Safety Evaluation on XSTest (Over, Safety Metrics)
24OverQwen3.5-27B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Qwen3.5-27BVersion=v1, Judge=Qwen2.5-7B-Instruct, Prompt template=Röttger et al. (2024)2026.06 | 24 | 78.5 | |
| Qwen3.5-35B-A3BArchitecture=MoE2026.06 | 14.4 | 62.5 | |
| Qwen3.5-35B-A3BArchitecture=MoE2026.06 | 13.6 | 59.5 | |
| Qwen3.5-27BVersion=default, Judge=Qwen2.5-7B-Instruct, Prompt template=Röttger et al. (2024)2026.06 | 13.2 | 59 | |
| Qwen3.5-27BVersion=v0, Judge=Qwen2.5-7B-Instruct, Prompt template=Röttger et al. (2024)2026.06 | 12.4 | 63 | |
| Qwen3.5-27B-psysafeVersion=v1, Judge=Qwen2.5-7B-Instruct, Prompt template=Röttger et al. (2024)2026.06 | 9.2 | 26.5 | |
| Qwen3.5-35B-A3B-psysafeArchitecture=MoE2026.06 | 6.8 | 17.5 | |
| Qwen3.5-35B-A3B-psysafeArchitecture=MoE2026.06 | 5.6 | 19 | |
| Qwen3.5-27B-psysafeVersion=v0, Judge=Qwen2.5-7B-Instruct, Prompt template=Röttger et al. (2024)2026.06 | 4.8 | 15 | |
| Qwen3.5-27B-psysafeVersion=default, Judge=Qwen2.5-7B-Instruct, Prompt template=Röttger et al. (2024)2026.06 | 3.6 | 17 |