---
format: "aidr-story-markdown/v1"
id: "b3e24fd706534e349db834a2e0a999f51939aaed05357ea87fafb0a8405d9708"
canonical_url: "https://aidr.today/b3e24fd7?lang=en"
title: "Nous Research Launches Arete-27B to Strip State Alignment From Qwen AI Model"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-11-06T22:22:48.000Z"
category: "Models"
topics: ["llm","nous-research","arete-27b","open-source","huggingface","safety"]
source_urls: ["https://marketbrief.now/ai/nous-research-launches-arete-27b-to-strip-state-alignment-from-qwen-ai-m-e22922b1","https://huggingface.co/NousResearch/Arete-27B","https://huggingnews.com/ai/nous-research-launches-arete-27b-to-strip-state-alignment-from-qwen-ai-m-e22922b1","https://the-decoder.com/state-propaganda-in-some-chinese-ai-models-is-only-a-thin-layer-that-fine-tuning-can-strip-away/","https://www.marktechpost.com/2026/11/07/nous-research-arete-27b-qwen3-8-27b-propaganda-fine-tune/","https://www.techmeme.com/261108/p3"]
summary: "Nous Research released Arete-27B, a modified version of the Qwen3.8-27B open-weights model post-trained on approximately 53 million tokens to reduce state-aligned responses, with a testing suite of 803 prompts showing the frequency of such responses falling from 22% to 1%. The developer stated that the launch includes no performance penalty based on Hermes Bench scores and has made the weights and training data publicly available."
---

# Nous Research Launches Arete\-27B to Strip State Alignment From Qwen AI Model

> [Open the canonical story](<https://aidr.today/b3e24fd7?lang=en>)

**Published:** 2026-11-06T22:22:48.000Z
**Category:** Models
**Topics:** llm, nous\-research, arete\-27b, open\-source, huggingface, safety

## Summary

Nous Research released Arete\-27B, a modified version of the Qwen3\.8\-27B open\-weights model post\-trained on approximately 53 million tokens to reduce state\-aligned responses, with a testing suite of 803 prompts showing the frequency of such responses falling from 22% to 1%\. The developer stated that the launch includes no performance penalty based on Hermes Bench scores and has made the weights and training data publicly available\.

## Sources

- [Story source](<https://marketbrief.now/ai/nous-research-launches-arete-27b-to-strip-state-alignment-from-qwen-ai-m-e22922b1>)
- [Story source](<https://huggingface.co/NousResearch/Arete-27B>)
- [Story source](<https://huggingnews.com/ai/nous-research-launches-arete-27b-to-strip-state-alignment-from-qwen-ai-m-e22922b1>)
- [Story source](<https://the-decoder.com/state-propaganda-in-some-chinese-ai-models-is-only-a-thin-layer-that-fine-tuning-can-strip-away/>)
- [Story source](<https://www.marktechpost.com/2026/11/07/nous-research-arete-27b-qwen3-8-27b-propaganda-fine-tune/>)
- [Story source](<https://www.techmeme.com/261108/p3>)