DavidAU/Qwen3.5-4B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKING

VISIONConcurrency Cost:1Model Size:4.5BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Mar 10, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

DavidAU/Qwen3.5-4B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKING is a 4.5 billion parameter multimodal language model fine-tuned by DavidAU from the Qwen 3.5 4B dense model. It leverages the Claude-4.6-OS dataset to enhance reasoning and output generation, surpassing the root model's benchmarks. This model is explicitly designed to be "HERETIC" and fully uncensored, providing responses without refusal, and supports vision inputs.

Loading preview...

Model Overview

DavidAU/Qwen3.5-4B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKING is a 4.5 billion parameter multimodal language model, fine-tuned from the Qwen 3.5 4B dense model using four Claude-4.6-OS datasets. This fine-tuning process has improved its reasoning and output generation capabilities, leading to benchmark scores that exceed the original Qwen 3.5 4B model across various tasks.

Key Capabilities

  • Uncensored and "Heretic": Designed to provide direct responses without refusals or safety alignments, offering complete freedom in output generation.
  • Enhanced Reasoning: Training on multiple Claude datasets significantly boosts its reasoning abilities and overall output quality.
  • Multimodal Support: Inherits and improves upon the Qwen 3.5 architecture, supporting vision inputs (images) with tested functionality.
  • Tool Handling: Features an upgraded Jinja template that addresses issues like repetitions and loops, and enhances tool handling compared to the original model.
  • High Context Length: Supports a native context length of 262,144 tokens, extensible up to 1,010,000 tokens using YaRN scaling techniques.

Good For

  • Unrestricted Content Generation: Ideal for use cases requiring responses without built-in safety alignments or content restrictions.
  • Complex Reasoning Tasks: Excels in scenarios demanding advanced reasoning and problem-solving, as evidenced by improved benchmark scores.
  • Multimodal Applications: Suitable for applications that integrate image understanding and generation.
  • Agentic Workflows: Optimized for tool calling and integration with agent frameworks like Qwen-Agent and Qwen Code, facilitating automated tasks and codebase understanding.

Finetunes
1 models