Nous Research fine-tunes Qwen to strip political bias

Nous Research retrained Alibaba's Qwen3.8-27B into Arete-27B using 53 million tokens of supervised fine-tuning, cutting its pro-Chinese propaganda rate from 22% to under 1% while general capabilities held up in tests. The team says the bias is applied late and thin.

• NousResearch releases Arete-27B, a Qwen fine-tune2026-11-06
• Real-World Test: AI Coding Agents Need Cleanup on ~3 of 4 Tasks2026-11-08
• Study: 8-bit KV cache near-lossless, 4-bit measurably worse2026-11-07
• Alibaba quietly announces Qwen 42026-11-06
• Same model, different harness: task completion rises from 30% to 81%2026-11-03