Nous Research has introduced Arete-27B, a fine-tuned version of Alibaba's Qwen3.8-27B designed to produce responses more aligned with Western values.

Following additional training on approximately 53 million tokens, the share of responses adopting Chinese government narratives dropped from 22% to 0.9% in the company's 803-prompt evaluation. Nous Research says the model retains its general capabilities and has made its weights, training data, and evaluation tools publicly available.