Share your thoughts, 1 month free Claude Pro on usSee more

Long-document Visual Question Answering on MMLongBench 64K context

93.1MMLB-D

GPT-5.5

Updated 2mo ago

Evaluation Results

Method	Links
GPT-5.5 2026.05		93.1	92.52	95.77	93.8
Gemini-3.1-Pro 2026.05		79.4	78.25	93	83.55
Gemini-3.1-Flash 2026.05		74.43	79.32	91	81.58
Gemini-2.5-Pro 2026.05		65.4	68.13	89	74.18
GPT-5.4 2026.05		61.56	72.37	87.1	73.68
Gemma4-31B 2026.05		56.33	62.42	84	67.58
Qwen2.5-VL-72B 2026.05		51.97	58.96	77.5	62.81
Qwen2.5-VL-32B 2026.05		44.06	60.58	77	60.55
Gemma4-26B-A4B 2026.05		40.67	42.82	64	49.16
InternVL3.5-38B 2026.05		39.17	48.37	52	46.51
Gemma3-27B 2026.05		38	51.72	69	52.91
MMProLong 2026.05		36	62.69	80	59.56
Gemma4-E4B 2026.05		36	30.21	45	37.07
InternVL3-38B 2026.05		36	49.32	74	53.11
InternVL3-14B 2026.05		35.33	49.69	67	50.67
Qwen2.5-VL-7B 2026.05		32.17	49.57	75	52.24
Gemma3-12B 2026.05		31.52	49.59	63	48.03
InternVL3.5-8B 2026.05		29.26	47.94	38	38.4
InternVL3.5-14B 2026.05		29.04	48.57	53	43.54
Gemma4-E2B 2026.05		28	29.86	48	35.29
InternVL3-8B 2026.05		27.72	52.84	70	50.19
Gemma3-4B 2026.05		25	33.33	57	38.44