Multimodal RAG on VimRAG Overall
50.1Overall AccuracyVimRAG
Evaluation Results
| Method | Links | |
|---|---|---|
| VimRAGBackbone=Qwen3-VL-8B-Instruct2026.02 | 50.1 | |
| VimRAGBackbone=Qwen3-VL-4B-Instruct2026.02 | 45.2 | |
| Mem1Backbone=Qwen3-VL-8B-Instruct2026.02 | 43.6 | |
| Mem1Backbone=Qwen3-VL-4B-Instruct2026.02 | 40.6 | |
| MemAgentBackbone=Qwen3-VL-8B-Instruct2026.02 | 40.3 | |
| VideoRAGBackbone=Qwen3-VL-8B-Instruct2026.02 | 38.9 | |
| ReActBackbone=Qwen3-VL-8B-Instruct2026.02 | 37.7 | |
| Vanilla RAGBackbone=Qwen3-VL-8B-Instruct2026.02 | 37.6 | |
| MemAgentBackbone=Qwen3-VL-4B-Instruct2026.02 | 37.4 | |
| VideoRAGBackbone=Qwen3-VL-4B-Instruct2026.02 | 36.4 | |
| Vanilla RAGBackbone=Qwen3-VL-4B-Instruct2026.02 | 35 | |
| ReActBackbone=Qwen3-VL-4B-Instruct2026.02 | 33.6 | |
| UniversalRAGBackbone=Qwen3-VL-8B-Instruct2026.02 | 28.5 | |
| UniversalRAGBackbone=Qwen3-VL-4B-Instruct2026.02 | 25.2 |