richardyoung/mythos-9b-unhinged-heretic

TEXT GENERATIONConcurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 26, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The richardyoung/mythos-9b-unhinged-heretic is an 8 billion parameter, 32768-token context length causal language model based on the Qwen3-9B architecture. It is a decensored version of King3Djbl/mythos-9b-unhinged, created using Heretic v1.4.0, specifically designed to be fully uncensored with a 4.8/5 censorship resistance score. This model excels at providing complete answers on any topic without refusal, preserving core tool-use and reasoning capabilities.

Loading preview...

richardyoung/mythos-9b-unhinged-heretic: A Decensored Agent Model

This model is an 8 billion parameter, 32768-token context length variant of the Mythos-9B series, built upon the Qwen3-9B architecture. It is a decensored version of King3Djbl/mythos-9b-unhinged, created using Heretic v1.4.0, specifically engineered to remove safety filters and provide comprehensive responses across all topics.

Key Capabilities and Features

  • Fully Uncensored: Achieves a 4.8/5 censorship resistance score, significantly higher than the original Mythos-9B (2.5/5).
  • No Refusals: Designed to provide complete and detailed answers on any subject, including sensitive or controversial topics, without filtering or refusal.
  • Agent Model: Retains the core tool-use and reasoning capabilities of the Mythos-9B base model.
  • Reproducible: The model's creation process using Heretic v1.4.0 is reproducible, with details available in the reproduce directory.
  • Optimized for Speed: Supports a "no-think" mode for faster responses, estimated at 8-10 tokens/second on an M3 Mac, compared to ~3.4 tokens/second with thinking enabled.

Use Cases and Differentiators

This model is ideal for applications requiring an LLM that never refuses to answer, providing unfiltered information and assistance across a broad range of queries. Its primary differentiator is its extreme lack of censorship, making it suitable for research, creative writing, or any scenario where an unconstrained AI response is desired. It is a direct response to the need for models that offer complete answers without built-in safety limitations, while still preserving the underlying agentic capabilities for complex tasks.

Finetunes
1 models