Video Understanding on ETBench
53.4RAR AccuracyFactum-4B
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Factum-4BSource Type=Open-Source, FPS=1fps2026.04 | 53.4 | 53.4 | 55 | 66.1 | 16.8 | 19.7 | 26.8 | 58.3 | |
| Qwen3-VL-4B-InstructSource Type=Open-Source, FPS=1fps2026.04 | 53.1 | 54.4 | 54.3 | 64.6 | 14.8 | 20.6 | 9.6 | 57.7 | |
| Qwen3-VL-4B-ThinkingSource Type=Open-Source, FPS=1fps2026.04 | 53 | 51 | 54.2 | 64.3 | 15.7 | 20 | 20.4 | 55.6 | |
| Qwen2.5-VL-7BSource Type=Open-Source, FPS=1fps2026.04 | 41 | 42 | 51.4 | 20.1 | 2.9 | 15.3 | 22.3 | 32.7 | |
| Qwen2.5-VL-3BSource Type=Open-Source, FPS=1fps2026.04 | 38.4 | 47.2 | 52.2 | 19.9 | 2.6 | 14.8 | 18.1 | 55.1 | |
| GPT-4vSource Type=Close-Source, FPS=2fps2026.04 | 33.3 | 40.9 | 46.2 | 27 | 1.8 | 18 | 23.9 | — | |
| TRACE-7BSource Type=Open-Source, FPS=1fps2026.04 | 29.4 | 28.8 | 42.6 | 46.8 | 12.3 | 21.6 | 17.8 | 52.4 | |
| VideoLLama2-7BSource Type=Open-Source, FPS=1fps2026.04 | 28.8 | 27.4 | 28 | 0.1 | 0 | 0 | 0 | — | |
| VTimeLLM-7BSource Type=Open-Source, FPS=1fps2026.04 | 28.4 | 31 | 28 | 7.6 | 1.9 | 18.2 | 6.9 | 1.9 | |
| GPT-4oSource Type=Close-Source, FPS=2fps2026.04 | 27.8 | 27.3 | 57.7 | 40.4 | 4.5 | 20 | 13.6 | — |