Long-video Question Answering on LV-Bench (val)
66.4Overall AccuracyVideo-RAG
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Video-RAGBackbone=25B, Native Input=false2026.04 | 66.4 | 3.4 | |
| VideoStirBackbone=72B, Native Input=true2026.04 | 66 | 6.6 | |
| VideoStirBackbone=25B, Native Input=true2026.04 | 65.8 | 2.5 | |
| TV-RAGBackbone=25B, Native Input=false2026.04 | 65.5 | 2 | |
| Video-RAGBackbone=72B, Native Input=false2026.04 | 65.4 | 5.7 | |
| TV-RAGBackbone=72B, Native Input=false2026.04 | 64.6 | 4.3 | |
| AriaBackbone=25B, Native Input=true2026.04 | 64.2 | — | |
| LLaVA-VideoBackbone=72B, Native Input=true2026.04 | 61.9 | — | |
| VideoStirBackbone=7B, Native Input=true2026.04 | 60.3 | 6.5 | |
| TV-RAGBackbone=7B, Native Input=false2026.04 | 58.8 | 2.1 | |
| Video-RAGBackbone=7B, Native Input=false2026.04 | 58.7 | 3.7 | |
| LLaVA-VideoBackbone=7B, Native Input=true2026.04 | 56.6 | — | |
| VideoStirBackbone=8B, Native Input=true2026.04 | 55.1 | 5.8 | |
| TV-RAGBackbone=8B, Native Input=false2026.04 | 54.8 | 5.2 | |
| Video-RAGBackbone=8B, Native Input=false2026.04 | 54.5 | 4.6 | |
| VideoStirBackbone=26B, Native Input=true2026.04 | 52.7 | 2.9 | |
| TV-RAGBackbone=26B, Native Input=false2026.04 | 52.4 | 2.3 | |
| Video-RAGBackbone=26B, Native Input=false2026.04 | 52.2 | 2 | |
| mPLUG-Owl3Backbone=8B, Native Input=true2026.04 | 52.1 | — | |
| InternVL-1.5Backbone=26B, Native Input=true2026.04 | 51.2 | — |