Long Audio-Video Question Answering on WorldSense benchmark leaderboard · SOTA2 Research

Long Audio-Video Question Answering on WorldSense

61.2Average AccuracyOmniAgent

Evaluation Results

Method	Links
OmniAgent2025.12		61.2	64.3	66.3	59.4	63.1	62.2	59.2	55.8	60.3
video-SALMONN 2+2025.12		56.5	59	63.1	54	59.9	58.1	54.1	51.9	54.4
Qwen3-Omni2025.12		54	—	—	—	—	—	—	—	—
Gemini 2.5 Flash2025.12		50.9	55.1	48.2	53	48.8	56.2	47.2	46.3	50
Gemini 1.5 Pro2025.12		48	53.7	47.2	50.3	50.4	52.4	46.8	40.2	42
Qwen2.5-Omni2025.12		45.4	47.8	49.8	43.6	43.8	48.3	39.1	43.5	47.3
GPT-4o2025.12		42.6	48	44	38.3	43.5	41.9	41.2	42.6	42.7
Gemini 2.5-Flash + RAG + Agent2026.02		39.84	57.89	30.77	42.59	33.33	31.82	41.18	25	45.95
Qwen2.5-Omni-7B + RAG + Agent + RL2026.02		38.28	44.74	23.08	40.74	40	40.91	41.18	25	45.95
Gemini 2.5-Flash + RAG2026.02		37.5	55.26	38.46	29.63	36.67	31.82	41.18	21.88	45.95
Qwen2.5-Omni-3B + RAG + Agent + RL2026.02		36.71	36.84	30.77	40.74	40	36.36	35.29	15.62	51.35
Qwen2.5-Omni-7B + RAG + Agent2026.02		34.38	31.58	50	25.93	30	50	35.29	21.88	43.24
Qwen2.5-Omni-3B + RAG + Agent2026.02		33.98	36.84	26.92	31.48	26.67	27.27	29.41	40.62	45.95
Gemini 2.5-Flash2026.02		33.59	47.37	23.08	27.78	40	27.27	29.41	25	43.24
Qwen2.5-Omni-7B + RAG2026.02		32.81	21.05	42.31	37.04	40	13.64	29.41	25	45.95
Qwen2.5-Omni-3B + RAG2026.02		31.25	34.21	26.92	25.93	30	31.82	41.18	31.25	35.14
Qwen2.5-Omni-7B2026.02		30.47	34.21	30.77	29.63	33.33	22.73	29.41	18.75	40.54
Qwen2.5-Omni-3B2026.02		29.69	34.21	42.32	27.78	30	13.64	29.41	15.62	40.54