Arete-27B: Nous Research post-trains Qwen3.8 to strip state-aligned framing
Nous Research fine-tuned Alibaba's Qwen3.8-27B with SFT only, cutting answers that adopt the Chinese state's framing from 22% to 0.9% on its own 803-prompt eval.
Why it matters
First public proof that a small, carefully chosen SFT run can re-align a frontier open model's values with no capability loss — weights, training data and eval all public under Apache 2.0.
What you could build with it
A research team could rerun Nous's open Propaganda_Eval on every open-weight model their organisation considers deploying, and publish a public values-alignment leaderboard that buyers consult before adopting a model.
Does it hold up?
Usable now: BF16 weights (53.8 GB) and a 17 GB Q4 quant run on consumer GPUs; but the headline numbers are Nous's own eval judged by Grok 4.7, so independent reruns are still needed before trusting the alignment claims.
Built with Arete-27B
- Nous Research Releases Arete-27B that Cuts Qwen3.8-27B Propaganda From 22% to 0.9%article · MarkTechPost covers the release: 27B dense model, SFT only, teacher Inkling, Apache 2.0 weights on Hugging Face.
- Nous Research releases Arete-27B, a Qwen model post-trained to remove CCP-aligned answersarticle · AI Brief summarises the 803-prompt eval and notes the US government moves to restrict Chinese models from federal use.
- State propaganda in some Chinese AI models is only a thin layer that fine-tuning can strip awayarticle · The Decoder frames the result as evidence that state-aligned behaviour in Chinese open models is a thin, removable layer.
Learn more
First spotted on article: source.
More AI models
VoiceStudio: open-source, fully-local ElevenLabs alternative for voice cloning, dubbing, and audiobooksdebpalash/VoiceStudio is a desktop app that runs voice cloning, voice design, video dubbing, dictation, transcription…models · JEV 0.75Soup: fine-tune an 8B LLM on a 4 GB laptop GPU from one YAML fileAn open-source CLI that fine-tunes LLMs from a single YAML file, with layer streaming that trains an 8B model on a 4…models · JEV 0.73Tencent Hunyuan Hy3: open 295B MoE with 21B active under Apache 2.0Tencent open-sourced Hy3, a 295B-parameter mixture-of-experts model with only 21B parameters active per token, under…models · JEV 0.73NVIDIA Isaac Lab 3.0 GA + Isaac OS open runtime + Agile ONE SN humanoidMajor Isaac platform expansion at GTC Berlin Oct 21: Isaac Lab 3.0 is generally available as the open-source robot…models · JEV 0.73NVIDIA Nemotron 3.5 Retriever collection: Embed 8B/1B, Rerank 8B, Parse 631MNVIDIA's Oct 21 collection drop for agentic retrieval: Nemotron-3.5-Embed in 8B BF16, 1B BF16 and 1B NVFP4…models · JEV 0.72Kimi K3.1 Pro: Moonshot's 2.8T open-weight flagship officially launchesMoonshot AI officially launched Kimi K3.1 Pro on Oct 23, a mixture-of-experts model with 2.8 trillion total / 104…models · JEV 0.72
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.