Nous Research fine-tunes Qwen to cut propaganda answers from 22% to 0.9%

Nous Research released Arete-27B, Alibaba's Qwen3.8-27B fine-tuned on about 52.8M tokens with no reinforcement learning. On Nous's 803-prompt test, graded by Grok 4.7, answers containing propaganda fell from 22% to 0.9% and the fact score rose from 2.32 to 2.98 of 5. Weights, eval and training data are on Hugging Face. Results are Nous's own.

HEAT

OUTLETS

1

NEW IN 6H

+1

FIRST SEEN

Nov 7, 2026, 07:32 ET

LAST SIGNAL

1h ago