Back to Gap Map Model components / Document conversion & OCR

MinerU

source available / Overall score: 3.4

Document extraction pipeline that converts PDFs and Office documents into LLM-ready Markdown and JSON, preserving reading order and handling multi-column layouts, tables, formulas and scanned pages through OCR. Published by OpenDataLab and distributed as a Python package, a Docker image and a hosted web application.

OpenDataLab is a project identity under Shanghai AI Laboratory, whose contact address the GitHub org publishes as OpenDataLab@pjlab.org.cn, so the record is filed under the lab rather than under a separate org.

Openness

2 high confidence

2.0

license: MinerU-Open-Source-License(Apache-2.0 plus commercial thresholds
source: public(the repository builds the extraction pipeline the package installs)
core features withheld: no — nothing is withheld from the published tree

MinerU's LICENSE.md states it is Apache-2.0 "subject to additional terms": an organization above 100 million monthly active users or $20 million in monthly revenue needs a separate commercial license, and the grant ends if that or the required attribution is not met. That clause, not anything withheld from the build, is why the repository reads as source-available rather than fully open.

https://api.github.com/repos/opendatalab/MinerU recorded 2026-08-30
Repo metadata - license spdx_id NOASSERTION, archived false, 78,805 stars - for opendatalab/MinerU.
https://raw.githubusercontent.com/opendatalab/MinerU/master/LICENSE.md recorded 2026-10-07
LICENSE.md body: 'MinerU is licensed under Apache License 2.0 and is subject to the additional terms below.' Section 1 requires a separate commercial license where an entity and its affiliates exceed 100 million monthly active users or USD 20 million total monthly revenue; Section 2 imposes a prominent-attribution obligation on online services; Section 3 terminates the license automatically where either is not met.
https://raw.githubusercontent.com/opendatalab/MinerU/master/README.md recorded 2026-10-07
README, read for the distribution and gating question: "MinerU use neural network models for local `basic`, `standard`, and `advanced` parsing", "A Standard server provides `basic`, `standard`, and `advanced` request tiers.", and "Documents are not automatically uploaded to the official service; remote parsing requires explicit configuration." Every quality tier runs from the published package, and the official remote service is an opt-in alternative rather than a carve-out.

Adoption

3 high confidence

3.0

Adoption is measured as PyPI downloads of the mineru package over the trailing month.

https://pypistats.org/api/packages/mineru/recent recorded 2026-08-30
last_month = 269,753 downloads for mineru.

Capability

4 medium confidence

4.0

MinerU parses PDFs and images into one document model that keeps tables, formulas, figures and page layout, rendered as Markdown, HTML or structured JSON, but stops short of Docling, which can also return the fields a user defines in a template. Its MinerU2.5-Pro model ranks fourth on OmniDocBench, a benchmark maintained by MinerU's own team.

https://raw.githubusercontent.com/opendatalab/MinerU/master/README.md recorded 2026-09-29
The raw README for MinerU 4.0 lists input formats "PDF, images, DOC/DOCX, PPT/PPTX, XLS/XLSX, RTF, ODT/ODS/ODP, EPUB", says "one document model supports nine rendering targets: Markdown, HTML, LaTeX, DOCX, EPUB, PDF, Structured Content, and Content List V1/V2", offers "Basic for OCR and model-based parsing", and calls MinerU "especially useful when documents contain OCR text, tables, formulas, figures, or complex page layouts".
https://raw.githubusercontent.com/opendatalab/OmniDocBench/main/README.md recorded 2026-09-29
The OmniDocBench README, published by OpenDataLab, ranks "Comprehensive evaluation of document parsing on OmniDocBench (v1.6_full)" with "TeleOCR" first at "96.91", then "OvisOCR2" at "96.47", "PaddleOCR-VL-1.6" at "96.34" and "MinerU2.5-Pro" at "95.75".

Verified 2026-09-17