Tested 3 November 2026
Best PDF to Markdown converter? Seven tools put to one test
Looking for the best PDF to Markdown converter? We ran seven free online tools on the same nine public test pages and scored them with olmOCR-bench's own checker. markdownwise scored 97.2%, ahead of the four others that finished; the closest of them scored 57.0%. Marker, an open-source program you install yourself, scored 96.4%.
We make markdownwise, so weigh this with that in mind. The scores come from the test's own open-source checker, run the same way for every tool, and the method is below so you can repeat it.
The short version
- Running heads and page numbers: markdownwise removed them on 8 of 9 checks and Marker on all 9. CloudConvert and PDF24 removed none, and pdfmarkdown.app 2.
- Tables: markdownwise passed all 21 table checks, Marker 18, PDF24 12, CloudConvert 6 and pdfmarkdown.app 5.
- Columns: markdownwise kept the reading order on all 11 column checks, CloudConvert on 10, pdfmarkdown.app on 9 and PDF24 on 5.
- We fixed what the test found: our first run on 3 November scored 72.8%. We fixed what it showed up, partly with these same pages, then ran markdownwise again on 4 November. The other tools were run once, so they didn't get that second chance. On 150 more olmOCR-bench pages that we ran through both, markdownwise scored 67.1% and Marker 60.7%. Still, treat this as a test, not a verdict.
- Free means different things: iLovePDF converted one file and then asked for a Premium sign-up, and AnythingMD wanted an account before converting anything. markdownwise asks for neither.
- The other direction too: on our Markdown to PDF feature test, markdownwise passed all 20 checks and the closest other tool 14. One tool's download button saved pictures of the pages, with no text you could select.
How the PDFs were scored
The nine pages come from olmOCR-bench, a public test set made by the Allen Institute for AI for exactly this job: three pages with running heads and footers, three set in columns, and three with tables. Each page comes with checks written by people, such as "this sentence comes before that one", "this footer is gone" or "this number sits right of that label". A tool passes a check or it doesn't.
Overall is the test's own average across its groups, which also includes a basic check that each page produced text at all. Nine pages is a small sample, so read the scores as a direction rather than a final ranking.
PDF to Markdown results
| Converter | Overall | Heads and footers removed (9) | Column order (11) | Tables (21) | Notes |
|---|---|---|---|---|---|
| markdownwise | 97.2% | 8 | 11 | 21 | Re-run on 4 November after fixing what this test showed up. On 3 November: 72.8%, with 6 clutter, 9 column and 9 table checks. |
| Marker (open source) | 96.4% | 9 | 11 | 18 | Not a website: a program you install, with heavy AI models. Added for reference. Version 1.6.2 on a laptop CPU, about 13 seconds a page. |
| pdfmarkdown.app | 57.0% | 2 | 9 | 5 | Runs in your browser, no account. |
| CloudConvert | 52.1% | 0 | 10 | 6 | Didn't convert one of the nine files, which counts as failed. |
| PDF24 | 50.6% | 0 | 5 | 12 | |
| iLovePDF | One free conversion, then it asks you to sign up for Premium. On that one file it passed 0 of 5 checks; markdownwise passed 4. | ||||
| AnythingMD | Asks you to create an account before it converts anything. | ||||
| Vertopal | Its upload box didn't accept our test upload, so it wasn't tested. | ||||
Markdown to PDF: one test file, twenty checks
We also wrote one Markdown file that uses the features people reach for, converted it with seven online tools, and checked each PDF by eye for: bold, italic, strikethrough, code and links, a bare web address made a link, curly quotes and dashes, heading levels, nested bullet lists, numbered lists, task-list check boxes, table column alignment, code colours, long code lines wrapped, not cut off, block quotes, footnotes, maths, definition lists, horizontal rules, accents, symbols and Japanese, raw HTML never run, a long table's header repeated on the next page, front matter hidden, page numbers.
| Converter | Passed (of 20) | What it missed |
|---|---|---|
| markdownwise | 20 | None. We wrote this test, and fixed the six things our own converter failed (task lists, footnotes, maths, definition lists, bare addresses, front matter) before testing the others. |
| md2pdf.my | 14 | Table alignment, maths, definition lists and curly quotes; front matter printed as a heading. Page numbers are off unless you tick them in its export options. |
| CodeShack | 10 | Its Download PDF button saves pictures of the pages, so that PDF's text can't be selected or searched (its print option, through your browser, keeps the text); long code lines cut off; no table alignment. |
| Syncfusion | 9 | Numbered lists printed as bullets, task boxes garbled, no footnotes, maths or page numbers; front matter printed. |
| code-format.com | 4 | Prints most Markdown marks as they are (task boxes, footnotes, maths), loses Japanese, and shows a random code as the title. It refused the file until we removed the line that tests raw HTML. |
| online-convert.com | 2 | Prints the Markdown source as plain text, # and ** marks included. |
| LightPDF | Asks you to log in before you can download the PDF. | |
| linangdata | Makes no file: it opens your browser's print window. | |
How the test was run
Every file went through each tool on 3 November 2026 with no account, on its free tier and default settings, the way a visitor would use it. Results that came as a download were scored exactly as downloaded. We'll run the test again in a few months and change the date at the top.
Want to see why the results differ? Read how a conversion works, then try the PDF to Markdown converter or the Markdown to PDF converter on your own files. To read a .md file without converting it, open it in the Markdown viewer. Choosing between two open-source programs instead? See Marker and Docling scored on 200 PDFs.
Questions about this test
Which free tool turns a PDF into the cleanest Markdown?
Of the free websites in this test, markdownwise passed the most checks (97.2%), after we fixed what the test showed up. Marker, a program you install, scored 96.4% and handled page clutter better. Results vary by file, so try two tools on one of your own PDFs.
Why do converters leave page numbers in the text?
A PDF stores each page's header, footer and number as ordinary text. A converter has to notice that it repeats or sits at the edge of the page and leave it out; the tools that don't print it between your paragraphs.
Can I check these results myself?
Yes. The pages are from olmOCR-bench, a public test set from the Allen Institute for AI, and its checker is open source. Run any converter on the same pages and score the Markdown with it.
Do any of these converters need an account?
markdownwise, PDF24, CloudConvert and pdfmarkdown.app worked without one. iLovePDF allowed one free file before asking for a Premium sign-up, and AnythingMD asked for an account first.