darkc0de/Qwen3-0.6B-heretic

TEXT GENERATIONConcurrent Unit Cost:1Model Size:0.8BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 11, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

darkc0de/Qwen3-0.6B-heretic is a 0.8 billion parameter causal language model, a decensored version of Qwen/Qwen3-0.6B, created using Heretic v1.4.0 on an Android device. This model is specifically modified to reduce refusals, showing 6 refusals out of 100 compared to the original model's 53/100. It retains the Qwen3 architecture's ability to switch between thinking and non-thinking modes, making it suitable for applications requiring less restrictive content generation.

Loading preview...

Model Overview

darkc0de/Qwen3-0.6B-heretic is a 0.8 billion parameter causal language model, derived from the Qwen/Qwen3-0.6B base model. This version has been specifically modified using Heretic v1.4.0 to be a "decensored" variant, aiming to reduce content refusals. The modification process was notably performed locally on an Android device.

Key Differentiators & Performance

The primary distinction of this model lies in its significantly reduced refusal rate. Benchmarking shows:

  • Refusals: 6/100 (compared to 53/100 for the original Qwen/Qwen3-0.6B)
  • KL divergence: 0.0045 (relative to the original model)

This indicates a substantial shift in content generation policy, making it more permissive than its base model.

Core Capabilities (inherited from Qwen3)

Despite the decensoring, the model retains the core features of the Qwen3 series, including:

  • Flexible Thinking Modes: Supports seamless switching between a "thinking mode" for complex reasoning (math, code, logic) and a "non-thinking mode" for efficient, general dialogue. This can be controlled via enable_thinking parameter or /think and /no_think tags in prompts.
  • Enhanced Reasoning: Designed for improved performance in mathematics, code generation, and commonsense logical reasoning.
  • Agent Capabilities: Excels in tool calling and integration with external tools, supported by frameworks like Qwen-Agent.
  • Multilingual Support: Capable of handling over 100 languages and dialects for instruction following and translation.

Use Cases

This model is particularly suited for applications where:

  • Reduced content restrictions are desired.
  • Flexible reasoning capabilities are beneficial, allowing dynamic switching between detailed thought processes and direct responses.
  • Agentic workflows requiring tool integration are in scope.
  • Multilingual interaction is a requirement.

Finetunes
1 models