Qwen3.6-35B-A3B-Uncensored-Claude-4.6-Genesis-APEX-GGUF: A High-Performance Delta Merge

A new GGUF quantization of a specialized delta merge model is now available, combining the Qwen3.6-35B architecture with Claude 4.6 Genesis APEX reasoning capabilities, offering uncensored outputs and enhanced stability for complex coding and tool-calling tasks.

Model Overview and Architecture

The Qwen3.6-35B-A3B-Uncensored-Claude-4.6-Genesis-APEX-GGUF represents a sophisticated integration of multiple model strengths. Developed via a delta merge process, this model leverages the base efficiency of the Qwen3.6-35B architecture while incorporating the reasoning logic associated with Claude 4.6 Opus.

The release is distributed in GGUF format, making it highly accessible for local LLM deployment via frameworks such as llama.cpp, allowing users to run high-parameter models on consumer-grade hardware through various quantization levels.

Key Technical Enhancements

Coding Stability and Quantization

One of the primary improvements noted is the stability of the model during coding tasks. Even when utilizing the Q4_K_M quantization (referred to as APEX Compact), the model maintains high coherence and reliability, even when paired with complex roleplay System Prompts that typically degrade performance in smaller or more heavily quantized models.

Reasoning and "Thinking" Modes

The model supports dual operational modes: a standard non-thinking mode and a "thinking" mode characterized by a short thinking chain. The latter is recommended for tasks requiring deeper logical deduction, as it integrates the reasoning capabilities derived from the Claude 4.6 Opus lineage.

Uncensored Output and Tool Integration

This iteration is fully uncensored, removing typical safety guardrails to allow for unrestricted generation. Additionally, the model demonstrates improved capabilities in function calling and tool use, enhancing its utility for developers building agentic workflows.

Deployment and Access

The model is available for download via Hugging Face, provided by the user LuffyTheFox. The integration was achieved through a delta merge, blending the specific weights of the source releases to optimize for both reasoning and versatility.

Original Source

LLM GGUF Delta Merge Qwen3.6 Claude 4.6 Reasoning Uncensored AI LocalLLaMA