--- title: "MinerU" type: "AI Tool" url: "https://aidemos.com/tools/mineru" description: "We tested MinerU on complex PDFs and kept figures, charts, and reading order intact, but tables, punctuation, and mojibake broke on dense files." category: "developer-tools" website: "https://github.com/opendatalab/MinerU" published: "2026-08-13T05:01:38.301049+00:00" updated: "2026-08-13T05:01:38.301049+00:00" --- # MinerU Open-source PDF-to-markdown that keeps figures and charts in place, but tables and punctuation can get messy on complex files. ## TL;DR Verdict **Strong at preserving visuals and structure, but table fidelity and punctuation cleanup are uneven.** **Where it wins:** - You need charts, figures, logos, or photos preserved as real embedded images in markdown. - You want open-source PDF-to-markdown with decent heading order and OCR on scanned pages. - You can tolerate slower processing in exchange for local/self-hosted control and image retention. **Main limitation:** You need punctuation-perfect OCR with no mojibake in em-dashes, quotes, or apostrophes. **Pricing:** Open Source $0 `Open source` · `Figures retained` · `HTML tables` · `Slow on long PDFs` **Website:** [Visit MinerU](https://github.com/opendatalab/MinerU) > **Strong at preserving visuals and structure, but table fidelity and punctuation cleanup are uneven.** > > MinerU is a solid self-hosted choice when your top priority is preserving figures, charts, logos, and reading order in a markdown output. It is less reliable on dense tables, and recurring mojibake plus slow runtimes make it a mixed fit for production pipelines that need exact text and fast throughput. ## Feature-by-Feature Breakdown ### Embedded Visual Preservation **Verdict:** Strong Keeps charts, figures, photos, logos, and other embedded visuals as real image references in the markdown output. In the hybrid earnings report it preserved financial charts and the CEO portrait; in the scanned paper it preserved the figure and cover logo as images. **Input:** > **Image** **Output:** > **Image** **Input:** > **Image** **Output:** > **Image** **Input:** > **Image** **Output:** > **Image** **Input:** > **Image** **Output:** > **Image** **Bottom line:** This is MinerU's strongest capability: visuals are consistently preserved as real images instead of placeholders. ### OCR Text Extraction **Verdict:** Mixed Converts native text and scanned pages into markdown text while keeping most content readable. It handled the hybrid earnings CEO letter, the Japanese/English management-plan excerpt, and the scanned paper's title/header text, though punctuation and some characters can be corrupted by mojibake. **Input:** > **Image** **Output:** > **Image** **Input:** > **Image** **Output:** > **Image** **Input:** > **Image** **Output:** > **Image** **Bottom line:** Text extraction is strong overall, but punctuation cleanup is inconsistent and exact OCR fidelity is not perfect. ### Table Reconstruction **Verdict:** Mixed Rebuilds tables with real rowspan/colspan HTML structure in markdown. It works well for many multi-level financial tables, but some dense or merged-cell tables can still break into misaligned, truncated, or collapsed rows. **Input:** > **Image** **Output:** > **Image** **Input:** > **Image** **Output:** > **Image** **Input:** > **Image** **Output:** > **Image** **Input:** > **Image** **Output:** > **Image** **Input:** > **Image** **Output:** > **Image** **Bottom line:** Good on simpler or well-structured tables, but brittle on the exact financial and segment tables that matter most. ### Reading-Order and Heading Preservation **Verdict:** Strong Keeps section order and heading hierarchy mostly intact across mixed digital and scanned PDFs. It preserved the scanned paper's title-before-abstract order and linearized the multicolumn brand list correctly, with only occasional OCR noise in table-of-contents entries. **Input:** > **Image** **Output:** > **Image** **Input:** > **Image** **Output:** > **Image** **Input:** > **Image** **Output:** > **Image** **Input:** > **Image** **Output:** > **Image** **Bottom line:** Hierarchy and order are usually preserved well, with the main blemish being OCR noise in some TOC entries. ### Caption and Footnote Association **Verdict:** Mixed Keeps captions, notes, and segment explanations adjacent to the elements they describe. The financial report's notes stay under their table, and the scanned paper's figure/table captions stay adjacent, though the hybrid earnings footnote block can be displaced ahead of its section heading. **Input:** > **Image** **Output:** > **Image** **Input:** > **Image** **Output:** > **Image** **Input:** > **Image** **Output:** > **Image** **Bottom line:** Association is usually correct, but the hybrid earnings footnote displacement shows it can misplace notes on complex pages. ### Form Checkbox State Extraction **Verdict:** Strong Recognizes simple SEC filing checkbox states on the cover page and preserves them as x/o markers. On the tested filing, it kept the checked annual-report option and the unchecked transition-report option. **Input:** > **Image** **Output:** > **Image** **Bottom line:** Works for simple filing checkboxes on this test. ## Free and self-hosted | Plan | Price | Notes | | --- | --- | --- | | Open Source ★ | $0 | Apache-2.0-based MinerU Open Source License | ## Is It Right For You? **Use it if** - You need charts, figures, logos, or photos preserved as real embedded images in markdown. - You want open-source PDF-to-markdown with decent heading order and OCR on scanned pages. - You can tolerate slower processing in exchange for local/self-hosted control and image retention. **Skip it if** - You need punctuation-perfect OCR with no mojibake in em-dashes, quotes, or apostrophes. - You need every complex table to preserve row and column boundaries without any structural defects. - You need fast turnaround on long, visually dense PDFs. ## Classification - **Category:** developer-tools - **Subcategory:** pdf-tools - **Type:** text - **Built for:** Other ## Frequently Asked Questions **Q: Does MinerU keep charts and figures instead of dropping them?** Yes. In this research, MinerU retained every chart, figure, photo, logo, and seal as a real embedded image across all three test PDFs. **Q: How good is MinerU at table reconstruction?** Mixed. Some tables were reconstructed well with real HTML rowspan/colspan structure, but others were misaligned, truncated, or collapsed into merged cells, especially the hybrid earnings summary and the sales-by-segment table. **Q: How accurate is the OCR text extraction?** Mostly accurate, but recurring mojibake affects punctuation such as em-dashes, curly quotes, apostrophes, and some isolated characters. The scanned research paper text was otherwise very clean. **Q: Does MinerU preserve reading order and headings?** Generally yes. The scanned paper kept title-before-abstract order, the financial report TOC preserved section hierarchy, and the multicolumn brand list was linearized correctly. The hybrid earnings TOC had mojibake in many item labels, though the structure was still recognizable. **Q: How does MinerU handle captions and footnotes?** Usually well. Table notes and figure captions stayed adjacent to their content in the financial report and scanned paper, but one hybrid-earnings footnote block was displaced ahead of its own section heading. **Q: Is MinerU free to use?** Yes. The report identifies it as open source under an Apache-2.0-based MinerU Open Source License, with a $0 price. ## Similar Tools AI tools similar to MinerU: - [Docling](https://aidemos.com/tools/docling) — Open-source PDF-to-markdown conversion that is strong on text, headings, and standard tables, but drops charts and other visual assets. - [PyMuPDF4LLM](https://aidemos.com/tools/pymupdf4llm) — Open-source PDF-to-markdown for clean native PDFs, but unreliable on scans, dense tables, and images. - [LiteParse](https://aidemos.com/tools/liteparse) — Open-source PDF-to-markdown parsing that works well on native-digital reports, but degrades on scans, tables, and charts. - [doc2mark](https://aidemos.com/tools/doc2mark) — Open-source PDF-to-markdown that preserves native text and headings, but still struggles with tables, charts, images, and scans. - [MarkItDown](https://aidemos.com/tools/markitdown) — Fast native-PDF text extraction for markdown, but structure, charts, images, and scans are unreliable. ## Need a custom AI solution for this use case? If you are looking to build a custom PDF parsing, document conversion, or structured data extraction system for your business or internal workflow, email us at [contact@futuresmart.ai](mailto:contact@futuresmart.ai). ### Found something inaccurate or missing? We try to keep our AI research accurate and useful. If you found outdated information, an issue, or have a suggestion, email us at [collaborate@aidemos.com](mailto:collaborate@aidemos.com).