Multimodal Document QA on VisDoMBench SPIQA (full)
75.44AccuracyVisDoMRAG
Evaluation results
| Method | Links | |
|---|---|---|
| VisDoMRAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 75.44 | |
| LightRAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 73.88 | |
| G2-ReaderFramework=Dual Evolving Graphs2026.01 | 73.19 | |
| MMGraphRAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 69.91 | |
| ViDoRAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 68.18 | |
| RAGAnythingFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 67.69 | |
| Deepseek-OCRFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 63.6 | |
| GraphRAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 62.65 | |
| GPT-5Framework=Single-VLM2026.01 | 55.22 | |
| MA-RAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 45.52 | |
| Qwen3-VL-32BFramework=Single-VLM2026.01 | 29.86 |