Multi-Image Reasoning on VideoEspresso benchmark leaderboard · SOTA2 Research

Multi-Image Reasoning on VideoEspresso

50.6Narration ScoreROVER-LSW

Evaluation Results

Method	Links
ROVER-LSW2026.05		50.6	45.9	42.9	33.8	45.9	43.1	37.5	42.9	41.9	26.3	36.9	49.1	56.6	44	42.7
ROVER-LSW2026.05		49.4	45.2	41.8	33.6	44.3	41.3	36.1	41.3	40.3	24.6	35.4	47.2	53.3	40	41
VideoEspresso2026.05		45.2	27	33.7	26.1	39.3	36.7	55.6	41.3	30.6	29.8	30.8	35.8	20	26	34.1
LongVA-DPO2026.05		35.5	14.9	16.3	19	34.4	22	37.5	23.8	29	22.8	20	37.7	16.7	12	24.4
Qwen-VL-Max2026.05		33.9	22.4	23.5	21.4	26.2	30.3	41.7	30.2	27.4	26.3	20	20.8	16.7	24	26
InternVL22026.05		33.9	24.1	27.6	24.4	42.6	33	45.8	28.6	19.4	22.8	21.5	34	20	24	28.7
GPT-4o2026.05		32.3	16.7	25.5	22.8	32.8	27.5	37.5	28.6	24.2	19.3	30.8	30.2	20	22	26.4
LLaVA-1.52026.05		32.3	21.3	19.4	17.1	26.2	20.2	36.1	33.3	21	21.1	20	35.8	16.7	18	24.2
LLaVA-N-Video2026.05		31.2	20.2	16.2	17.6	36.5	32.7	30.6	24.5	26.4	24.5	34.7	20.8	20.3	17	25.2
mPLUG-Owl32026.05		30.6	23.6	20.4	22.3	37.7	29.4	48.6	34.9	30.6	24.6	27.7	24.5	13.3	24	28
Qwen2-VL2026.05		27.4	23	24.5	23.5	29.5	31.2	47.2	31.7	22.6	28.1	40	22.6	30	18	28.5
LLaVA-N-Inter2026.05		24.2	23.6	26.5	19.2	31.1	32.1	31.9	17.5	24.2	21.1	26.2	30.2	13.3	20	24.4