chartreuse-verte/prose-rewriter-4b-v1.3

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 29, 2026License:agpl-3.0Architecture:Transformer Open Weights Featherless Exclusive Cold

chartreuse-verte/prose-rewriter-4b-v1.3 is a 4 billion parameter Qwen3-based model specifically designed for paragraph-level prose rewriting. It takes AI-generated text and re-renders it to sound more human, preserving original semantics. This model excels at making large model output less artificial, with improved performance over its predecessor in reducing invention and copying.

Loading preview...

Model Overview

chartreuse-verte/prose-rewriter-4b-v1.3 is a specialized 4 billion parameter model built on Qwen/Qwen3-4B-Base. Its primary function is to rewrite AI-generated prose at the paragraph level, making it sound more natural and human-like while maintaining the original meaning. This model is a successor to prose-rewriter-4b-v1.2, featuring a larger training pool and two new data gates, resulting in significantly reduced "invention" (hallucination) and "copying" rates.

Key Capabilities

  • Prose Rewriting: Transforms AI-generated text into more human-sounding prose.
  • Semantic Preservation: Designed to maintain the original meaning and intent of the input text.
  • Reduced Fabrication: Compared to its predecessor, it invents less and copies less, leading to higher fidelity rewrites.
  • Flexible Length Modes: Supports match (rewrite in place), inflate (cut padded source), and compress (expand flattened source) modes, though match is strongly recommended.
  • Optimized for Fictional Prose: Best suited for narrative registers (first/third person fiction, dialogue).

When to Use This Model

  • Humanizing AI Output: Ideal for refining text generated by other large language models to improve its naturalness.
  • Paragraph-Level Edits: Works best on inputs from a full sentence up to about 80 words; longer inputs should be split.
  • Specific Rewriting Tasks: Use the match mode for general rewriting without length changes. The inflate and compress modes offer more control over output length based on the input's original state.

Limitations

  • Not an Instruct Model: It has a single, specific function and does not respond to general instructions.
  • Fictional Prose Only: Not intended for technical documentation or other non-narrative text types.
  • English Only: Trained exclusively on English narrative prose.
  • AI Detector Ineffectiveness: Will not bypass AI detection tools, as it preserves core linguistic patterns.
  • Short Input Issues: Inputs below 15 words may lead to padding and fabrication.

Finetunes
1 models