kawaimasa/Wanabi-Novelist-24B

TEXT GENERATIONPricing:Input $0.7 / Cached $0.04 / Output $1.16Concurrent Unit Cost:2Model Size:24BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Oct 12, 2025License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

kawaimasa/Wanabi-Novelist-24B is a 24 billion parameter Japanese large language model, fine-tuned specifically for novel writing assistance. Based on a vision-removed variant of Mistral-Small-3.1-24B-Base-2503, it is designed to integrate with the Project Wannabe desktop application for generating novel text, continuations, and ideas. This model excels at structured creative writing tasks in Japanese, offering control over elements like rating and dialogue proportion.

Loading preview...

Wanabi-Novelist-24B: Japanese Novel Writing Assistant

Wanabi-Novelist-24B is a 24 billion parameter Japanese large language model, meticulously fine-tuned for novel writing support. It is built upon a modified version of Mistral-Small-3.1-24B-Base-2503 (with vision capabilities removed) and is primarily intended for use with the dedicated desktop application, Project Wannabe. This model is not a general-purpose instruction-following LLM but specializes in structured creative text generation.

Key Capabilities (via Project Wannabe)

  • Text Generation (GEN): Generates initial novel sections or new scenes based on metadata (title, synopsis, plot) or freely.
  • Continuation Generation (CONT): Produces natural continuations of existing text, guided by context and "author's notes" for specific plot directions.
  • Idea Generation (IDEA): Expands fragmented information (keywords, genre) into comprehensive novel ideas, including title, synopsis, plot, and settings.

Unique Features & Enhancements

  • Refreshed Dataset: Features a completely rebuilt dataset, incorporating approximately 24,000 web novels and 7,000 Aozora Bunko texts for diverse writing styles. Data quality improvements ensure proper formatting like line breaks and indentations.
  • Enhanced Control: Allows users to specify content rating (general/r18) and the proportion of dialogue for generated text, offering fine-grained control over output.
  • Improved Idea Task: The idea generation function has been significantly enhanced to expand brief or incomplete inputs into detailed and varied novel concepts.

Important Considerations

  • Specialized Use: This model is highly specialized for novel writing within the Project Wannabe ecosystem and does not offer general instruction-following capabilities.
  • Context Length: While the base model supports 128K context, this fine-tuned version was trained with a maximum context of 32K. Performance may degrade with longer contexts.
  • Potential Biases: Due to training data characteristics, generated content may exhibit biases in genre, expression, or plot development.

Finetunes
2 models