A developer named Mia (Mia-AiLab on Hugging Face) created Qwable 27B, a full fine-tune of Alibaba's Qwen3.6-27B base, trained on a Fable 5-style reasoning dataset to replicate the structured, deliberate thinking style of Anthropic's newest flagship model. The abliterated version removes the model's built-in refusal behavior by surgically modifying its weights using llama.cpp's cvector-generator. Both models run locally, cost nothing per query, and require neither Anthropic's API nor its mandatory policies. Anthropic spent last week apologizing for Fable 5's invisible safeguards, and then the U.S. government ordered the model pulled for all foreign nationals over a disputed jailbreak finding. Qwable runs in GGUF format and fits in roughly 16.5 GB in its Q4 quantized build. It sends nothing to Anthropic's servers, which matters given that Fable 5 required mandatory 30-day data retention on all traffic, even for enterprise customers who previously had zero-retention agreements. Shortly after Qwable appeared on Hugging Face, huihui-ai, an open-source contributor known for uncensored GGUF releases, took the abliterated version further.