Nous Research ships Arete-27B, a Qwen3-27B fine-tune trained to reinforce Western values

altryne · x · 2026-11-07

Nous Research introduced Arete-27B, a post-trained version of Qwen3.8-27B aimed at reinforcing Western values. The team argues that the strongest current open models come from Chinese labs and ship with state-aligned behavior baked into the weights.

They built an 803-prompt eval across seven value categories—including freedom of expression, democracy, and privacy—to measure how often a model adopts China's party line, then fine-tuned on roughly that data to shift the model's behavior. The release doubles as a value-alignment experiment with its own benchmark.

Original post →

Nous Research ships Arete-27B, a Qwen3-27B fine-tune trained to reinforce Western values

More from Models