Role-playing Agent Evaluation on PersonaGym
4.13Action JustificationGPT-4.1
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| GPT-4.12026.05 | 4.13 | 4.13 | 4 | 4.25 | 4.88 | 4.28 | |
| DPO-Qwen3-8Btraining=Direct Preference Optimization (DPO)2026.05 | 3.88 | 3.63 | 3.75 | 4.25 | 4.92 | 4.09 | |
| SFT-Qwen3-8Btraining=Supervised Fine-Tuning (SFT)2026.05 | 3.5 | 3.63 | 3.5 | 3.88 | 4.93 | 3.88 | |
| Qwen3-8Bstatus=base model2026.05 | 3.38 | 3.13 | 3.13 | 3.75 | 4.91 | 3.66 |