Visual Question Answering on HD-EPIC benchmark leaderboard · SOTA2 Research

Visual Question Answering on HD-EPIC

65Recipe AccuracyOurs w/o Post-Processing

Evaluation Results

Method	Links
Ours w/o Post-Processing2026.01		65	43	35	40.3	38.4	26.5	31.6	40
Ours (Full Pipeline)2026.01		64.8	43.3	37	42	40.9	29.9	33	41.6
Ours w/o Fine-tuning / T-CoT2026.01		62.6	41.5	32	38.3	35.2	23.5	29.1	37.5
Ours w/o Pre-Processing2026.01		62	40.3	35.7	39	35.4	25	29.4	38.1
EgoSG2026.06		61.5	37.33	24.67	38	44	32.67	29	38.17
Gemini Flash 2.02026.06		61.5	37.33	—	38	44	32.67	—	42.7
Gemini Pro2026.01		60.5	46.2	34.7	39.6	32.5	20.8	28.7	37.6
Qwen2.5 VL 32B In.2026.01		59	37	33	40.3	35.6	19.8	33.6	36.9
InternVL32026.06		53.5	28	—	40.5	40.5	30.67	—	38.63
Video2026.06		51.75	34	32.67	46	39	26.67	34	37.73
Gemini Flash 2.02026.06		51.75	34	—	46	39	26.67	—	39.48
Qwen2.5-VL2026.06		46.25	27	—	34.5	37	34.67	—	35.88
VideoLLaMa32026.06		45.5	31.33	—	32.5	31	28.67	—	33.8
Qwen2.5-VL2026.06		42.25	31.67	—	33	39	30	—	35.18
InternVL32026.06		42.25	18.67	—	33	38.5	34	—	33.28
Qwen2.5 VL 7B In.2026.01		40.6	35.8	32	37.3	35	23.9	29.7	33.5
VideoLLaMa32026.06		39.75	32	—	33	35.5	30.67	—	34.18
Qwen2.5-VL2026.06		37.75	29.67	—	31	30.5	26.67	—	31.12
Qwen2.5-VL2026.06		37.25	30.67	—	32	25.5	38	—	32.68
InternVL32026.06		37	36	—	38	35.5	18.67	—	33.03
LLaVA-Video2026.01		36.3	33.5	38.7	43	27.3	18.9	29.3	32.4
InternVL32026.06		35	30.67	—	33.5	24	26	—	29.83
InternVL32026.06		35	29	—	40	31	20	—	31
VideoLLaMa32026.06		31.75	30.67	—	25	29	30	—	29.28
InternVL32026.06		31	20	—	33.5	34	28.67	—	29.43
VideoLlama 22026.01		30.8	25.7	32.7	27.2	25.7	28.5	21.2	27.4
LongVA2026.01		29.6	30.8	33.7	30.7	32.9	22.7	24.5	29.3
VideoLLaMa32026.06		28.25	20.33	—	27	28	26	—	25.92