Nous Re-Taught Qwen to Stop Sounding Like Beijing. It Took 53 Million Tokens
Nous Research post-trained Alibaba's Qwen3.8-27B with supervised fine-tuning alone — 52.8M tokens, 1,202 steps, no RL. On its own 803-prompt eval, answers adopting Beijing's official line fell from 22% to 0.9% and the fact score rose from 2.32 to 2.98 of 5, with capability flat. Nous says the result shows the values in an open model are 'only ever one fine-tune away' — which cuts both ways.