Back to AIBriefs
AnalysisAI Models

Abliterated Qwen 3.6 27B reduces refusal rate from 92% to 7.6%

Abliterated Qwen 3.6 27B reduces refusal rate from 92% to 7.6%View on Reddit

A community project called Apostate produced an abliterated Qwen 3.6 27B model, cutting refusal rate from 92% to 7.6% with minimal capability loss (0.120). The model is available on Hugging Face.

·
16 hours ago
Abliterated Qwen 3.6 27B reduces refusal rate from 92% to 7.6% — AIBriefs