Nous Research releases Arete-27B, cutting propaganda-framed answers on Qwen3.8-27B from 22% to 0.9%

Nous Research released Arete-27B, an open-weight supervised fine-tune of Alibaba’s Qwen3.8-27B aimed at Western value alignment. Using Inkling as the teacher, the lab ran about 52.8 million tokens of text-only SFT over two epochs (1,202 steps) with no RL or preference optimization. On its self-reported Propaganda_Eval of 803 prompts across seven value categories, the fact mean rose from 2.32 to 2.98 out of 5 and answers containing propaganda fell from 22% to 0.9%, with no meaningful regression on Hermes Bench or Terminal-Bench-2.1 in its own runs. Weights, the evaluation, and training data are public under Apache 2.0.

Written by dalao.ai: the title, summary and why it matters are our own words. The original reports are the record; their publishers keep authorship.