MarkPrep vs Marker
Marker is a GPU deep-learning pipeline for converting PDFs to Markdown, and it is very good at it. On complex layouts, dense tables, and scanned pages, Marker delivers accuracy that a text-layer tool cannot match. We will not pretend otherwise: if raw extraction quality on hard documents is your single priority, Marker is the stronger converter.
MarkPrep competes on a different axis. It runs entirely in your browser with no GPU and no install, keeps every file on your machine, is free for unlimited use, and ships the AI-ready tooling that Marker was never built to include. The two tools solve overlapping problems in very different ways.
Side by side
| MarkPrep | Marker | |
|---|---|---|
| Where it runs | Browser (WebAssembly), local | GPU machine or hosted API |
| Install / hardware | None — open a web page | GPU required, or pay for API |
| Accuracy on hard scans & tables | Text-layer tier | Best-in-class |
| Files leave your device | Never | Yes if using hosted API |
| Token counting | Live, per-model | None |
| Chunking for RAG | Heading-aware preview + JSONL | None |
| OCR | Local Tesseract, offline | Deep-learning OCR |
| License / price | Free unlimited; Pro $4/mo | GPL + model-weight terms |
When to use which
Use Marker when your inputs are genuinely hard — scanned documents, multi-column academic PDFs, or pages packed with intricate tables — and you have a GPU or budget for the hosted API. Its deep-learning models are honestly better than MarkPrep on those inputs, and accuracy there is worth the setup.
Use MarkPrep when you want to convert a digital document right now without installing anything, when the file is sensitive and cannot be uploaded, or when the whole point is to feed an LLM. The live token count, context-window verdict, heading-aware chunk preview, and JSONL export with heading-path and page metadata turn a Markdown dump into something you can drop straight into a RAG pipeline — all for free, in the tab you already have open.
Frequently asked questions
Which is more accurate on hard PDFs?+
Marker is. It runs a GPU deep-learning pipeline that is genuinely strong on complex layouts, dense tables, and scanned pages. If your input is a stack of difficult scans and raw extraction accuracy is the only thing that matters, Marker will usually beat MarkPrep. MarkPrep targets parity with the text-layer quality tier, not with GPU tools.
Then why would I use MarkPrep instead?+
Because Marker needs a GPU or their hosted API, and MarkPrep needs neither. MarkPrep runs entirely in your browser via WebAssembly, so there is nothing to install, nothing to provision, and your files never leave the tab. It is also free for unlimited use and gives you token counts and chunking that Marker does not.
Does Marker count tokens or chunk for RAG?+
No. Marker focuses on turning documents into Markdown and structured output. It has no live token counting, no context-window verdict, and no heading-aware chunking. MarkPrep adds all three, plus JSONL export with heading-path and page metadata, so the output is ready to drop into a retrieval pipeline.
What about licensing?+
Marker carries GPL plus model-weight terms, which matters if you plan to embed it in a commercial product. MarkPrep runs client-side and is free to use unlimited, with an optional Pro plan at $4/mo or $29/yr for the local MCP server, CLI, watch folders, and BYOK OCR.
Can I use both?+
Yes, and for some workflows that is the right call. Reach for Marker when you have a batch of hard scans and access to a GPU, and use MarkPrep for everyday digital documents where you want zero install, privacy, and LLM-ready output in seconds.
Try it in your browser
No install, no GPU, no upload. Convert a file and see the token count in seconds.