Nous Research releases open-source value-aligned Arete-27B

Nous Research has open-sourced Arete-27B under Apache 2.0, a post-trained Qwen 27B model aimed at realigning open-weight models toward Western values, claiming to reduce pro-China messaging rates from 22% to 1% on their seven-category value benchmark.

2026-11-07 ~ 2026-11-07 · 3 related posts

• Nous Research releases Arete-27B, a value-alignment fine-tune of Qwen3.5 — NousResearch · 2026-11-07
• Nous Research releases Arete-27B, cutting a Chinese open model's propaganda rate from 22% to 1% — NousResearch · 2026-11-07
• Nous Research ships Arete-27B, a Qwen3-27B fine-tune trained to reinforce Western values — altryne · 2026-11-07