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

Hugging Face
VISIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 24, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Warm

The richardyoung/Qwythos-9B-Claude-Mythos-5-1M-heretic is a 9 billion parameter reasoning model, a decensored version of Empero AI's Qwythos-9B, built on a Qwen3.5-9B base. It features a 1,048,576-token context window via YaRN rope-scaling and excels in complex reasoning tasks across cybersecurity, biomedical, and quantitative domains. This model is specifically fine-tuned for native function calling and self-correction with tools, making it suitable for retrieval-augmented agentic applications.

Loading preview...

Model Overview

This model, richardyoung/Qwythos-9B-Claude-Mythos-5-1M-heretic, is a 9 billion parameter reasoning model derived from Empero AI's Qwythos-9B, utilizing a deeply uncensored Qwen3.5-9B base. It has been post-trained on over 500 million tokens of high-quality Claude Mythos and Claude Fable traces, with chain-of-thought generated by Empero AI's internal tool, rethink. This process results in a compact and highly capable reasoning model.

Key Capabilities

  • Extended Context Window: Features a 1,048,576-token (1M) context window, enabled by YaRN rope-scaling, making it one of the longest context windows available in its class for whole-codebase reasoning, multi-document research, and long agentic trajectories.
  • Enhanced Reasoning Performance: Demonstrates significant improvements over its base model, with +34 pts on MMLU, +30 pts on gsm8k-strict, and +19 pts on gsm8k-flex under matched evaluation conditions.
  • Native Function Calling: Supports OpenAI/Qwen3.5-style function calling out-of-the-box, requiring no additional wrappers or tool-specific fine-tuning.
  • Self-Correction with Tools: Proven to self-correct and provide source-cited, factually correct answers across diverse and technically demanding prompts when integrated with Python executors and web search tools.
  • Uncensored Design: Intentionally uncensored to engage seriously with complex technical questions in domains like cybersecurity, red-teaming, biology, pharmacology, and clinical medicine, where over-aligned models might refuse or provide boilerplate responses.

Ideal Use Cases

  • Retrieval-Augmented Agentic Systems: Its strong tool-use and self-correction capabilities make it deployment-ready for applications requiring factual verification and integration of external information.
  • Complex Technical Reasoning: Excels in domains such as cybersecurity, red-teaming methodology, biology, biochemistry, pharmacology, clinical medicine, and advanced mathematics.
  • Long-Context Applications: Suitable for tasks requiring analysis of extensive documents, entire codebases, or multi-round agentic interactions due to its 1M token context window.

Finetunes
1 models