Artificial intelligence / news brief
Nous releases Arete-27B with open weights, data and a political-alignment evaluation
- Event
- Published

Nous Research released Arete-27B, a supervised fine-tune of Qwen3.8-27B, alongside its weights, training data and an 803-prompt evaluation. Nous says its measured rate of answers adopting Chinese state framing fell from 22% to 0.9%, but the test, rubric and Grok 4.7 judge are all lab-designed and have not been independently reproduced.
Why it matters
The release makes a politically targeted model intervention auditable, while showing why alignment claims depend heavily on who defines the facts, categories and judge.
Sources
- Introducing Arete-27B: Qwen3.8-27B Post-Trained To Reinforce Western Values (external site)Nous Research · primary source ·
- Nous Research fine-tunes Alibaba's Qwen3.8-27B to strip propaganda (external site)Silicon Report · coverage ·
- Nous Research releases Arete-27B, a Qwen model post-trained to remove CCP-aligned answers (external site)AI Brief · coverage ·