empero-ai/Qwythos-9B-v2

VISIONConcurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 9, 2026License:apache-2.0Architecture:Transformer0.1K Open Weights Featherless Exclusive Cold

Qwythos-9B-v2 by Empero AI is a 9 billion parameter language model built on the Qwen3.5 architecture, featuring a 1 million token context window. This iteration significantly reduces looping behavior (0% greedy looping rate) while preserving strong reasoning capabilities across MMLU, GSM8K, and ARC benchmarks. It is intentionally uncensored for research in fields like cybersecurity and clinical work, making it suitable for applications requiring deep chain-of-thought and robust technical analysis.

Loading preview...

Qwythos-9B-v2: Enhanced Reasoning with Eliminated Looping

Empero AI's Qwythos-9B-v2 is an improved version of the Qwythos-9B model, designed to maintain its predecessor's deep chain-of-thought reasoning and 1 million token context while addressing critical issues like repetitive output. Built on a Qwen3.5-9B hybrid architecture, this model is specifically optimized for robustness and cleaner behavior.

Key Enhancements & Capabilities

  • Looping Behavior Eliminated: Achieves a 0.0% looping rate under greedy decoding, a significant improvement from 6.7% in the base Qwythos, allowing for more coherent and reliable generation without heavy reliance on repetition_penalty.
  • Reasoning Preserved: Benchmarks like MMLU (83.8% CoT), GSM8K (93.6%), GPQA (49.0%), and ARC-Challenge (96.4%) are maintained at or above v1 levels, confirming no regression in core reasoning abilities.
  • Restored MTP Head: The native multi-token-prediction module is re-integrated, ensuring configuration and weights align for speculative decoding setups.
  • Uncensored for Research: Intentionally designed to be uncensored, supporting research in sensitive technical domains such as cybersecurity, red-teaming, biology, chemistry, pharmacology, and clinical work.
  • 1M-Token Context: Retains the 1,048,576-token context window via YaRN rope-scaling, enabling processing of extensive documents and complex queries.

Ideal Use Cases

  • Advanced Reasoning Tasks: Excels in problems requiring deep logical deduction, such as mathematical word problems and complex analytical queries.
  • Code Generation & Explanation: Capable of generating functional code and providing clear explanations of algorithms.
  • Technical & Scientific Research: Its uncensored nature and robust reasoning make it suitable for exploring sensitive or specialized topics in scientific, medical, and cybersecurity fields.
  • Applications Requiring Coherent Output: Developers needing reliable, non-repetitive text generation, especially under greedy or low-temperature decoding, will benefit from its eliminated looping behavior.

Finetunes
2 models