15	28	41.54	38.66
Qwen/Qwen3-VL-2B-Instruct	37.28	50.36	35.66	28.9	34.52	36.97
inclusionAI/ViLaSR	36.88	44.63	38.74	30.2	35.1	35.71
hongxingli/SpatialLadder-3B	36.71	44.86	27.96	27.4	43.46	39.85
RUBBISHLIKE/SpaceR-SFT-7B	34.36	36.29	34.27	27.4	37.98	35.85
Qwen/Qwen2.5-7B	33.93	32.3	37.64	26.8	36.05	36.85
MLL-Lab/MindCube-Qwen2.5VL-Plain-CGMap-FFR-Out-SFT	32.03	29.93	28.74	30.4	39.9	31.2
OpenGVLab/InternVL3_5-2B	31.91	32.98	30.03	26.5	37.5	32.56
MLL-Lab/MindCube-Qwen2.5VL-Aug-CGMap-FFR-Out-SFT	31.78	29.6	28.23	29.1	41.06	30.9
Qwen/Qwen2.5-3B	31.66	27	33.14	28.6	37.6	31.97
Diankun/Spatial-MLLM-subset-sft	30.06	45.98	10.11	26.1	33.46	34.66
ByteDance-Seed/BAGEL-7B-MoT	28.01	30.9	null	33.1	34.71	41.32
MLL-Lab/MindCube-Qwen2.5VL-RawQA-SFT	20.22	17.24	6.31	1.7	51.73	24.14