Cross-Video Reasoning on CrossVid benchmark leaderboard · SOTA2 Research

Cross-Video Reasoning on CrossVid

85.6BU ScoreHuman

Evaluation Results

Method	Links
Human2026.05		85.6	92.3	90.7	83.9	88.13	91.6	85.2	89.9	88.9	93.2	94.2	93.7	85.2	89.18
Gemini-2.5-Pro2026.05		54.2	51.8	68.7	36.4	52.78	76.5	13.4	78.2	56.03	32	25.3	28.65	59.8	49.63
Doubao-1.5-VL-pro2026.05		51.2	58.1	69.5	36.4	53.8	66.9	4.6	36.8	36.1	37.4	32	34.7	50.1	44.3
Qwen3-VL-32B2026.05		46.75	37.5	59.54	39.77	45.89	73	4.7	7.5	28.4	26.62	36.56	31.59	47.51	38.37
GPT-4.12026.05		46.2	34.6	58.5	51.2	47.63	70.9	8.6	60.5	46.67	38.6	38.2	38.4	44.6	45.19
AgentCVR-8B2026.05		33.88	43.65	70.71	41.23	47.35	67.43	16.63	30.41	38.68	37.18	33.2	35.19	41.85	42.03
Qwen2.5-VL-32B2026.05		31.4	30.5	48.6	39.7	38.3	65.7	5.2	8.7	26.53	23.7	39.6	31.65	41.2	33.78
VCA-8B2026.05		27.5	30.8	39.2	32.8	32.53	64.6	8.3	21.6	31.5	34.4	26.9	30.65	33.6	30.71
AgentCVR-4B2026.05		26.41	38.71	51.66	32.61	35.35	62.31	12.31	19.42	31.35	28.32	31.43	29.88	26.43	32.06
VideoAgent-8B2026.05		24.31	31.42	42.16	31.73	32.41	67.2	5.93	17.1	30.08	32.7	28.5	30.6	30.65	30.01
Qwen3-VL-8B2026.05		23.48	27.35	40.72	34.94	31.6	62.54	5.71	7.22	25.16	34.51	27.15	30.83	31.18	29.47
Qwen3-VL-4B2026.05		17.93	21.47	26.62	28.61	23.66	49.89	3.13	2.97	18.66	24.94	28.71	26.83	11.59	21.59