Omni-video Question Answering on SOVBench-O
89.5AV Context: Real-Time AccuracyGemini 3.0 Flash
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Gemini 3.0 FlashAudio=true, frames=-2026.05 | 89.5 | 87.6 | 82.6 | 85.6 | 90.3 | 89.7 | 87.2 | 88.6 | |
| Gemini 2.5 FlashAudio=true, frames=-2026.05 | 87 | 80.4 | 66.8 | 75.6 | 88.9 | 85.6 | 70.8 | 78.8 | |
| StreamOVAudio=true, frames=642026.05 | 86.9 | 73.2 | 78.6 | 81.6 | 86.7 | 80.4 | 82.1 | 83.8 | |
| Qwen3-Omni-30B-A3BAudio=true, frames=1 fps2026.05 | 85.3 | 77.3 | 64.8 | 73.7 | 87.3 | 82.5 | 74.3 | 79.9 | |
| Qwen2.5-Omni-7BAudio=true, frames=1 fps2026.05 | 78.4 | 75.3 | 52.1 | 63.9 | 79.5 | 81.4 | 53.6 | 65.5 | |
| ROMA-7BAudio=true, frames=1 fps2026.05 | 78 | 73.2 | 46.3 | 60.4 | 78.1 | 79.4 | 47.5 | 61.5 | |
| Qwen3-VL-30B-A3BAudio=false, frames=1 fps2026.05 | 69.5 | 67 | 42.3 | 54.5 | 65 | 70.1 | 49.5 | 56.8 | |
| Qwen2.5-VL-7BAudio=false, frames=1 fps2026.05 | 59.7 | 60.8 | 36 | 46.8 | 63.8 | 70.1 | 38.9 | 50.6 |