ChatDOC/OCRFlux-3B

VISIONPricing:Input $0.32 / Cached $0.016 / Output $1.6Concurrent Unit Cost:1Model Size:3BQuant:BF16Context Size:32kPublished:Jun 12, 2025License:qwen-researchArchitecture:Transformer0.4K Featherless Exclusive Cold

OCRFlux-3B by ChatDOC is a 3 billion parameter multimodal large language model (VLM) fine-tuned from Qwen2.5-VL-3B-Instruct, specifically designed for high-quality document parsing. It excels at converting PDFs and images into clean, readable Markdown text, offering superior parsing quality with an average Edit Distance Similarity (EDS) of 0.967 on the OCRFlux-bench-single benchmark. A key differentiator is its native support for cross-page table and paragraph merging, a feature not commonly found in open-source projects, making it ideal for robust document digitization and information extraction.

Loading preview...

OCRFlux-3B: Advanced Document-to-Markdown Conversion

OCRFlux-3B is a 3 billion parameter multimodal large language model (VLM) developed by ChatDOC, fine-tuned from Qwen2.5-VL-3B-Instruct. It specializes in converting PDFs and images into clean, readable Markdown text, aiming to significantly advance the state-of-the-art in document parsing. The model leverages private document datasets and data from the olmOCR-mix-0225 dataset for its fine-tuning.

Key Capabilities

  • Superior Parsing Quality: Achieves an average Edit Distance Similarity (EDS) of 0.967 on the OCRFlux-bench-single benchmark, outperforming baselines like olmOCR-7B-0225-preview (0.872), Nanonets-OCR-s (0.858), and MonkeyOCR (0.780).
  • Native Cross-Page Merging: Uniquely supports the detection and merging of tables and paragraphs that span across multiple pages, a critical feature for accurate document digitization.
  • Efficient Inference: Designed to run efficiently, even on GPUs like the GTX 3090, making it accessible for various deployment scenarios.
  • Robust Table Parsing: Demonstrates strong performance in table parsing, with an average Tree Edit Distance-based Similarity (TEDS) of 0.861 on the OCRFlux-pubtabnet-single benchmark.

Good For

  • High-Quality Document Digitization: Ideal for converting complex PDFs and images into structured Markdown with high fidelity.
  • Automated Data Extraction: Excellent for scenarios requiring accurate extraction of text and tables from multi-page documents.
  • Applications with Limited Resources: Its 3B parameter size allows for deployment on more modest hardware compared to larger models.
  • Developers needing a comprehensive OCR toolkit: Best utilized via the OCRFlux toolkit for efficient, scalable inference and pipeline management.

Finetunes
1 models