Multimodal Document QA on VisDoMBench SciGraphQA (full)
75AccuracyLightRAG
Evaluation results
| Method | Links | |
|---|---|---|
| LightRAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 75 | |
| GraphRAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 65.76 | |
| MMGraphRAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 64.11 | |
| GPT-5Framework=Single-VLM2026.01 | 64.08 | |
| VisDoMRAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 63.36 | |
| Deepseek-OCRFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 61.91 | |
| G2-ReaderFramework=Dual Evolving Graphs2026.01 | 61.56 | |
| RAGAnythingFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 41.6 | |
| ViDoRAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 37.86 | |
| MA-RAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 29.32 | |
| Qwen3-VL-32BFramework=Single-VLM2026.01 | 23.06 |