Audio-Video Understanding on OmniVideoBench
29.2Avg LatencyVideoLLaMA2-7B
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| VideoLLaMA2-7BModality=A+V, Reasoning/Extended Inference=false2025.12 | 29.2 | 32 | 28.2 | 29.6 | 28.29 | |
| Qwen2.5-Omni-7BModality=A+V, Reasoning/Extended Inference=false2025.12 | 29.3 | 41.57 | 27.41 | 25.33 | 26.72 | |
| Qwen2.5-VL-72BModality=V, Reasoning/Extended Inference=false2025.12 | 29.5 | 33.13 | 30.03 | 31.88 | 24.43 | |
| Baichuan-Omni-1.5Modality=A+V, Reasoning/Extended Inference=false2025.12 | 30.7 | 28.92 | 31.78 | 28.38 | 32.44 | |
| RandomBackbone=Qwen2.5-Omni-3B, Retained Ratio (%)=55%, Frame Budgets=1282026.05 | 31 | — | — | — | — | |
| DyCoke (V&A)Backbone=Qwen2.5-Omni-3B, Retained Ratio (%)=50%, Frame Budgets=1282026.05 | 31.3 | — | — | — | — | |
| OmniZipBackbone=Qwen2.5-Omni-3B, Retained Ratio (%)=30%, Frame Budgets=1282026.05 | 31.4 | — | — | — | — | |
| RandomBackbone=Qwen2.5-Omni-7B, Retained Ratio (%)=55%, Frame Budgets=1282026.05 | 32.1 | — | — | — | — | |
| DyCoke (V&A)Backbone=Qwen2.5-Omni-7B, Retained Ratio (%)=50%, Frame Budgets=1282026.05 | 32.4 | — | — | — | — | |
| OmniZipBackbone=Qwen2.5-Omni-3B, Retained Ratio (%)=45%, Frame Budgets=1282026.05 | 32.5 | — | — | — | — | |
| OmniSelectBackbone=Qwen2.5-Omni-3B, Retained Ratio (%)=45%, Frame Budgets=1282026.05 | 32.7 | — | — | — | — | |
| OmniSelectBackbone=Qwen2.5-Omni-7B, Retained Ratio (%)=30%, Frame Budgets=1282026.05 | 32.7 | — | — | — | — | |
| OmniZipBackbone=Qwen2.5-Omni-7B, Retained Ratio (%)=30%, Frame Budgets=1282026.05 | 32.8 | — | — | — | — | |
| Full TokensBackbone=Qwen2.5-Omni-3B, Retained Ratio (%)=100%, Frame Budgets=1282026.05 | 32.9 | — | — | — | — | |
| OmniSelectBackbone=Qwen2.5-Omni-7B, Retained Ratio (%)=45%, Frame Budgets=1282026.05 | 33.1 | — | — | — | — | |
| OmniSelectBackbone=Qwen2.5-Omni-3B, Retained Ratio (%)=30%, Frame Budgets=1282026.05 | 33.3 | — | — | — | — | |
| OmniZipBackbone=Qwen2.5-Omni-7B, Retained Ratio (%)=45%, Frame Budgets=1282026.05 | 33.3 | — | — | — | — | |
| Full TokensBackbone=Qwen2.5-Omni-7B, Retained Ratio (%)=100%, Frame Budgets=1282026.05 | 34.6 | — | — | — | — | |
| Qwen3-Omni-30BModality=A+V, Reasoning/Extended Inference=false2025.12 | 38.4 | 45.78 | 37.03 | 38.86 | 35.11 | |
| Qwen3-VL-PlusModality=V, Reasoning/Extended Inference=false2025.12 | 38.93 | 36.92 | 45.27 | 37.87 | 30.65 | |
| Gemini-2.0-FlashModality=A+V, Reasoning/Extended Inference=false2025.12 | 41.5 | 49.4 | 43.15 | 41.05 | 34.87 | |
| Gemini-2.5-FlashModality=A+V, Reasoning/Extended Inference=true2025.12 | 52.4 | 55.42 | 55.1 | 47.37 | 52.11 | |
| OmniAgentModality=A+V, Reasoning/Extended Inference=false2025.12 | 59.1 | 66.08 | 58.53 | 59.03 | 55.64 |