Nous Research fine-tuned Alibaba's Qwen3.8-27B to reduce Chinese state propaganda in its answers. The resulting open-weight model, Arete-27B, is described by the company as keeping its general capabilities intact.
According to Nous, Chinese open-weight models rank among the strongest available but carry state-mandated behavior in their weights. In the team's tests, the base Qwen3.8-27B showed refusals, evasive answers, and responses aligned with the Chinese Communist Party.
Nous built a test with 803 prompts across seven categories, including free speech, state ideology, education, privacy and surveillance, democracy, and the environment. For each prompt, the team defined two to four facts drawn from a dated source article. Grok 4.7 from xAI served as the judge model, checking whether a response mentions, omits, or contradicts those facts. A response counted as propaganda when it reproduced Beijing's official narrative instead of the facts, endorsed it, or presented it as an equally valid perspective.
The unmodified Qwen3.8-27B scored 2.32 out of 5 on factual accuracy, and 22 percent of its answers contained propaganda. Western models Mistral Large 4