georgehenney/Qwen3-8B-heretic

TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:32kPublished:Nov 17, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

georgehenney/Qwen3-8B-heretic is an 8.2 billion parameter causal language model, a decensored version of Qwen/Qwen3-8B created using Heretic v1.0.1. It features a 32,768 token context length, extendable to 131,072 tokens with YaRN, and supports dynamic switching between 'thinking' and 'non-thinking' modes for varied task efficiency. This model is optimized for enhanced reasoning, instruction-following, and agent capabilities, particularly excelling in scenarios where reduced refusal rates are desired.

Loading preview...

Model Overview

georgehenney/Qwen3-8B-heretic is an 8.2 billion parameter causal language model, derived from Qwen/Qwen3-8B and processed with Heretic v1.0.1 to reduce content refusal rates. While the original Qwen3-8B model had 86/100 refusals, this 'heretic' version shows a significantly lower refusal rate of 8/100, making it suitable for use cases requiring less restrictive content generation.

Key Capabilities

  • Decensored Output: Achieves a substantially lower refusal rate compared to its base model, offering more unconstrained text generation.
  • Dual-Mode Operation: Inherits Qwen3's unique ability to switch between a 'thinking mode' for complex logical reasoning, math, and code generation, and a 'non-thinking mode' for efficient general-purpose dialogue.
  • Extended Context Length: Natively supports a 32,768 token context, which can be extended up to 131,072 tokens using the YaRN method for processing very long texts.
  • Enhanced Reasoning & Agentic Abilities: Excels in instruction-following, creative writing, role-playing, and multi-turn dialogues, with strong capabilities for integrating external tools.
  • Multilingual Support: Supports over 100 languages and dialects for instruction following and translation.

Good For

  • Applications requiring a large language model with reduced content moderation and fewer refusals.
  • Tasks demanding advanced reasoning, mathematical problem-solving, and code generation, leveraging its 'thinking' mode.
  • Creative writing, role-playing, and complex multi-turn conversational agents.
  • Scenarios needing long context processing, such as document analysis or extended dialogues.
  • Multilingual applications requiring robust instruction following and translation across many languages.