heafy99/Qwen3-4B-Instruct-2507-heretic

TEXT GENERATIONConcurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:May 30, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The heafy99/Qwen3-4B-Instruct-2507-heretic is a 4 billion parameter causal language model, based on the Qwen3-4B-Instruct-2507 architecture, that has been decensored using the Heretic v1.3.0 tool. This model retains the original's enhanced capabilities in instruction following, logical reasoning, and 256K long-context understanding, while significantly reducing refusal rates. It is optimized for applications requiring less restrictive content generation and improved subjective task alignment.

Loading preview...

Model Overview

This model, heafy99/Qwen3-4B-Instruct-2507-heretic, is a decensored version of the Qwen3-4B-Instruct-2507, created using the Heretic v1.3.0 tool. It is a 4 billion parameter causal language model with a native context length of 262,144 tokens, making it suitable for extensive long-context understanding tasks. A key differentiator is its significantly reduced refusal rate (7/100 compared to 100/100 for the original model), achieved through 'abliteration' parameters.

Key Capabilities & Enhancements

Based on the Qwen3-4B-Instruct-2507, this model inherits and builds upon several core strengths:

  • Instruction Following & Reasoning: Demonstrates strong improvements in general capabilities, including logical reasoning, text comprehension, mathematics, science, and coding.
  • Long-Context Understanding: Features enhanced capabilities in processing and understanding contexts up to 256K tokens.
  • Subjective & Open-Ended Tasks: Shows markedly better alignment with user preferences, leading to more helpful responses and higher-quality text generation.
  • Multilingual Support: Offers substantial gains in long-tail knowledge coverage across multiple languages.
  • Agentic Use: Excels in tool-calling capabilities, with recommendations to use Qwen-Agent for optimal performance.

Performance Highlights

Compared to the original Qwen3-4B-Instruct-2507, this 'heretic' version maintains the strong performance across various benchmarks while offering a decensored output. The original model itself showed significant improvements over its predecessor, Qwen3-4B Non-Thinking, in metrics like MMLU-Pro (69.6 vs 58.0), GPQA (62.0 vs 41.7), AIME25 (47.4 vs 19.1), and Creative Writing v3 (83.5 vs 53.6).

When to Use This Model

This model is particularly well-suited for applications where:

  • Unrestricted Content Generation is desired, as it is specifically designed to be decensored.
  • Long-Context Processing is critical, given its 262K native context length.
  • Complex Instruction Following and logical reasoning are required.
  • Subjective and Creative Tasks benefit from its enhanced alignment with user preferences.

Finetunes
1 models