Marker is a document conversion tool that transforms PDFs, images, and other document formats into markdown, JSON, and HTML output with high accuracy.
The tool addresses the challenge of extracting structured content from diverse document types by combining layout detection with optical character recognition. It handles complex document elements including tables, forms, mathematical equations, code blocks, and images while removing extraneous content like headers and footers. The system operates across multiple hardware configurations—GPU, CPU, or Apple Silicon—making it adaptable to different deployment environments. For users requiring higher accuracy, an optional LLM integration can refine output by merging tables across pages, properly formatting complex structures, and extracting form values, with support for multiple LLM providers including Gemini, Claude, OpenAI, Azure, and Ollama.
Marker suits organizations processing large volumes of documents that need reliable structured extraction without manual intervention. The tool works across all languages and document types, from born-digital PDFs to scanned images and office formats like PPTX, DOCX, and XLSX. It performs competitively on standard benchmarks, scoring ahead of comparable open-source tools while maintaining faster processing speeds than larger vision language models. Teams with specific formatting requirements can extend the tool with custom logic. The project also offers a managed platform alternative for high-volume workloads, which runs an improved model variant with enhanced accuracy and includes infrastructure management for batch processing at scale.
Development activity shows consistent engagement with the codebase through regular updates and refinements to the core conversion pipeline. The project maintains active community channels and provides accessible documentation alongside a public playground for testing functionality. Ongoing work addresses edge cases and performance optimization across different document types and hardware configurations.
