opendatalab/MinerU

100% speed boost!

The ultimate document parser MinerU releases version 3.4. Boasting over 69,000 stars on GitHub, this tool easily converts complex PDFs, Office documents, handwritten text, multi-column layouts, and cross-page tables into LLM-ready Markdown or JSON format in one click.

Key highlights of this update:

🔹 **OCR Accuracy Boost**: The pipeline backend has been upgraded to PP-OCRv6, improving OCR accuracy by approximately 11% on OmniDocBench v1.6, and natively supports OCR recognition across 109 different languages.

🔹 **Doubled Inference Speed**: Optimizations in the OCR inference and processing pipeline have increased overall speed by 100%, significantly enhancing parsing efficiency for batch files and text-heavy documents.

🔹 **Deployment Optimization**: MinerU now features automatic model source selection based on network environments and prioritizes local cache hit checks, preventing repetitive downloads and reducing configuration errors.

MinerU provides three product formats for developers: a zero-install web version, a full-featured desktop client, and an instant API access option. It supports private, fully offline local deployments, and is natively compatible with more than ten domestic AI hardware chips, including Huawei Ascend, Cambricon, Moore Threads, MetaX, Kunlunxin, and Hygon, making it extremely suitable for enterprise-grade Agentic workflows.

https://github.com/opendatalab/MinerU