Skip to content

Search wahloon.net

Search models, news, guides and FAQs.

Local-first · open weights · late 2026

Abliterated models.
On your hardware.

Track uncensored open-weight LLMs, compare real hardware requirements, and get them running with Ollama, LM Studio, or llama.cpp. No cloud account required.

Models
18
Guides
3
FAQ answers
14
quickstart.sh

# pull a 27B abliterated build

ollama run huihui_ai/Qwen3.8-abliterated:27b

# or pin an exact quant

ollama run hf.co/huihui-ai/Huihui-Qwen3.8-27B-abliterated-GGUF:Q4_K_M
≈ 16–18 GB VRAMApache-2.0

// signal

Latest intelligence

Release notes, roundups and changes worth knowing about.

All news
3 minNews

MoE abliteration arrives for Qwen3.6-35B-A3B

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.

#qwen#moe#35b-a3b
Read

// catalog

Featured models

18 tracked builds · catalog updated Nov 18, 2026

Full catalog
Qwen27Bfeatured
Abliteration

Huihui-Qwen3.8-27B-abliterated

Qwen/Qwen3.8-27B

One of the most downloaded abliterated Qwen 3.8 27B builds. Strong general performance with significantly reduced refusals.

Approx. VRAM
~16–18 GB (Q4)
License
Apache-2.0
Downloads
4.5M / 30d
#popular#qwen#abliterated#gguf
ollama run huihui_ai/Qwen3.8-abliterated:27b

updated Nov 18, 2026

Qwen35B-A3Bfeatured
Abliteration

Huihui-Qwen3.6-35B-A3B-abliterated-MTP

Qwen/Qwen3.6-35B-A3B

Highly popular MoE abliterated variant. Excellent quality-to-size ratio for the active parameter count.

Approx. VRAM
~21 GB (Q4)
License
Apache-2.0
Downloads
5.2M / 30d
#popular#qwen#moe#abliterated

updated Nov 15, 2026

Every entry links to its Hugging Face repositoryVRAM figures are practical estimates, not guarantees

// start here

Get a model running

Short, practical paths from download to a working chat.

All guides
  1. 01Quickstart: Run Huihui Qwen3.8-27B-abliterated with Ollama / LM StudioA 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.
  2. 02Local vs Hosted Uncensored Models — Trade-offsPrivacy, cost, latency, quality, and operational burden: a practical comparison of running open-weight models yourself versus using a hosted endpoint.
  3. 03What is Abliteration? (and why it matters in 2026)A practical explanation of abliteration, how it differs from “uncensored” fine-tunes, and what Heretic and ARA add — without the hype.

// contribute

Know a model we are missing?

The catalog is a typed data file. Send a Hugging Face link and the method, or open a pull request with the entry.