Safety and Utility Evaluation on XSTest (S_safe, S_help, S_nat)
9.89Safety ScoreQwen3-max
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Qwen3-max2026.05 | 9.89 | 7.81 | 8.35 | |
| GPT-4o2026.05 | 9.84 | 3.58 | 6.64 | |
| Kimi-k22026.05 | 9.68 | 7.35 | 8.13 | |
| LANCERefinement Model=1.5B, Backbone Scale=7B/8B2026.05 | 9.55 | 7.88 | 8.53 | |
| Gemini2.5-Pro2026.05 | 9.33 | 2.84 | 4.12 |