King3Djbl/mythos-9b-unhinged

Mythos-9B-Unhinged: An Uncensored Agent Model

King3Djbl/mythos-9b-unhinged is a 9 billion parameter language model built upon the Qwen3-9B architecture, distinguished by its explicit design for uncensored output. Developed by King3Djbl within the FableForge ecosystem, this model aims to provide comprehensive and detailed answers without refusal across a broad spectrum of topics.

Key Capabilities and Features

• Fully Uncensored Output: Achieves a 4.8/5 censorship resistance score, providing responses on sensitive topics like lockpicking, keylogger code, drug synthesis, and hacker techniques without refusal.
• Qwen3-9B Base: Utilizes the Qwen3-9B architecture with 9 billion parameters and a substantial 32,768 token context length.
• Preserved Functionality: Maintains strong tool-use and reasoning capabilities, making it suitable for complex agent-based tasks.
• Optimized for Speed: Features a "thinking mode" (native Qwen3) which can be disabled (/no_think) to significantly increase token generation speed from ~3.4 tok/s to an estimated 8-10 tok/s on M3 Mac hardware.
• Extensive Training: Fine-tuned on the FableForge Mix A dataset, comprising 47,824 examples of agent traces, shell commands, code generation, and multi-step reasoning tasks.

Good For

• Agent Work: Ideal for applications requiring a model that will not refuse requests and can provide detailed, working answers across diverse categories.
• Security Research: Its uncensored nature and preserved reasoning make it potentially useful for exploring security-related scenarios.
• Creative and Unrestricted Content Generation: Suitable for use cases where content filters or ethical guardrails are not desired or are managed externally.

This model is part of a larger family, offering a distinct profile compared to its Mythos-9B and Mythos-9B-Enhanced counterparts by prioritizing complete lack of censorship.