Allocentric Spatial Reasoning on COMFORT# benchmark leaderboard · SOTA2 Research

Allocentric Spatial Reasoning on COMFORT#

69Left/Right AccuracySymPL

Evaluation Results

Method	Links
SymPL2026.02		69	97.33	91.41	91.5
Qwen2.5-VL + SCAFFOLD2026.02		52.17	71.17	50.39	47.42
GPT-52026.02		49.83	84.25	54.22	49.83
Random2026.02		48.75	48.67	47.27	52.33
Qwen2.5-VL2026.02		48.17	72.33	51.17	51.33
LLaVA-OneVision2026.02		47.92	57.83	51.88	50.42
APC-Num2026.02		47.83	52.5	34.14	36.92
LLaVA-NeXT2026.02		47.33	51.58	51.8	52.42
SpatialVLM2026.02		46.83	63.67	50.78	49.58
Molmo2026.02		46.58	46.17	53.36	53.58
Qwen2.5-VL + SoM2026.02		46.58	67.25	51.88	46.42
SpatialBot2026.02		46.33	58.83	50.08	53.08
SD-VLM2026.02		45.83	45.17	48.91	52.25
APC-Vis2026.02		43.75	54.08	49.77	30.92
Qwen2.5-VL + CoT2026.02		43.25	70.75	50.39	44.67
SpatialRGPT2026.02		43.08	70.25	53.75	47.75
Cambrian-12026.02		41.17	80.42	50.7	41.33
Gemini-2.5-Flash2026.02		38.33	77.83	52.34	51.58
SAT2026.02		35	48.75	34.92	39.5