Role-playing on DynSess-Eval Human Evaluation 1.0 (test)
3.38Average ScoreDoubao-1.5-pro-character
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Doubao-1.5-pro-characterDomains=Character Models, Params=∼200B12026.05 | 3.38 | 3.29 | 3.42 | 3.47 | 3.35 | |
| DynSess-Character-32B (DSPO)Domains=Ours, Params=32B2026.05 | 3.37 | 3.23 | 3.34 | 3.56 | 3.34 | |
| DynSess-Character-32B (GSRPO)Domains=Ours, Params=32B2026.05 | 3.35 | 3.39 | 3.25 | 3.46 | 3.31 | |
| GPT-5.4Domains=General Models, Params=-2026.05 | 3.31 | 3.23 | 3.34 | 3.38 | 3.28 | |
| Claude Sonnet 4.6Domains=General Models, Params=-2026.05 | 3.24 | 3.12 | 3.21 | 3.37 | 3.27 | |
| Deepseek v3.2Domains=General Models, Params=685B2026.05 | 3.21 | 3.11 | 3.17 | 3.34 | 3.22 | |
| Gemini-3-proDomains=General Models, Params=-2026.05 | 3.17 | 3.05 | 3.06 | 3.36 | 3.2 | |
| Qwen-plus-characterDomains=Character Models, Params=-2026.05 | 3.01 | 2.86 | 2.88 | 3.22 | 3.06 | |
| MiniMax-M2-HERDomains=Character Models, Params=32B2026.05 | 2.99 | 2.95 | 2.91 | 3.13 | 2.97 | |
| CoserDomains=Character Models, Params=70B2026.05 | 2.96 | 2.89 | 2.9 | 3.09 | 2.98 |