Neura Market
Marker logo

Marker

Free

Convert documents to markdown, JSON, and HTML quickly and accurately.

FreeFree tier
Inputs: image
Type
Open Source
Company
Datalab
Visit Website GitHub (38 stars)

About Marker

Marker is an open-source document conversion tool that accurately transforms PDFs, images, and Office documents (PPTX, DOCX, XLSX) along with HTML and EPUB files into markdown, JSON, chunks, and HTML. It handles complex elements such as tables, forms, equations, inline math, links, references, and code blocks, while removing headers, footers, and other artifacts. The tool is extensible with custom formatting and logic. It optionally integrates with LLMs (Gemini, Claude, OpenAI, etc.) via a hybrid mode for even higher accuracy. Marker achieves 76% overall on the olmocr-bench benchmark (83.5% on born-digital PDFs), outperforming MinerU and docling. It runs on GPU, CPU, or Apple Silicon (MPS) and is licensed under Apache 2.0 (code) with separate model weights licensing.

Key Features

Converts PDF, image, PPTX, DOCX, XLSX, HTML, and EPUB to markdown, JSON, chunks, and HTML
Formats tables, forms, equations, inline math, links, references, and code blocks
Extracts and saves images from documents
Removes headers, footers, and other artifacts
Extensible with custom formatting and logic
Optionally boosts accuracy with LLMs (Gemini, Claude, OpenAI, etc.) via hybrid mode
Works on GPU, CPU, or Apple Silicon (MPS)
High accuracy on olmocr-bench: 76% overall, 83.5% on born-digital PDFs
Fast mode available for speed in addition to balanced and no-OCR modes

Pros & Cons

Pros
  • High accuracy on diverse document types, outperforming similar tools on benchmarks
  • Supports multiple input formats including PDF, Office, HTML, and EPUB
  • Handles complex elements like tables, equations, and forms
  • Optional LLM integration for further accuracy improvements
  • Open source with permissive Apache 2.0 license for code
  • Works on CPU, GPU, and MPS, making it accessible on various hardware
Cons
  • Requires Python 3.10+ and PyTorch setup, which may be heavy for some environments
  • LLM-enhanced mode introduces external API dependencies and potential costs
  • Model weights have restrictive license for commercial entities exceeding $5M funding/revenue
  • Not a native GUI application; primarily used via command line or API

Best For

Preprocessing documents for RAG (Retrieval-Augmented Generation) systemsExtracting structured data from PDFs, invoices, and academic papersConverting scanned documents and images to editable markdownArchiving and digitizing large volumes of documents and formsIntegrating document conversion into data pipelines with custom formatting

FAQ

What input formats does Marker support?
Marker supports PDF, image (JPG, PNG, etc.), PPTX, DOCX, XLSX, HTML, and EPUB files.
Is Marker free to use?
Yes, the code is licensed under Apache 2.0, free for any use including commercial. The model weights have a separate modified AI Pubs Open Rail-M license (free for research, personal use, and startups under $5M funding/revenue).
Can I use Marker with an LLM for higher accuracy?
Yes, by using the --use_llm flag, Marker can leverage LLMs such as Gemini, Claude, OpenAI-compatible models, Azure, Vertex, OpenRouter, or Ollama to improve accuracy on complex documents.
How accurate is Marker according to benchmarks?
Marker scores 76% overall on the olmocr-bench benchmark of 1,403 PDFs, with 83.5% on born-digital PDFs. This places it ahead of MinerU and docling.