Role Fidelity on RoleBench (test)
36.4RAW ScoreDASE
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DASEBase Model=Kimi-K2-Instruct, Size=1T, Training Type=Training-Free2026.01 | 36.4 | 34.44 | 35.42 | |
| DASEBase Model=GPT-5.2, Size=Undisclosed, Training Type=Training-Free2026.01 | 31.11 | 40.09 | 35.6 | |
| Gemma-2 + SaRFTSize=9B, Training Type=Training-Based2026.01 | 30.69 | 26.7 | 28.69 | |
| Kimi-K2-InstructSize=1T, Training Type=Training-Free2026.01 | 29.42 | 19.8 | 24.61 | |
| GPT-5.2Size=Undisclosed, Training Type=Training-Free2026.01 | 28.99 | 17.73 | 23.36 | |
| Qwen2.5 + SaRFTSize=7B, Training Type=Training-Based2026.01 | 28.87 | 23.8 | 26.33 | |
| LLaMA-3 + SaRFTSize=8B, Training Type=Training-Based2026.01 | 28.58 | 25.23 | 26.91 | |
| Qwen3-235B-A22B-Instruct-2507Size=235B, Training Type=Training-Free2026.01 | 28.07 | 16.54 | 22.3 | |
| Qwen3-32BSize=32B, Training Type=Training-Free2026.01 | 27.95 | 19.27 | 23.61 | |
| gpt-oss-120bSize=120B, Training Type=Training-Free2026.01 | 27.07 | 17.06 | 22.06 |