Cross-video Understanding on XVBench
37.1AccuracyVimRAG
Evaluation results
| Method | Links | |
|---|---|---|
| VimRAGBackbone=Qwen3-VL-8B-Instruct2026.02 | 37.1 | |
| VimRAGBackbone=Qwen3-VL-4B-Instruct2026.02 | 34.2 | |
| Mem1Backbone=Qwen3-VL-8B-Instruct2026.02 | 32.2 | |
| Mem1Backbone=Qwen3-VL-4B-Instruct2026.02 | 31.9 | |
| VideoRAGBackbone=Qwen3-VL-8B-Instruct2026.02 | 30.7 | |
| VideoRAGBackbone=Qwen3-VL-4B-Instruct2026.02 | 29.8 | |
| Vanilla RAGBackbone=Qwen3-VL-8B-Instruct2026.02 | 29.7 | |
| Vanilla RAGBackbone=Qwen3-VL-4B-Instruct2026.02 | 27.2 | |
| MemAgentBackbone=Qwen3-VL-8B-Instruct2026.02 | 26.9 | |
| MemAgentBackbone=Qwen3-VL-4B-Instruct2026.02 | 24.4 | |
| ReActBackbone=Qwen3-VL-8B-Instruct2026.02 | 24 | |
| ReActBackbone=Qwen3-VL-4B-Instruct2026.02 | 21.3 | |
| UniversalRAGBackbone=Qwen3-VL-8B-Instruct2026.02 | 9.6 | |
| UniversalRAGBackbone=Qwen3-VL-4B-Instruct2026.02 | 7.5 |