Multimodal Document QA on VisDoMBench FetaTab (full)
72.4AccuracyMMGraphRAG
Evaluation results
| Method | Links | |
|---|---|---|
| MMGraphRAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 72.4 | |
| Deepseek-OCRFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 70.32 | |
| G2-ReaderFramework=Dual Evolving Graphs2026.01 | 66.89 | |
| LightRAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 64.71 | |
| GPT-5Framework=Single-VLM2026.01 | 63.94 | |
| GraphRAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 61.35 | |
| VisDoMRAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 61.02 | |
| ViDoRAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 58.74 | |
| RAGAnythingFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 57.76 | |
| Qwen3-VL-32BFramework=Single-VLM2026.01 | 37.39 | |
| MA-RAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 27.7 |