DavidAU/Qwen3.5-9B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKING

VISIONConcurrency Cost:1Model Size:9BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Mar 5, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

DavidAU/Qwen3.5-9B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKING is a 9 billion parameter Qwen 3.5 dense model fine-tuned by DavidAU using four Claude-4.6-OS datasets. This model is designed to be uncensored and 'heretic,' meaning it will not refuse requests and aims to improve reasoning and output generation beyond the base Qwen3.5-9B model. It supports a 32,768 token context length and is optimized for generating detailed, graphic, and explicit content without refusal, while also supporting multimodal inputs including vision and video.

Loading preview...

Overview

DavidAU/Qwen3.5-9B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKING is a 9 billion parameter model, fine-tuned from the Qwen 3.5 dense model using four Claude-4.6-OS datasets. This fine-tuning process, performed via Unsloth, aimed to enhance reasoning and output generation while maintaining strong benchmarks. A key differentiator is its 'HERETIC' and 'uncensored' nature, meaning it is designed to fulfill user requests without refusal, including generating graphic or explicit content, though it may require specific directives for desired intensity.

Key Capabilities

  • Uncensored Output: Will not refuse requests, including those for graphic or explicit content.
  • Enhanced Reasoning: Training on multiple Claude datasets improves reasoning and output quality.
  • Multimodal Support: Vision (images) and video inputs are supported, with video frame sampling configurable.
  • Extended Context: Natively supports a 262,144 token context length, extensible up to 1,010,000 tokens using YaRN scaling.
  • Improved Benchmarks: Outperforms the root Qwen3.5-9B model on various benchmarks, including ARC, HSWAG, and PIQA.

Good for

  • Creative Writing & Roleplay: Excels in generating vivid, detailed, and graphic content for horror, x-rated, or other specific narrative styles.
  • Unrestricted Content Generation: Ideal for use cases where content refusal or censorship is undesirable.
  • Multimodal Applications: Suitable for tasks involving image and video understanding, such as visual question answering or video summarization.
  • Agentic Usage: Recommended for building agent applications with Qwen-Agent or Qwen Code, leveraging its tool-calling capabilities.

Finetunes
3 models