Multimodal Document QA on VisDoMBench PaperTab (full)
57.1AccuracyG2-Reader
Evaluation results
| Method | Links | |
|---|---|---|
| G2-ReaderFramework=Dual Evolving Graphs2026.01 | 57.1 | |
| MMGraphRAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 56.36 | |
| VisDoMRAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 56.21 | |
| Deepseek-OCRFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 51.58 | |
| LightRAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 51.02 | |
| ViDoRAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 43.67 | |
| GraphRAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 42.9 | |
| RAGAnythingFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 42.02 | |
| GPT-5Framework=Single-VLM2026.01 | 37.08 | |
| Qwen3-VL-32BFramework=Single-VLM2026.01 | 34.32 | |
| MA-RAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 33.43 |