Hirundo Westernized Qwen: CCP political alignment unlearned from Qwen weights
Israeli AI safety lab Hirundo released Westernized versions of Qwen3.6-35B-A3B and Qwen3.5-4B with political alignment edited directly out of the weights; CCP-aligned answers fell from 89.8% to 2.8% on a 500-prompt benchmark.
Why it matters
First public de-politicized Qwen weights — enterprises can adopt Chinese open models without the baked-in censorship, with capabilities held within 0.72 points of the original.
What you could build with it
A startup building assistants for compliance-sensitive Western enterprises could deploy the Westernized Qwen weights as a drop-in base model, offering Qwen-level coding and reasoning capability without the political framing learned into the original weights.
Does it hold up?
Too early to judge for builders — the HF weights are brand new; the release's own benchmarks (DECCP refusals 65.26% to 3.16%, ChinaBench non-compliance 96.67% to 6.67%) and CBS's independent testing suggest the de-alignment claim is real.
Built with Hirundo Westernized Qwen
- CBS News independently tested original vs Westernized Qwen (Oct 2, broadcast on CBS Evening News)article · Per the release, CBS News ran its own side-by-side tests of both models and covered the research — rare third-party validation for a weights release.
Learn more
First spotted on article: source.
More AI models
ElevenLabs Eleven v4 + v4 Turbo: new emotive TTS architecture with a 100ms real-time variantElevenLabs released Eleven v4 and Eleven v4 Turbo on September 28: a new text-to-speech architecture with inline…models · JEV 0.72VoiceStudio: open-source, fully-local ElevenLabs alternative with voice cloning, dubbing and dictation in 646 languagesVoiceStudio is an open-source, fully-local voice AI studio: voice cloning from a 3-second sample, voice design, video…models · JEV 0.71Mistral Forge: enterprise platform for training and continuously improving proprietary modelsMistral AI launched Forge (Sep 28) — a full-lifecycle model training platform (pre-training, SFT, DPO/ODPO, RL…models · JEV 0.69DeepSeek V4-Flash official API: public beta with upgraded agent capabilities, Responses API and Codex supportDeepSeek's official V4-Flash is now a public beta API with massively upgraded agent capabilities - benchmark scores…models · JEV 0.69kev: open, trainable Jev-like family of small decision models on Qwen3.5/3.8A family of small decision models (0.8B to 27B) built on Qwen3.5/3.8 that reproduces Jev's prefill-only typed-decision…models · JEV 0.68GPT-Synopsys: OpenAI and Synopsys multi-year deal to build an AI model that operates EDA toolsOpenAI and Synopsys signed a multi-year agreement on Sep 30 to jointly develop GPT-Synopsys, a specialized model that…models · JEV 0.66
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.