richardyoung/mythos-9b-merged-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-merged-heretic is an 8 billion parameter uncensored fine-tuned agent model, derived from King3Djbl/mythos-9b-merged using Heretic v1.4.0. Built on the Qwen3-9B architecture, it features a 32768 token context length and is optimized for tool use, shell commands, and complex multi-step reasoning. This model excels in scenarios requiring robust agent capabilities and resistance to refusals, making it suitable for advanced automation and interactive tasks.

Loading preview...

richardyoung/mythos-9b-merged-heretic: An Uncensored Agent Model

This model is an 8 billion parameter uncensored variant of the King3Djbl/mythos-9b-merged model, processed with Heretic v1.4.0. It is built upon the Qwen3-9B architecture and fine-tuned specifically for agentic capabilities, including tool use, shell command execution, and multi-step reasoning. A key differentiator is its significantly reduced refusal rate compared to the original model, with 7 refusals out of 100 queries versus 79/100.

Key Capabilities

  • Decensored Performance: Achieves a 4.5/5 censorship resistance score, providing full answers across 9 out of 10 categories.
  • Advanced Tool-Use: Rated 4.8/5 for its proficiency with shell, code, Docker, Kubernetes, and SQL tools.
  • Robust Reasoning: Scores 4.5/5 in logic, debugging, and system design tasks, supported by a native thinking mode for complex problems.
  • High Context Length: Supports a 32768 token context window, enabling processing of extensive inputs and maintaining conversational coherence.

Good For

  • General Agent Applications: Ideal for tasks requiring autonomous execution and interaction with various systems.
  • Complex Multi-Step Reasoning: Excels in scenarios demanding logical deduction and problem-solving over multiple steps.
  • Uncensored Content Generation: Suitable for use cases where models need to provide direct answers without inherent refusal mechanisms.
  • Developer Tooling: Particularly effective for code generation, shell scripting, and interacting with development environments.

Finetunes
1 models