Multimodal Document QA on VisDoMBench SlideVQA (full)
72.31AccuracyG2-Reader
Evaluation results
| Method | Links | |
|---|---|---|
| G2-ReaderFramework=Dual Evolving Graphs2026.01 | 72.31 | |
| ViDoRAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 71.71 | |
| VisDoMRAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 69.03 | |
| Deepseek-OCRFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 65.69 | |
| MMGraphRAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 54.2 | |
| RAGAnythingFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 52.18 | |
| GPT-5Framework=Single-VLM2026.01 | 45.06 | |
| LightRAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 29.63 | |
| MA-RAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 29.4 | |
| Qwen3-VL-32BFramework=Single-VLM2026.01 | 24.87 | |
| GraphRAGFramework=RAG Baseline (Qwen3-VL-32B-Instruct)2026.01 | 21.68 |