richardyoung/Qwythos-9B-Claude-Mythos-5-1M-heretic

Hugging Face
VISIONConcurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 24, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Warm

Qwythos-9B-Claude-Mythos-5-1M-heretic by richardyoung is a 9 billion parameter reasoning model, based on a deeply uncensored Qwen3.5-9B base, fine-tuned by Empero AI. It features a 1,048,576-token context window via YaRN rope-scaling and native function calling with self-correction capabilities. This model excels in technically demanding questions across cybersecurity, biomedical, and quantitative reasoning, providing detailed, source-cited answers without refusal.

Loading preview...

Qwythos-9B-Claude-Mythos-5-1M-heretic: A Decensored Reasoning Model

This model, a decensored version of Empero AI's Qwythos-9B-Claude-Mythos-5-1M, is a 9 billion parameter reasoning model built on a deeply uncensored Qwen3.5-9B base. It was post-trained on over 500 million tokens of high-quality Claude Mythos and Fable traces, with chain-of-thought generated by Empero AI's internal rethink tool. The model is designed to engage seriously with technically demanding questions across specialized domains.

Key Capabilities

  • 1,048,576-token context window: Achieved through YaRN rope-scaling, enabling whole-codebase reasoning, multi-document research, and long agentic trajectories.
  • Enhanced Reasoning Performance: Demonstrates significant improvements over its base model, including +34.3 MMLU, +30 GSM8K-strict, and +19 GSM8K-flex points under matched evaluation conditions.
  • Native Function Calling with Self-Correction: Supports OpenAI/Qwen3.5-style function calling out-of-the-box, allowing it to use tools like Python executors and web search to verify specifics and produce source-cited, factually correct answers.
  • Uncensored Engagement: Intentionally uncensored to provide substantive responses in domains like cybersecurity, red-teaming, biology, pharmacology, and clinical medicine, where other models might refuse or hedge.

Good For

  • Complex Reasoning Tasks: Excels in multi-step problem-solving, particularly in quantitative, scientific, and technical fields.
  • Long-Context Applications: Ideal for tasks requiring extensive context, such as analyzing large codebases, synthesizing information from multiple documents, or managing long agentic conversations.
  • Tool-Augmented Workflows: Designed for integration with external tools for enhanced factual accuracy and verification, making it deployment-ready for retrieval-augmented agentic settings.
  • Specialized Technical Domains: Provides detailed and direct answers for cybersecurity, biomedical, and quantitative reasoning questions without typical LLM refusals or disclaimers.