Nous Research Removes Propaganda from Chinese AI

Nous Research fine-tuned Alibaba's Qwen3.8-27B model using 53 million tokens to strip Chinese state propaganda, dropping the rate from 22 percent to under one percent. The resulting Arete-27B model scores 2.98 on factual accuracy while maintaining general capabilities, proving political bias is a thin layer. This development follows Nous raising $90 million in a Series B round in October.