Qwen3.5-9B-Claude-4.6-Opus-Uncensored-Distilled-GGUF
A developer merges two popular models to create a 9B parameter AI with zero refusals and enhanced creativity.
An independent developer, known as LuffyTheFox, has released a novel open-source language model by merging two existing fine-tunes of Alibaba's Qwen 3.5 9B model. The goal was to solve common frustrations with the base model, such as excessive refusals, lack of creativity, and repetitive 'thinking loops.' To achieve this, they downloaded two popular community variants: the aggressively uncensored 'HauhauCS' model and the 'Jackrong' model, which was distilled to mimic the reasoning patterns of Anthropic's high-end Claude Opus 4.6.
Using a custom script written with the help of Claude Opus itself, the developer performed a tensor-level merge. They extracted the modified tensors responsible for the HauhauCS model's lack of censorship and combined them with the tensors from the Jackrong model that encode Claude-like reasoning. The final model is packaged in the efficient GGUF format, includes a custom system prompt baked into the file, and is optimized to run locally on hardware like a 12GB NVIDIA RTX 3060. Initial testing in LM Studio with specific parameters (Temperature 0.7, Top K 20) reports zero refusals and highly creative, coherent outputs for its size, making it a potent tool for unrestricted creative writing and AI image generation prompt engineering.
- Model merges an uncensored Qwen 3.5 9B variant with a Claude Opus 4.6 reasoning-distilled version via tensor extraction and merging.
- Designed for local execution on consumer GPUs (e.g., RTX 3060 12GB) and optimized for creative tasks like roleplay and prompt crafting.
- Developer reports 'zero refusals' and enhanced creativity, addressing common issues of refusal and repetitive loops in the base model.
Why It Matters
Demonstrates the power of community model merging to create specialized, uncensored AI tools that run efficiently on local hardware for creative professionals.