The Best PDF to Markdown Converters, Compared

For hard documents such as scans, equations and complex tables, the best PDF to Markdown converter is Marker or MinerU, if you can install Python. For a quick conversion with no install, a browser converter such as MarkPrep is the easier choice for digital PDFs in English.

The ranking below states its criteria, so you can re-rank it for your own case. No speed or accuracy benchmark was run for this piece. Every tool fact comes from the tool's own README, licence or pricing page, checked on 12 October 2026.

The shortlist at a glance

Rank	Tool	Runs where	Hard PDFs (scans, maths, complex tables)	Cost
1	Marker	Your machine (Python), or Datalab's hosted API	Strong: OCR in many languages, equations, tables	Free below $5 million revenue or funding; paid hosted API
2	MinerU	Your machine (Python)	Strong: built for tables, formulas and complex layouts	Free below large-company limits
3	Docling	Your machine (Python)	Good: reading order, table structure, formulas, OCR	Free (MIT)
4	MarkPrep	Your browser tab, no install	Basic: digital PDFs, English OCR only, no maths	Free; Pro £19 once
5	MarkItDown	Your machine (Python)	Basic: text extraction; OCR through a plugin or Azure	Free (MIT)

How these PDF to Markdown tools were ranked

The order uses five criteria, in this order of weight:

1. Output on hard PDFs. Does the tool's documentation say it handles scanned pages, multi-column reading order, equations and tables?
2. Where the file goes. Does conversion happen on your own computer, or on someone's server?
3. Effort to start. Is it a web page, or a Python install with model downloads?
4. Cost and licence. Is it free for your use, and are there limits for companies?
5. Extras for AI work. Does it help you fit the result into a model, for example with chunking or token counts?

If your PDFs are clean, digital and single-column, criterion 1 matters much less. In that case the browser tools move up the list.

The ranked shortlist

1. Marker

Marker converts PDF, images, PPTX, DOCX, XLSX, HTML and EPUB. Its README says it handles tables, forms, equations and inline maths, and it removes running headers and footers. It runs on a graphics card (GPU), on an ordinary processor (CPU), or on a Mac through Apple's Metal graphics (MPS). For higher accuracy, the optional --use_llm flag sends each document through a language model as well. Gemini is the default, and Claude, OpenAI-compatible models and Ollama also work.

The code is Apache 2.0. The model weights are free for research, personal use and companies below $5 million in revenue or funding, and paid above that. Datalab, the company behind Marker, runs a hosted API from $4 per 1,000 pages. Choose Marker when the PDFs are hard and you can run Python.

2. MinerU

MinerU reads PDF, images, Word, PowerPoint, Excel, EPUB, HTML and more. Its README says it is most useful for documents with OCR text, tables, formulas, figures or complex page layouts. It has four parsing tiers, from a fast "Flash" preview up to "Advanced". The default install runs on a CPU, and an NVIDIA GPU gives the best throughput. Files stay on your machine unless you configure remote parsing.

The licence is Apache 2.0 with extra terms. A separate commercial licence is needed above 100 million monthly active users or $20 million monthly revenue. Online services built on it must say they use MinerU.

3. Docling

Docling was started at IBM Research and is now run by the LF AI & Data Foundation. It is MIT-licensed. Its README lists reading order, page layout, formulas and table structure, plus OCR for scanned PDFs. It runs locally, ships its own MCP server, and has built-in chunkers for retrieval (RAG) pipelines. It suits a developer building a document pipeline in Python.

4. MarkPrep

MarkPrep is this site's converter, so read this entry with that in mind. It runs in your browser tab with nothing to install, and the PDF is not uploaded. It rebuilds headings from font size and tables from column alignment. Pages with no text layer go through local Tesseract OCR, which reads English only.

Two-column pages are read straight across each line, so sentences from side-by-side columns get interleaved. Equations and handwriting are not converted. MarkPrep shows an estimated token count for named models and says whether the result fits the model's context window. It also previews how the text splits into chunks, with JSONL export. Fifteen other converters were checked for these features on 12 October 2026. Two of them, markdown.new and Cloudflare's conversion API, show an estimated token count. None of the 15 offers all three: a count for each named model, a check against its context window, and a chunk-by-chunk preview. See MarkPrep's PDF converter for the details.

5. MarkItDown

MarkItDown is Microsoft's MIT-licensed Python utility. It converts many formats, including PDF, Word, PowerPoint, Excel, images, audio and HTML. Its own README warns that it "may not be the best option for high-fidelity document conversions for human consumption." OCR comes through a plugin that sends images to a vision model, or through Azure Document Intelligence. It has its own MCP server. Choose it when you need one Python command for many file types and can accept plain output on complex layouts.

Which is the best free or online PDF to Markdown converter?

All five tools on the list convert for free. Marker's weights and MinerU carry licence terms for larger companies. MarkPrep sells an optional £19 Pro licence for its command-line converter and MCP server.

For the best free PDF to Markdown converter with no install, use a browser converter. MarkPrep, markitdown.online and pdf2md.net all state that conversion runs in the browser without an upload. The browser tools also take Word, Excel and HTML files. So the same pick is the best free markdown converter, and the best online markdown converter, for those formats too. If you searched for the best PDF to Markdown converter online, check where the file goes before you drop it in.

Hosted services are the other online route. LlamaParse charges by credits ($1.25 per 1,000 credits, with 10,000 free credits a month), and Datalab sells Marker-style conversion by the page. Both receive your file. The MarkPrep and LlamaParse comparison covers the hosted route in more detail.

Best docx to markdown converter, and the best LLM for the job

Word files

Word files are easier to convert than PDFs. A .docx file already stores headings, lists and tables as named parts, so a converter reads them directly instead of guessing from the page layout. That makes the best docx to markdown converter, or the best Word to Markdown converter in general, a question of convenience more than accuracy. Pandoc reads .docx but cannot read PDF at all. MarkItDown and MarkPrep both use mammoth, an open-source library that turns Word files into simple HTML. MarkPrep runs it in the browser and saves images as separate files in an images/ folder. MarkPrep's Word page has the details.

Is there a best LLM to convert PDF to Markdown?

No single model is best at it, and the tools above don't need one. Marker, MinerU and Docling first read the page with their own layout models, then build the Markdown. Calling a general model such as Gemini or Claude is an optional extra step, as with Marker's --use_llm. MarkPrep can also send the converted text to Anthropic for an optional clean-up, using your own API key.

The best PDF to Markdown converter for AI work

A model reads whatever the converter gives it, mistakes included. So for AI work, pick a converter that lets you check the output before you send it. Token counts tell you whether the document fits the model. A chunk preview shows where a retrieval system will split it.

How to narrow it down without testing every tool

If you need the best PDF parser for LLMs or for a RAG workflow in 2026, answer three questions in order before you install anything:

1. Are the PDFs scanned, two-column, or full of equations? For PDFs with complex structure, start with Marker or MinerU. Docling is the next choice.
2. May the files leave your computer? If no, rule out hosted APIs. The local Python tools and the browser converters keep files on your machine, unless you switch on an LLM or cloud option.
3. Can you install Python? If no, use a browser converter for digital PDFs, and accept its limits on scans and maths.

Then convert one awkward page from your own documents in your top two picks. MarkPrep flags the problems it finds under the result. A scanned page with OCR switched off is flagged as "appears to be a scan with no text layer." A low-confidence OCR page is flagged with its confidence score, so you know which pages to check.

Other questions about Markdown converters

Is there one best Markdown converter or MD converter for every format?

No. The best PDF to Markdown tool for scanned reports needs a Python install and model downloads. A simple Word file needs far less. Match the tool to the format you convert most.

What is the best document to Markdown converter for a folder of mixed files?

Marker, MinerU, Docling and MarkItDown all take PDF, Word, Excel and HTML files in one install. MarkPrep takes the same formats in the browser, plus EPUB, Jupyter notebooks and images, and accepts several files at once.

Is a vendor's list worth more than a "best pdf to markdown reddit" thread?

Only if you can check it. Threads such as "Which PDF to MD converter to use?" and "best .md conversion tools for LLM parsing?" collect people's experience with their own files. This list gives the criteria and the documented features instead, and places its own tool fourth. Use both, then test on your own PDFs.

To try the browser route on your own file, open MarkPrep's PDF converter. To see where the same file will split for retrieval, drop it into the chunk preview.