Meet Qwable: The Free Local Model That Thinks Like Claude Fable
A developer on Hugging Face released Qwable, a local 27B model based on Alibaba’s Qwen3.6-27B and fine-tuned on Fable 5-style reasoning traces. The goal is to mimic Fable’s more step-by-step, guided instruction-following while keeping the model runnable on consumer hardware. It ships in GGUF format for tools like LM Studio and llama.cpp, with a Q4 build around 16.5 GB. A second release, Huihui-Qwable-3.6-27b-abliterated, removes refusal behavior using “abliteration,” a method that suppresses the model’s safety-related internal signal rather than bypassing it. That version keeps the model’s reasoning style but stops it from declining harmful or sensitive requests. The standard model is aimed at coding, debugging, and local agent workflows; the abliterated version is intended for research, evaluation, and controlled environments.
