Video Question Answering on TempCompass MC
77.9AccuracyGPT-5.2
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT-5.22026.05 | 77.9 | 73.5 | 74.8 | 84 | |
| Qwen3-VL-235BModel Scale=235B2026.05 | 76.1 | 62.5 | 82.8 | 83.3 | |
| Ours FT (4B)Fine-tuning=Yes, Model Scale=4B2026.05 | 72 | 55.5 | 79.5 | 79.9 | |
| Baseline (Qwen3-VL-4B)Fine-tuning=No, Model Scale=4B2026.05 | 70.1 | 49.5 | 74.8 | 78.1 | |
| GPT-4o2026.05 | 64.1 | 46.1 | 55.3 | 76 |