DavidAU/OpenAI-gpt-oss-20B-GPT5.1-5.2-DISTILL-Heretic-Uncensored-MXFP4

TEXT GENERATIONConcurrency Cost:1Model Size:20BQuant:FP8Ctx Length:32kPublished:Feb 7, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

The DavidAU/OpenAI-gpt-oss-20B-GPT5.1-5.2-DISTILL-Heretic-Uncensored-MXFP4 is an experimental 20 billion parameter language model, distilled from GPT 5.1 and 5.2 datasets and compressed to MXFP4 format. It features a 128k context length and is fully uncensored, designed to generate explicit or sensitive content without refusal. This model is specifically fine-tuned for creative and uncensored content generation, requiring specific directives for optimal output quality.

Loading preview...

Model Overview

DavidAU/OpenAI-gpt-oss-20B-GPT5.1-5.2-DISTILL-Heretic-Uncensored-MXFP4 is an experimental 20 billion parameter language model, developed by DavidAU. It is a distillation of GPT 5.1 and 5.2 datasets, with its reasoning capabilities specifically replaced and enhanced by these versions. The model is provided in an MXFP4 compressed format, with a BF16 version potentially following.

Key Characteristics

  • Uncensored Design: This model is fully uncensored, achieved through an "Abliteration" process followed by fine-tuning via Unsloth. It is designed not to refuse requests for sensitive or explicit content.
  • Context Length: Features a substantial 128k token context window, allowing for extended interactions and complex prompts.
  • Content Generation: While uncensored, the model may require specific directives (e.g., using slang, including desired terms) to generate graphic, cursing, or explicit content at the expected intensity. Without such guidance, the output might be "bland" compared to other uncensored models.
  • Experimental Status: Currently a work in progress, indicating that not all functionalities may operate perfectly.

Recommended Usage

  • Creative & Uncensored Content: Ideal for use cases requiring the generation of horror, swear words, or x-rated content without refusal.
  • System Prompts: Optional system prompts are provided to enhance thinking and output generation, including a "business" prompt and a "Joker" persona prompt for creative scenarios.
  • Optimal Settings: For smoother operation in interfaces like KoboldCpp, oobabooga/text-generation-webui, or Silly Tavern, setting a "Smoothing_factor" (or "Smoothing") to 1.5 is recommended. Increasing repetition penalty to 1.1-1.15 can also be beneficial if smoothing is not used. Detailed guidance on maximizing performance and specific settings for various use cases is available here.

Finetunes
1 models