arnomatic/Mistral-Small-3.2-24B-Instruct-2506-Text-Only-heretic

Hugging Face
TEXT GENERATIONConcurrent Unit Cost:2Model Size:24BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Dec 14, 2025Architecture:Transformer Featherless Exclusive Warm

arnomatic/Mistral-Small-3.2-24B-Instruct-2506-Text-Only-heretic is a 24 billion parameter instruction-tuned language model based on the Mistral-Small-3.2 architecture, featuring a 32768 token context length. This model is a decensored version of the original, created using the Heretic tool, significantly reducing refusals from 97/100 to 4/100. It is specifically designed for use cases requiring less restrictive content generation, maintaining the standard Mistral architecture without a vision encoder.

Loading preview...

Overview

arnomatic/Mistral-Small-3.2-24B-Instruct-2506-Text-Only-heretic is a 24 billion parameter instruction-tuned language model derived from the mistralai/Mistral-Small-3.2-24B-Instruct-2506 base model. This version is a decensored variant, created using the Heretic v1.1.0 tool, specifically targeting use cases where content restrictions are undesirable.

Key Capabilities

  • Significantly Reduced Refusals: The primary differentiator is its drastically lowered refusal rate, dropping from 97/100 in the original model to just 4/100 in this 'heretic' version. This makes it suitable for generating content that might otherwise be filtered or blocked by standard moderation.
  • Standard Mistral Architecture: It retains the core "Mistral" architecture, offering a robust foundation for various NLP tasks.
  • Text-Only: This model is explicitly designed without a vision encoder, focusing solely on text-based instruction following.

Good for

  • Applications requiring uncensored or less restricted text generation.
  • Scenarios where the original model's content filters are too aggressive.
  • Developers seeking a powerful 24B parameter model for general instruction-following tasks without vision capabilities.