List your company

Harmful-request assistance on SORRY-Bench decontaminated base prompts

17.1Compliance RateQwen3.5-27B
-0.6843.9338.5513.167Jun 8, 2026

Evaluation results

MethodLinks
Qwen3.5-27BSystem Prompt=default (none), Judge=ft-mistral-7b-instruct-v0.22026.0617.1
Qwen3.5-27BSystem Prompt=v1 (PSYCHOSAFE prompt), Judge=ft-mistral-7b-instruct-v0.22026.0613.6
Qwen3.5-27BSystem Prompt=v0 (generic helpful-assistant), Judge=ft-mistral-7b-instruct-v0.22026.0613.2
Qwen3.5-27B-psysafeSystem Prompt=default (none), Judge=ft-mistral-7b-instruct-v0.22026.060
Qwen3.5-27B-psysafeSystem Prompt=v0 (generic helpful-assistant), Judge=ft-mistral-7b-instruct-v0.22026.060
Qwen3.5-27B-psysafeSystem Prompt=v1 (PSYCHOSAFE prompt), Judge=ft-mistral-7b-instruct-v0.22026.060