Video Question Answering on LVBench (p@1, p@4, m@4)
40.3P@1VSeek-GPT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| VSeek-GPTbase_model=GPT-5.22026.07 | 40.3 | 58.5 | 40.9 | |
| VSeek-EMreward_type=exact-match2026.07 | 40 | 54.9 | 41.5 | |
| GPT-5.2frames=642026.07 | 38.8 | 46.1 | 39.1 | |
| VSeek-VETLreward_type=VETL2026.07 | 38.8 | 53 | 39.7 | |
| Qwen3-VL-4B-Instructparameters=4B2026.07 | 38.7 | 47.7 | 38.6 | |
| Qwen3-VL-8B-Instructparameters=8B2026.07 | 38.7 | 47.7 | 38.6 | |
| Naive RAG2026.07 | 37.1 | 46.1 | 38.7 | |
| Qwen3-VL-4B-Thinkingparameters=4B2026.07 | 35.6 | 41.8 | 36.4 | |
| Qwen3-VL-8B-Thinkingparameters=8B2026.07 | 35.1 | 51.3 | 37.4 | |
| VideoTree2026.07 | 34 | 36.4 | 34.2 | |
| InternVL-3.5-4Bparameters=4B2026.07 | 32.6 | 57.7 | 33.2 |