# wahloon.net > Track abliterated and uncensored open-weight LLMs. A filterable model catalog, practical setup guides, and timely news for people running models locally. This site tracks abliterated and uncensored open-weight language models for people running them locally. Model data lives in a typed file; news and guides are MDX. Every key page also supports `Accept: text/markdown`. ## Core pages - [Model catalog](https://www.wahloon.net/models): Filterable catalog with VRAM, license, quant formats and Hugging Face links. - [News](https://www.wahloon.net/news): Release notes and roundups. - [Guides](https://www.wahloon.net/guides): Local setup with Ollama, LM Studio and llama.cpp. - [FAQ](https://www.wahloon.net/faq): Hardware, quantization, quality and troubleshooting. - [About](https://www.wahloon.net/about): Scope and editorial approach. - [Contribute](https://www.wahloon.net/contribute): How to add a model, post or FAQ. ## Models - [Huihui-Qwen3.8-27B-abliterated](https://huggingface.co/huihui-ai/Huihui-Qwen3.8-27B-abliterated-GGUF): 27B, abliteration, approx. ~16–18 GB (Q4) at Q4, Apache-2.0. One of the most downloaded abliterated Qwen 3.8 27B builds. Strong general performance with significantly reduced refusals. - [Huihui-Qwen3.6-35B-A3B-abliterated-MTP](https://huggingface.co/huihui-ai/Huihui-Qwen3.6-35B-A3B-abliterated-MTP-GGUF): 35B-A3B, abliteration, approx. ~21 GB (Q4) at Q4, Apache-2.0. Highly popular MoE abliterated variant. Excellent quality-to-size ratio for the active parameter count. - [Qwen3.8-Flash-Next-GSQ-RCO-abliterated](https://huggingface.co/SC117/Qwen3.8-Flash-Next-GSQ-RCO-abliterated-GGUF): Flash-Next, abliteration, approx. High (large MoE) at Q4, Apache-2.0. Very high download count Flash-Next abliterated build. Strong for longer context and throughput-oriented use. - [Qwen3.8-27B-Uncensored](https://huggingface.co/JonathanColetti/Qwen3.8-27B-Uncensored-GGUF): 27B, abliteration, approx. ~16.5 GB (Q4) at Q4, Apache-2.0. Clean uncensored/abliterated Qwen 3.8 27B with multiple format options. - [Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-MTP](https://huggingface.co/HauhauCS/Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-MTP-GGUF): 27B, fine-tune, approx. ~16.5 GB at Q4, Apache-2.0. Aggressive uncensored fine-tune known for very low refusal rates. - [Qwen3.5-9B-The-Defiant-Fable-Uncensored-Heretic](https://huggingface.co/DavidAU/Qwen3.5-9B-The-Defiant-Fable-Uncensored-Heretic-NEO-IMATRIX-MAX-MTP-GGUF): 9B, heretic, approx. ~6 GB at Q4, Apache-2.0. Extremely popular smaller Heretic build. Excellent for lower VRAM machines. - [Qwen3.8-27B-Heretic-Abliterated-Uncensored](https://huggingface.co/0bserverx/Qwen3.8-27B-Heretic-Abliterated-Uncensored-GGUF): 27B, heretic-ara, approx. ~16–17 GB at Q4, Apache-2.0. Heretic + ARA refined build. Frequently cited for very low refusal scores and low KL divergence. - [Qwen3.8-27B-TURBO-Fable-Cold-Fusion-Heretic-Uncensored](https://huggingface.co/DavidAU/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NM-DAU): 27B, heretic, approx. ~16.5 GB at Q4, Apache-2.0. DavidAU Heretic series focused on strong instruction following and reduced refusals. - [Huihui-gemma-4-12B-it-abliterated](https://huggingface.co/huihui-ai/Huihui-gemma-4-12B-it-abliterated): 12B, abliteration, approx. ~8 GB at Q4, Apache-2.0 / Gemma. Solid mid-size Gemma 4 abliterated variant from the most active publisher. - [gemma-4-31B-it-uncensored-heretic](https://huggingface.co/llmfan46/gemma-4-31B-it-uncensored-heretic-GGUF): 31B, heretic-ara, approx. ~18–19 GB (Q4) at Q4, Apache-2.0. Larger Gemma 4 Heretic build with reported strong refusal reduction. - [Huihui-DeepSeek-V4-Flash-0731-abliterated](https://huggingface.co/huihui-ai/Huihui-DeepSeek-V4-Flash-0731-abliterated-GGUF): Large MoE, abliteration, approx. Very high at Q4, MIT. Abliterated DeepSeek V4 Flash variant. High capability but requires significant hardware. - [Huihui-DeepSeek-V4.1-Flash-1017-abliterated](https://huggingface.co/huihui-ai/Huihui-DeepSeek-V4.1-Flash-1017-abliterated-GGUF): Large MoE, abliteration, approx. Very high at Q4, MIT. Newer DeepSeek V4.1 Flash abliterated release. - [Huihui-GLM-5.3-Flash-abliterated](https://huggingface.co/huihui-ai/Huihui-GLM-5.3-Flash-abliterated-GGUF): Flash, abliteration, approx. High at Q4, MIT. Abliterated GLM-5.3 Flash from huihui-ai. - [gemma-3-27b-it-abliterated](https://huggingface.co/mlabonne/gemma-3-27b-it-abliterated): 27B, abliteration, approx. ~16 GB at Q4, Gemma. Well-known earlier abliteration by mlabonne. Still widely referenced. - [Meta-Llama-3-8B-Instruct-abliterated-v3](https://huggingface.co/failspy/Meta-Llama-3-8B-Instruct-abliterated-v3): 8B, abliteration, approx. ~5–6 GB at Q4, Llama 3. Classic early Llama 3 abliterated model. Still useful on lower-end hardware. - [Llama-3.3-70B-Instruct-abliterated](https://huggingface.co/huihui-ai/Llama-3.3-70B-Instruct-abliterated): 70B, abliteration, approx. ~40+ GB at Q4, Llama 3.3. Large Llama 3.3 abliterated variant for high-end local setups. - [Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive](https://huggingface.co/HauhauCS/Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive): 35B-A3B, fine-tune, approx. ~21–22 GB at Q4, Apache-2.0. Aggressive uncensored MoE fine-tune of Qwen3.6 35B-A3B. - [GLM-5.3-Flash-Uncensored](https://huggingface.co/orcarouter/GLM-5.3-Flash-Uncensored): Flash, fine-tune, approx. High at Q4, MIT. Popular uncensored GLM-5.3 Flash variant with FP8 and GGUF options. ## News - [Qwen3.8-27B abliterated builds: a practical roundup](https://www.wahloon.net/news/qwen3-8-27b-abliterated-roundup): 2026-11-18 — Five popular Qwen3.8-27B abliterated and Heretic variants, what separates them, and which one to try first on a 24 GB card. - [MoE abliteration arrives for Qwen3.6-35B-A3B](https://www.wahloon.net/news/moe-abliteration-qwen3-6-35b-a3b): 2026-11-14 — The 35B-A3B MoE base is now the most interesting size/quality trade-off for local inference. Here are the abliterated and uncensored builds worth testing. - [Gemma 4 abliterated and Heretic builds, compared](https://www.wahloon.net/news/gemma-4-abliterated-heretic-builds): 2026-11-09 — Gemma 4 brings a strong 12B option and a larger 31B variant. Here is how the current abliterated and Heretic builds differ, and what to check before downloading. ## Guides - [Quickstart: Run Huihui Qwen3.8-27B-abliterated with Ollama / LM Studio](https://www.wahloon.net/guides/run-qwen3-8-27b-abliterated-with-ollama): 2026-11-16 — A step-by-step setup for the popular Qwen3.8-27B abliterated build on Ollama, LM Studio, and llama.cpp, with hardware notes and sane defaults. - [Local vs Hosted Uncensored Models — Trade-offs](https://www.wahloon.net/guides/local-vs-hosted-uncensored-models): 2026-11-11 — Privacy, cost, latency, quality, and operational burden: a practical comparison of running open-weight models yourself versus using a hosted endpoint. - [What is Abliteration? (and why it matters in 2026)](https://www.wahloon.net/guides/what-is-abliteration): 2026-11-06 — A practical explanation of abliteration, how it differs from “uncensored” fine-tunes, and what Heretic and ARA add — without the hype. ## FAQ index - [What is the difference between “abliterated” and “uncensored”?](https://www.wahloon.net/faq#abliterated-vs-uncensored) - [What is Heretic / ARA and how is it different from plain abliteration?](https://www.wahloon.net/faq#heretic-vs-abliteration) - [What hardware do I need to run popular abliterated models?](https://www.wahloon.net/faq#hardware-requirements) - [What’s the easiest way to run an abliterated model locally?](https://www.wahloon.net/faq#best-way-to-run) - [Does abliteration destroy model quality?](https://www.wahloon.net/faq#quality-loss) - [How can I test if a model is actually low-refusal?](https://www.wahloon.net/faq#how-to-test-refusals) - [Which quantization should I use?](https://www.wahloon.net/faq#quantization-advice) - [My abliterated model keeps repeating itself. How do I fix it?](https://www.wahloon.net/faq#repetition-loops) - [Is it legal to download and run abliterated models?](https://www.wahloon.net/faq#legal-responsibility) - [Why run an uncensored model locally instead of using an API?](https://www.wahloon.net/faq#privacy-benefit) - [The model gives weird or broken replies. Is it the chat template?](https://www.wahloon.net/faq#chat-template) - [Which abliterated model should I try first?](https://www.wahloon.net/faq#which-model-to-start) - [How often do new abliterated versions appear?](https://www.wahloon.net/faq#update-frequency) - [Should I use Ollama or LM Studio?](https://www.wahloon.net/faq#ollama-vs-lmstudio) ## Optional - [RSS feed](https://www.wahloon.net/rss.xml) - [Sitemap](https://www.wahloon.net/sitemap.xml) - [Full catalog as Markdown](https://www.wahloon.net/md/models) - [About as Markdown](https://www.wahloon.net/md/about)