Multi-Modal Long Document Question Answering on VisDoMBench (Full) benchmark leaderboard · SOTA2 Research

Multi-Modal Long Document Question Answering on VisDoMBench (Full)

78.85SPIQA ScoreQwen3-VL-32B + CDS

Evaluation Results

Method	Links
Qwen3-VL-32B + CDS2026.06		78.85	70.9	65.59	57.45	62.17	66.99
VisDoMRAG2026.06		75.44	61.02	56.21	63.36	69.03	—
VisDoMRAG2026.06		75.44	61.02	56.21	63.36	69.03	65.01
FLOWREADER2026.06		74.23	64.45	58.4	57.32	72.93	—
-w/o System 22026.06		74.11	63.26	57.95	55.89	72.36	—
LightRAG2026.06		73.88	64.71	51.02	75	29.63	—
LightRAG2026.06		73.88	64.71	51.02	75	29.63	58.85
G^2-Reader2026.06		73.19	66.89	57.1	61.56	72.31	—
G2-Reader2026.06		73.19	66.89	57.1	61.56	72.31	66.21
MMGraphRAG2026.06		69.91	72.4	56.36	64.11	54.2	—
MMGraphRAG2026.06		69.91	72.4	56.36	64.11	54.2	63.4
ViDoRAG2026.06		68.18	58.74	43.67	37.86	71.71	—
ViDoRAG2026.06		68.18	58.74	43.67	37.86	71.71	56.03
RAGAnything2026.06		67.69	57.76	42.02	41.6	52.18	—
RAGAnything2026.06		67.69	57.76	42.02	41.6	52.18	52.25
Deepseek-OCR2026.06		63.6	70.32	51.58	61.91	65.69	—
Deepseek-OCR2026.06		63.6	70.32	51.58	61.91	65.69	62.62
GraphRAG2026.06		62.65	61.35	42.9	65.76	21.68	—
GraphRAG2026.06		62.65	61.35	42.9	65.76	21.68	50.87
GPT-52026.06		55.22	63.94	37.08	64.08	45.06	—
GPT-52026.06		55.22	63.94	37.08	64.08	45.06	53.08
MA-RAG2026.06		45.52	27.7	33.43	29.32	29.4	—
MA-RAG2026.06		45.52	27.7	33.43	29.32	29.4	33.07
Qwen3-VL-32B2026.06		29.86	37.39	34.32	23.06	24.87	—
Qwen3-VL-32B2026.06		29.86	37.39	34.32	23.06	24.87	29.9