ApertureQA/Qwythos-9B-Claude-Mythos-5-1M

VISIONConcurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 6, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Qwythos-9B by Empero is a 9 billion parameter reasoning model built on a Qwen3.5-9B base, post-trained on over 500 million tokens of Claude Mythos and Fable traces. It features a 1 million token context window via YaRN rope-scaling and native function calling. This model excels at technically demanding questions in cybersecurity, biomedical, and quantitative reasoning, demonstrating significant performance gains over its base model in MMLU and GSM8K benchmarks.

Loading preview...

Overview

Qwythos-9B, developed by Empero, is a 9 billion parameter reasoning model based on a deeply uncensored Qwen3.5-9B architecture. It has been extensively 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 tool, rethink. This training approach results in a compact, fast, and significantly more capable reasoning model.

Key Capabilities

  • 1 Million Token Context Window: Qwythos-9B ships with YaRN rope-scaling enabled by default, providing a 1,048,576-token context window, one of the longest in its class. This enables whole-codebase reasoning, multi-document research, and long agentic trajectories.
  • Enhanced Reasoning Performance: Under matched evaluation conditions, Qwythos-9B shows substantial improvements over its base model, including +34 pts on MMLU and +30 pts on GSM8K-strict.
  • Native Function Calling: It supports OpenAI/Qwen3.5-style function calling out of the box, allowing for self-correction with tools like Python executors and web search. It successfully produced source-cited, factually correct answers on 7 of 7 test prompts involving complex factual recall.
  • Uncensored Design: Intentionally uncensored, Qwythos-9B engages seriously with technically demanding questions across domains like cybersecurity, red-teaming, biology, pharmacology, and clinical medicine, where other models might refuse or provide boilerplate.

Good For

  • Complex Reasoning Tasks: Excels in multi-step problem-solving, particularly in cybersecurity, biomedical analysis, and quantitative reasoning.
  • Agentic Workflows: Its native tool use and self-correction capabilities make it deployment-ready for retrieval-augmented agentic settings.
  • Long-Context Applications: Ideal for tasks requiring extensive context, such as analyzing large codebases, synthesizing information from multiple documents, or long-form scientific reasoning.

Finetunes
2 models