OS-Software/gemma-4-26B-A4B-it-qat-q4_0-unquantized-uncensored-heretic-v2

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:2Model Size:26BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 7, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

OS-Software/gemma-4-26B-A4B-it-qat-q4_0-unquantized-uncensored-heretic-v2 is a 26 billion parameter instruction-tuned multimodal Mixture-of-Experts (MoE) model, derived from Google DeepMind's Gemma 4 family, with a 32768 token context length. This version has undergone substantial reduction of its safety alignment using the Heretic v2.0.0.dev0+custom tool, making it uncensored. It is intended for research and experimentation in areas like safety research, alignment studies, and red-teaming, rather than public deployment.

Loading preview...

Model Overview

This model, OS-Software/gemma-4-26B-A4B-it-qat-q4_0-unquantized-uncensored-heretic-v2, is a 26 billion parameter instruction-tuned variant from the Gemma 4 family by Google DeepMind. It is a multimodal Mixture-of-Experts (MoE) model, featuring 25.2B total parameters with 3.8B active parameters, allowing for faster inference comparable to a 4B model. A key differentiator is its decensored nature, achieved by substantially reducing its safety alignment using the Heretic v2.0.0.dev0+custom tool, resulting in 0/100 refusals compared to 100/100 in the original model.

Key Capabilities (Base Gemma 4 Model)

  • Multimodal Input: Processes text and image inputs, with support for variable aspect ratio and resolution images.
  • Reasoning: Designed with configurable thinking modes for step-by-step reasoning.
  • Extended Context: Supports a 256K token context window.
  • Coding & Agentic Capabilities: Improved performance in coding benchmarks and native function-calling support.
  • Native System Prompt Support: Enables more structured and controllable conversations.

Important Notice

This model has significantly reduced safety alignment and is more likely to generate harmful, inaccurate, biased, or inappropriate content. It is explicitly intended for research and experimentation only, including safety research, alignment studies, and red-teaming. Users are solely responsible for evaluating outputs and implementing safeguards, as use is entirely at your own risk.

Finetunes
1 models