Long-video understanding on LVBench (Accuracy, Latency, Memory, Avg. Acc.)
50.1AccuracyGemini-2.0-Flash
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Gemini-2.0-FlashAugmentation=Light-Omni2026.07 | 50.1 | — | — | 59.6 | |
| Light-OmniBackbone=Qwen2.5-Omni-7B2026.07 | 49.9 | 2.62 | 24.23 | 58 | |
| M3-Agent2026.07 | 49.3 | 33.1 | 62.2 | 55.6 | |
| Qwen2.5-VL-7BAugmentation=Light-Omni2026.07 | 49.3 | 6.87 | 46.92 | 54.5 | |
| Qwen3-VL-8BAugmentation=Light-Omni2026.07 | 49.3 | 6.92 | 46.9 | 57.2 | |
| Qwen2.5-VL-72B2026.07 | 48.8 | — | — | 56.1 | |
| Gemini-2.0-Flash2026.07 | 48.6 | — | — | 55.8 | |
| Qwen3-VL-8B2026.07 | 48.3 | 15.1 | 29.9 | 54.7 | |
| RAG-Rewrite2026.07 | 47.6 | 7.4 | 27.9 | 54.2 | |
| Naive RAG2026.07 | 46.9 | 6.9 | 27.7 | 52.7 | |
| Qwen2.5-VL-7B2026.07 | 43.7 | 48.8 | 119 | 49.6 | |
| Qwen2.5-Omni-7B2026.07 | 41.6 | 51.9 | 80 | 48.5 | |
| HippoMM2026.07 | 38.2 | 14.3 | 34.8 | 39.9 | |
| GPT-4o2026.07 | 30.8 | — | — | 48.1 | |
| MovieChat2026.07 | 22.5 | — | 31.1 | 21 |