If downstream code reads a Word table by column name, try Docling Slim on a representative file first. MarkItDown kept our test values but left the Markdown header blank. Both tools preserved the same table and link in HTML. That makes the source format and what the next step needs more useful decision criteria than the fact that both tools produce Markdown.
If your input is mainly PDF, also compare PyMuPDF4LLM's PDF-to-Markdown workflow. Check reading order and tables on your own pages instead of applying these Word and HTML results to PDF.
What we gave both tools
We used the same synthetic shipping brief in DOCX and HTML. Each named two columns, Route and Cutoff, with Taipei at 08:30 and Tainan at 16:45. Both included reference BOX-17. The HTML file used <th> for its headers and contained a hyperlink. The DOCX had column labels in ordinary first-row cells, with no repeating-header setting. Each tool read both files.
| Check | Docling Slim 2.150.1 | MarkItDown 0.1.9 |
|---|---|---|
| DOCX Markdown header | Route and Cutoff |
Blank cells in the header, column names in a body row |
| City and cutoff pairs | Both pairs intact in both formats | Both pairs intact in both formats |
| DOCX heading | ## Shipping brief |
# Shipping brief |
| HTML link | Destination preserved | Destination preserved |
| Programmatic output used | One structured table from the document object | Markdown through the CLI |
Choose against one of your actual files
- Ingestion by column name: compare the first Markdown row or the structured table with the original labels. Docling exposes its parsed table list in
document.tables. - Readable Markdown or text search: check a known identifier and value. If a later step splits documents by heading, check the heading level too.
To repeat the inspection, install markitdown[pdf,docx]==0.1.9 in one Python 3.12 environment and run markitdown shipping.docx. In a second environment, install docling-slim[format-docx,format-html]==2.150.1. Import DocumentConverter from docling.document_converter, convert the same file, and call export_to_markdown() on the returned document.
What this comparison covers
The two tools ran on one simple table in two formats. We did not compare OCR, PDF layout recovery, processing speed, or large batches. Repeat the header check on the formats and styles you actually receive. Docling Slim version and extras · Microsoft MarkItDown 0.1.9.