Long-form Generation on LLaVA-Bench
96.6ScoreQwen2.5
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen2.5Type=AR2026.05 | 96.6 | |
| LLaVA-OVType=AR2026.05 | 83.8 | |
| InternVL3Type=AR2026.05 | 82.3 | |
| LLaVA-1.6Type=AR2026.05 | 75.3 | |
| LLaDA-V + OursType=Diff.2026.05 | 64.1 | |
| LLaDA-VType=Diff.2026.05 | 61.3 | |
| LaViDa + OursType=Diff.2026.05 | 46.5 | |
| LaViDaType=Diff.2026.05 | 39.5 |