Video Physical Reasoning on MASS-Bench 1.0 (test) benchmark leaderboard · SOTA2 Research

Video Physical Reasoning on MASS-Bench 1.0 (test)

65.82Spatial Understanding (SU) AccuracyGemini-2.5-Flash

Evaluation Results

Method	Links
Gemini-2.5-Flash2025.11		65.82	59.65	68.57	65.15	43.8	63.18
Qwen2.5-VL-7B + MASS2025.11		63.52	63.06	64.07	65.06	45.79	61.93
LLaVA-OneVision-7B + MASS2025.11		61.26	65.41	58.15	61.45	42.37	59.24
Qwen3-VL-32B2025.11		58.87	50.59	48.05	59.84	38.63	53.44
Video-R12025.11		57.49	55.19	47.65	63.53	33.91	53.08
Qwen3-VL-8B2025.11		57.29	50.77	51.92	58.65	37.07	53.23
Qwen2.5-VL-7B2025.11		55.87	53.03	54.89	52.36	30.21	52.3
GPT-4o2025.11		54.47	55.13	55.62	60.15	31.32	52.81
InternVL3.5-38B-Flash2025.11		52.19	49.88	43.29	61.45	32.71	48.71
VideoChat-R12025.11		51.2	48.66	47.34	52.3	31.09	47.05
LLaVA-OneVision-7B2025.11		49.18	51.99	38.12	56.02	30.46	45.79
InternVL3-8B2025.11		48.2	49.89	39.29	57.2	32.47	45.59
InternVL3.5-14B-Flash2025.11		47.68	49.23	41.96	57.52	31.03	45.71
InternVL3.5-8B-Flash2025.11		47.22	45.25	39.4	58.65	32.76	44.68
VideoLLaVA2025.11		36.39	31.76	21.01	38.55	17.76	30.72