MarkPrep

MarkPrep vs Marker

MarkPrep converts PDFs to Markdown in your browser, while Marker is a Python tool that runs deep-learning models on your own computer. Marker's models handle scans in many languages and equations. MarkPrep needs no install or GPU, and it estimates how many tokens the result uses.

Marker comes from Datalab and also reads images, Word, PowerPoint, Excel, HTML and EPUB files. It writes Markdown, JSON, HTML or its own chunks format. Its OCR uses Surya, a vision model that runs on your own computer through a local server.

Side by side

 MarkPrepMarker
Where it runsIn your browserOn your own computer, or Datalab's managed platform
Install and hardwareNone, open a web pagepip install marker-pdf; runs on GPU, CPU or Apple MPS
Files leave your computerNo, conversion runs in your browserNot when you run it yourself; yes with --use_llm or the managed platform
Input formatsPDF, Word .docx, Excel, PowerPoint, CSV, HTML, JSON, XML, EPUB, Jupyter, TXT, images, web pages by URLPDF, images, PowerPoint, Word .docx, Excel, HTML, EPUB
Layout, tables and mathsReads the PDF text layer; flags doubtful tables for you to check; no mathsLayout detection and table recognition; formats equations and inline maths
OCRLocal Tesseract, EnglishSurya vision model, multilingual
Token estimatesFor named Claude, GPT and Gemini models, on screenNot mentioned in its README
Chunking for RAGChunk preview split at headings and sized in tokens; JSONL exportA chunks output format, one flat list of layout blocks
Licence and priceFree converter; Pro £19 onceApache 2.0 code; model weights free under $5M revenue or funding (see Datalab's pricing page)

Where Marker wins

Marker's vision model reads scanned pages in many languages. In its balanced mode, if any of a page's embedded text is bad, Marker runs OCR on the whole page again. It formats tables, forms, equations and inline maths. MarkPrep's OCR reads English only, and it has no maths support.

With the optional --use_llm flag, Marker hands parts of the job to an LLM such as Gemini or Claude. That LLM can join a table split over several pages and extract the values filled into forms. Marker also removes headers and footers and saves the images it finds. It can process a whole folder in one command.

What MarkPrep does differently

MarkPrep needs no Python, model server or GPU. It runs in any modern browser, so you can convert a file on a work laptop where you can't install software.

After conversion, the free AI-Ready Pack gives a token estimate for the model you choose and a context-window check. Its chunk preview splits the Markdown at headings into pieces sized in tokens, ready for JSONL export. Each JSONL line stores the chunk's heading path, plus the page number for PDFs.

Frequently asked questions

How do I convert a PDF to Markdown with Marker?+

Install it with pip install marker-pdf, then run marker_single followed by the path to your PDF. For a whole folder, the marker command takes a folder path. Marker can also write JSON, HTML or its chunks format instead of Markdown.

Does Marker need a GPU?+

No. Marker works on an NVIDIA GPU, a normal CPU or Apple's MPS. Its default mode differs by device: a balanced mode on a GPU and a fast mode on CPU or MPS. With OCR switched off, it reads only the PDF text layer and starts no model server.

Is Marker good for RAG?+

Marker has a chunks output format that flattens each page's top-level blocks into one list, with the HTML of each block. Its README describes this as a way to make chunking for RAG easy. MarkPrep uses the word differently. Its chunks are pieces of the Markdown split at headings and sized in tokens.

What licence does Marker use?+

Marker's code is Apache 2.0, free to use, including commercially. Its model weights use a modified AI Pubs Open Rail-M licence. That licence is free for research, personal use and startups under $5M in funding or revenue. For commercial use beyond that, Datalab points to its pricing page.

Does my file leave my computer with Marker?+

Not when you run Marker yourself. Files go to an outside service only if you add the optional --use_llm flag, which calls an LLM such as Gemini or Claude. The same applies if you use Datalab's managed platform. MarkPrep converts the file in your browser, so it stays on your computer.

Can I use both?+

Yes. Marker suits batches of scanned or complex PDFs on a computer you control. MarkPrep suits a single file on a computer where you can't install Python.

Convert a PDF in your browser, with no GPU or model server, and check its token estimate.