We tested five open-source PDF-to-Markdown libraries on the same three mixed PDFs — a hybrid annual report, a table-heavy financial filing, and a scanned research paper — to compare text preservation, table reconstruction, chart handling, image retention, reading order, and OCR behavior.
The strongest overall open-source option here: it kept text, tables, headings, and reading order very well across all three PDFs, but it still dropped charts and standalone images.
The tool does not turn charts into usable content; in both chart-heavy tests, the visuals were simply removed instead of being translated into values or a faithful description, so this is a core failure.
The evidence-backed checks show the shape of the field; coverage explains the gaps.
Columns, left to right: Bonus: Unique or Anomalous Behaviors · Caption/Figure Association · Chart Reconstruction · Document Hierarchy · Image Retention · Long/Complex Document Robustness · Reading Order Preservation · Table Reconstruction · Text Preservation
Pick the tools you care about, then compare what they returned or how they scored.
It produced a clean result for the financial report overall, with strong text, hierarchy, and most tables, but it still dropped the visual assets and flattened one complex table header.
docling-docling-input2-financialpdf-fy2025-heade-5340c1f4d35a.png
Clean prose, headings, and table-of-contents handling come through well, but several tables flatten or split apart, which keeps it from being a top score.
liteparse-liteparse-input2-financialpdf-summary-se-75cdedba0497.png
It can pull through some narrative sections, but the report’s core tables disappear, the header/logo becomes garbled text, and reading order and overall structure become unreliable.
pymupdf4llm-pymupdf4llm-input2-financialpdf-quarterl-052a903bfa21.png
It gets the plain text and contents order right, but table formatting and heading consistency are shaky.
doc2mark-doc2mark-input2-financialpdf-output-64777dd9b209.md
It is solid for plain text and the table of contents stays in order, but the financial tables and section structure are too uneven to call it reliable for structured extraction.
markitdown-markitdown-input2-financialpdf-output-1b08db40e794.md
Open a tool to inspect every recorded check and finding.
It has a few standout quirks: it can preserve form state and financial parentheses correctly, but it also hoovers up scan artifacts as if they were real text. That mix is notable enough to earn a strong but not perfect score.
Captures scanned-page library stamps and barcode-like artifacts directly into the OCR text flow, adding noise such as `186153` and `CSX` to the title region.
permalink to this finding →
Preserves negative numbers in parentheses in financial tables without treating them as markdown syntax, including percentage changes and negative balance-sheet values.
permalink to this finding →
The report explicitly lists the tool as free/open source, with pricing given as `Free - $0 - Open Source`.
permalink to this finding →Extracts interactive form checkboxes and preserves their checked state, including the SEC Form 10-K mark-one field being rendered as selected in the parsed output.
permalink to this finding →
docling is the overall winner from these scorecards: it has the strongest hierarchy handling and long-document robustness, while also scoring well on text preservation, caption-figure association, and table reconstruction. The trade-off is clear: it is weak at keeping images and charts in place, so it is not the best pick when visual fidelity matters. liteparse is the closest alternative for native-digital prose and structure, with solid text and reading order, but it gives up some robustness and table fidelity. pymupdf4llm is a reasonable choice for clean digital text and tables, but it is less reliable on reading order, hierarchy, and complex or scanned layouts. doc2mark fits native-digital prose and headings, but its table output is especially inconsistent and it is weak on images and scans. markitdown is the simplest route for pulling plain text from native PDFs, but structural parsing is uneven and scanned PDFs need external OCR.
The tools we tested for this use case — each card opens its full tested review.
If you are looking to build a custom PDF to Markdown conversion, document parsing, or OCR extraction system for your business or internal workflow, email us at 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.
Comments (0)