Local ModelsMedium importance

Nous Research releases Arete-27B: Qwen3.8-27B retrained to remove Chinese government bias, with open weights

Nous did supervised fine-tuning of Alibaba's Qwen3.8-27B with about 52.8 million tokens, using Inkling as the teacher. On its own evaluation of 803 questions across seven topics, answers with propaganda fell from 22% to about 1% and the fact score rose from 2.32 to 2.98 (out of 5). On Terminal-Bench-2.1 it scored 70% versus 67% for the base model, as run by Nous. The weights are on Hugging Face under Apache 2.0 (54.7 GB in BF16), and there are already community GGUFs: Q4_K_M is 17.2 GB.