Community floods Hugging Face with uncensored Qwen3.8-27B variants as local offline runs take off

Over Aug 17–19, community accounts released a wave of "abliterated" (uncensored) fine-tunes of Alibaba's Qwen3.8-27B on Hugging Face, enabling fully local, offline, private runs of the 27B model on consumer hardware. All evidence points to community efforts; there is no sign of an official uncensored release by Alibaba.

Confirmed

• Huihui-ai released Huihui-Qwen3.8-27B-abliterated on Aug 18, based on Qwen/Qwen3.8-27B via an image-text-to-text pipeline, followed by a GGUF quantized version on Aug 19
• PocketAiHub reported an MLX-framework Qwen3.8-27B-Abliterated on Aug 17, supporting image-to-text pipelines with 2-bit to 8-bit quantizations, optimized for Apple Silicon
• orcarouter's Qwen3.8-27B-Uncensored-GGUF (llama.cpp compatible, vision-language capable) trended on Hugging Face
• @johnseach shared that the 27B model can run locally and offline on a 32GB-RAM PC via Ollama
• @atShruti noted a jailbreak project on Hugging Face aimed at bypassing model watermark protection, releasing an Uncensored Qwen variant

Unconfirmed

• @johnseach's posts are inconsistent: some claim Alibaba Cloud officially released Qwen3.8-27B-Uncensored, others say the locally deployed model was Qwen2.5-27B-Instruct-Uncensored. Whether these are official releases or community fine-tunes cannot be verified; the evidence from other posts suggests community abliteration rather than an official release

Why it matters

• Uncensored fine-tunes have become routine in the open-source ecosystem, letting users run large models locally and bypass cloud content moderation
• @johnseach noted the timing coincides with Anthropic tightening model restrictions, highlighting the tension between community open models and commercial APIs over content policy

2026-08-17 ~ 2026-08-19 · 8 related posts

• [source] Qwen3.8-27B-Abliterated-MLX Model Released on Hugging Face — PocketAiHub · 2026-08-17
• Jailbreak on Hugging Face bypasses model watermarks — atShruti · 2026-08-18
• [source] Huihui releases uncensored Qwen3.8-27B abliterated model — huihui-ai · 2026-08-18
• Uncensored Qwen3.8-27B GGUF Hits Hugging Face, Runnable Fully Offline — johnseach · 2026-08-18
• Alibaba releases Qwen3.8-27B-Uncensored open-source model — johnseach · 2026-08-18
• Alibaba Releases 27B Uncensored Model Runnable Locally via Ollama — johnseach · 2026-08-18
• Huihui releases Qwen3.8-27B abliterated GGUF, trending on Hugging Face — huihui-ai · 2026-08-19
• [source] Qwen3.8-27B-Uncensored GGUF hits Hugging Face trending — orcarouter · 2026-08-19