| video-question-answering-on-videoholmes | Qwen2.5-VL-7B (GRPO, concise reason, trained with token compression & 6x frames) | #12 | Accuracy: 41.6 |
| video-question-answering-on-mmvu | Qwen2.5-VL-7B (GRPO, concise reason, trained with token compression & 6x frames) | #23 | Accuracy: 63.6 |
| video-question-answering-on-mlvu | Qwen2.5-VL-7B (GRPO, concise reason, trained with token compression & 6x frames) | #45 | Accuracy: 67.0 |
| video-question-answering-on-mvbench | Qwen2.5-VL-7B (GRPO, concise reason, trained with token compression & 6x frames) | #48 | Avg.: 65.6 |
| video-question-answering-on-lvbench | Qwen2.5-VL-7B (GRPO, concise reason, trained with token compression & 6x frames) | #51 | Accuracy: 38.9 |