spinochenza/Qwen3.6-27B-uncensored-heretic-v2

VISIONConcurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 2, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The spinochenza/Qwen3.6-27B-uncensored-heretic-v2 is a 27 billion parameter causal language model, based on the Qwen3.6 architecture, with a 32768 token context length. This version is specifically decensored using the Heretic v1.2.0 tool and Magnitude-Preserving Orthogonal Ablation (MPOA) method, resulting in 94% fewer refusals while maintaining original model quality (0.0021 KL divergence). It is optimized for applications requiring less restrictive content generation and excels in agentic coding, knowledge tasks, and multimodal understanding.

Loading preview...

Model Overview

This model, spinochenza/Qwen3.6-27B-uncensored-heretic-v2, is a 27 billion parameter variant of the Qwen3.6 architecture, featuring a 32768 token context length. It has been specifically modified using the Heretic v1.2.0 tool and a variant of the Magnitude-Preserving Orthogonal Ablation (MPOA) method to significantly reduce content refusals.

Key Differentiators

  • Decensored Output: Achieves a remarkable 94% reduction in refusals (6/100 compared to 92/100 for the original model) while preserving the original model's quality with a low KL divergence of 0.0021.
  • Enhanced Agentic Coding: The base Qwen3.6 model is designed for improved agentic coding, handling frontend workflows and repository-level reasoning with greater precision.
  • Multimodal Capabilities: Supports text, image, and video inputs, making it suitable for diverse applications.
  • Thinking Preservation: Includes an option to retain reasoning context from historical messages, which can streamline iterative development and reduce overhead.

Performance

While significantly reducing refusals, the model maintains strong performance across various benchmarks. Its MMLU accuracy is 85.61%, closely matching the original Qwen3.6-27B's 86.65%. It also demonstrates competitive results in coding agent benchmarks like SWE-bench and Terminal-Bench 2.0, and strong performance in knowledge, STEM, and multimodal understanding tasks.

Use Cases

This model is ideal for applications requiring less restrictive content generation, advanced agentic coding, and multimodal understanding. It is particularly well-suited for developers building agents that need to operate with fewer content limitations.

Finetunes
1 models