DavidAU/Qwen3-4B-Thinking-2507-Claude-4.5-Opus-High-Reasoning-Distill-Heretic-Abliterated

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kPublished:Dec 9, 2025Architecture:Transformer0.0K Warm

DavidAU/Qwen3-4B-Thinking-2507-Claude-4.5-Opus-High-Reasoning-Distill-Heretic-Abliterated is a Qwen3-based language model with a 256K context window, specifically modified using the Heretic v1.0.1 method to significantly reduce refusal rates from 98/100 to 14/100. This model maintains a low KL divergence of 0.01, indicating minimal damage to its root state during the de-censoring process. It is designed for uncensored, honest responses across all use cases, including those typically refused by other models.

Loading preview...

Model Overview

DavidAU/Qwen3-4B-Thinking-2507-Claude-4.5-Opus-High-Reasoning-Distill-Heretic-Abliterated is a Qwen3-based model engineered for uncensored and honest responses across a broad spectrum of use cases. It features a substantial 256K context window.

Key Differentiators

  • Significantly Reduced Refusal Rate: Modified using the Heretic v1.0.1 method, its refusal rate has been drastically lowered from an original 98/100 to just 14/100, aiming for uninhibited content generation.
  • High Fidelity to Original State: Achieves a low KL divergence of 0.01, ensuring that the de-censoring process has not "brain damaged" the model, preserving its original performance characteristics.
  • Uncensored Output: Designed to generate content without refusals, including sensitive or explicit material, though it may require explicit directives (e.g., specific slang or graphic terms) to achieve desired intensity.

Optimal Usage

  • Adjusting for Smoother Operation: Users are advised to set a "Smoothing_factor" to 1.5 in interfaces like KoboldCpp, oobabooga/text-generation-webui, or Silly Tavern for improved chat and roleplay performance.
  • Controlling Output Intensity: While uncensored, the model might need specific instructions or keywords to generate content at expected graphic or explicit levels, as its default settings can be "tame" without direction.

This model is part of the broader Qwen3-24B-A4B-Freedom-Thinking-Abliterated-Heretic-NEO series, emphasizing user freedom and unconstrained generation.