MinerU

MinerU is an open-source document parsing tool with 82,000+ stars on GitHub that transforms complex PDFs and Office documents into clean, LLM-ready Markdown and JSON.

Overview

Large language models are only as good as the text you feed them, and real-world documents — academic papers, financial reports, scanned forms — resist simple extraction: multi-column layouts, formulas, tables and figures all break naive parsers. MinerU, from OpenDataLab, is built specifically for this job. It performs layout analysis, OCR, formula recognition to LaTeX, table conversion to HTML, and reading-order reconstruction, then emits Markdown or JSON that preserves the document’s structure for downstream LLM and agent workflows. It handles PDF, DOCX, PPTX and XLSX files, runs locally in Python or via Docker, and also powers a hosted API for teams that prefer not to run models themselves. The project is actively maintained, with repository updates in October 2026, and it has become a default choice in serious RAG ingestion pipelines.

Key Features

• High-fidelity parsing – preserves layout, reading order, headings, lists and paragraphs.
• Formula and tables – converts formulas to LaTeX and tables to structured HTML.
• Built-in OCR – handles scanned documents and image-based PDFs across many languages.
• Multi-format input – PDF plus Word, PowerPoint and Excel documents.
• LLM-ready output – clean Markdown and JSON designed for RAG and agentic workflows.
• Flexible deployment – Python package, command line, Docker or the hosted API.

💳 Pricing

MinerU is free and open source for self-hosting. An official hosted API and online service are also available for teams that want parsing without running the models — check the official site for current terms.

Who Is It For

Developers and data teams building RAG systems, research-paper tooling or document automation, who need faithful structured extraction from complex real-world documents.