ScottzillaSystems/Cydonia-24B-v4.1

TEXT GENERATIONConcurrency Cost:2Model Size:24BQuant:FP8Ctx Length:32kPublished:Feb 24, 2026Architecture:Transformer Cold

Cydonia-24B-v4.1 by TheDrummer is a 24 billion parameter language model based on the Mistral v7 Tekken architecture, featuring a 32768 token context length. This model is highly praised for its exceptional writing capabilities, producing descriptive and nuanced prose without being overly verbose. It excels in creative writing, roleplay, and maintaining focus over extended conversations, making it suitable for applications requiring high-quality, consistent text generation.

Loading preview...

Cydonia 24B v4.1: Enhanced Writing and Roleplay

Cydonia 24B v4.1, developed by TheDrummer, is a 24 billion parameter model built upon the Mistral v7 Tekken architecture. This iteration of the Cydonia series is specifically noted for its significant advancements in text generation quality, offering a 32768 token context length.

Key Capabilities

  • Exceptional Prose Quality: Users consistently highlight its ability to generate descriptive, nuanced, and engaging prose that avoids common pitfalls like purple prose or repetitive phrasing.
  • Strong Focus and Consistency: The model demonstrates excellent memory and focus, even in longer chat contexts (up to 12K-16K tokens), making it reliable for extended interactions.
  • Creative Writing and Roleplay: It excels at understanding and adapting to specific narrative requirements, accurately picking up on roleplay cues and maintaining character consistency.
  • Non-Mistral 'Feel': Despite its base architecture, users report that Cydonia 24B v4.1 does not exhibit typical 'Mistral-isms,' suggesting a unique fine-tuning that enhances its natural language generation.

Good For

  • Applications requiring high-quality, creative text generation.
  • Interactive storytelling and role-playing scenarios.
  • Long-form content creation where maintaining context and nuanced writing is crucial.
  • Use cases where a model's ability to avoid repetitive or overly generic output is valued.