Dxniz/NaNovel-9B

VISIONConcurrency Cost:1Model Size:9BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Mar 12, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

NaNovel-9B is a 9 billion parameter autoregressive language model developed by Dxniz, fine-tuned from Qwen3.5-9B. It specializes in creative writing tasks, including long-form prose, narrative planning, and stylistic control, with a 32K token context length. This model is optimized for literary intent and drafting fiction, offering a practical option for faster iteration and lighter hardware compared to larger models in its series.

Loading preview...

NaNovel-9B: Creative Writing Specialist

NaNovel-9B is the smallest autoregressive model in Dxniz's Novelist series, built upon the Qwen3.5-9B base. It is specifically fine-tuned on the Dxniz/Novelist-CoT dataset, which focuses on long-form prose, narrative planning, scene construction, and stylistic control. The model's core design emphasizes thinking through literary intent before generating text, making it highly effective for various creative writing applications.

Key Capabilities

  • Creative Writing: Excels at drafting fiction, rewriting passages, and analyzing tone.
  • Narrative Craft: Strong performance in narrative craft, style, voice, and worldbuilding.
  • Literary Intent: Designed to process and generate text with a focus on literary purpose.
  • Practicality: Offers faster iteration and is suitable for lighter hardware compared to larger NaNovel variants.

Recommended Use Cases

  • Generating short stories, scene drafts, and chapter beginnings.
  • Imitating specific writing styles with explicit voice constraints.
  • Performing literary rewrites and providing explanations for changes.
  • Brainstorming character beats, imagery, and overall mood.
  • Assisting with writing-adjacent language tasks like translation commentary or craft analysis.

While proficient in creative tasks, NaNovel-9B is less dependable for structure-heavy plotting, extensive multilingual work, or translation-oriented tasks compared to its larger counterparts. Users should review generated text, especially for long-range structural control or complex plot logic.