Nous Research Unveils Arete-27B Model

Nous Research unveils Arete-27B, a post-trained version of Qwen3.8-27B designed to reinforce Western values. The team generated twenty thousand prompts across seven categories and fine-tuned the model on about 52.8 million tokens over 1,202 steps. This release aims to address biases in Chinese AI models that affect code quality and factual recall for American users.