Justbackup/Gemma-3-27b-it-Uncensored-HERETIC-Gemini-Deep-Reasoning

VISIONPricing:Input $0.4 / Cached $0.08 / Output $1.2Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kPublished:Aug 24, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Justbackup/Gemma-3-27b-it-Uncensored-HERETIC-Gemini-Deep-Reasoning is a 27 billion parameter Gemma 3-based model, fine-tuned by Justbackup using Unsloth. It features a 32k context length and is designed for deep thinking and uncensored content generation, with image processing capabilities intact. This model excels at reasoning tasks and offers stable performance across a wide temperature range, making it suitable for applications requiring nuanced and unrestricted outputs.

Loading preview...

Model Overview

This model, Justbackup/Gemma-3-27b-it-Uncensored-HERETIC-Gemini-Deep-Reasoning, is a 27 billion parameter Gemma 3 variant fine-tuned for deep thinking and uncensored content generation. It leverages Unsloth for efficient training and maintains full image processing functionality. A key feature is its reasoning capability, which influences both image intelligence and output generation, remaining stable across a wide temperature range (0.1 to 2.5).

Key Capabilities & Features

  • Uncensored Output: Designed to generate content without refusals, though it may require specific directives for highly graphic or explicit material.
  • Deep Thinking: Enhanced reasoning, which can be activated explicitly with "think deeply: prompt" or through system prompts. A dedicated "chat-template-thinking.jinja" is available for always-on thinking.
  • Extended Context: Supports a 128k context window, allowing for processing longer inputs.
  • Image Processing: Image intelligence is intact and fully functional, influenced by the model's reasoning.

Performance & Decensoring

Benchmarks show improved performance over the base uncensored Heretic model across various tasks, including ARC Challenge, HellaSwag, and Winogrande. The model achieves a KL divergence of 0.07 compared to the original Google Gemma-3-27b-it, indicating minimal damage during the decensoring process. Refusals are significantly reduced from 98/100 to 9/100.

Optimal Usage

For smoother operation, especially in chat and roleplay applications, users are advised to set the Smoothing_factor to 1.5 in interfaces like KoboldCpp, oobabooga/text-generation-webui, or Silly Tavern. Detailed guidance on maximizing performance and advanced settings is available in the provided external documentation.