marklipi
Comparison · Last reviewed July 2026

marklipi vs Marker

Marker and marklipi both turn PDFs into Markdown, but they are different kinds of tools. Marker is an open-source Python library that runs deep-learning models locally — layout detection, table recognition, and OCR — which makes it strong on scientific papers, books, and scanned material, at the cost of a real installation (Python environment, model weights, and ideally a GPU). marklipi is a hosted HTTP API that does deterministic text-layer extraction only: one curl, no key, no models, with scans deliberately rejected instead of OCR'd. The right choice mostly follows from whether your documents are digital-native and whether you want to operate inference infrastructure.

Side by side

marklipi Marker
Type Hosted HTTP API Open-source Python library (self-hosted)
Setup None — one curl, no API key Python env + model weights; GPU recommended
Scanned PDFs / OCR No — returns 422 so you can route to OCR Yes — built-in OCR
Method Deterministic text-layer extraction Deep-learning layout models
Same input → same output Yes, always Model-dependent
Where files go Processed in memory, never stored Stays on your machine
Cost model Free tier (10/min, 100/day), no signup Free to run; your compute. Check repo for current license terms
Best for Agents & pipelines on statements, contracts, filings Papers, books, scans; offline/air-gapped work

Choose Marker if…

  • Your corpus includes scans, photos, or image-only PDFs — Marker's OCR handles what marklipi deliberately rejects.
  • Documents can't leave your infrastructure (compliance, air-gapped environments).
  • You're converting scientific papers or books where its layout models shine.
  • You have GPU capacity and a team comfortable operating Python ML dependencies.

Choose marklipi if…

  • Your PDFs are digital-native — bank statements, contracts, invoices, exports — where text-layer extraction is exact and OCR adds risk, not value.
  • You want zero infrastructure: no Python env, no model weights, no GPU bill.
  • You need deterministic output — the same PDF converts identically every run.
  • You're building an agent: keyless access, JSON out, machine-readable docs (llms.txt, OpenAPI).

FAQ

Is marklipi a replacement for Marker?

Only for text-layer PDFs. Marker covers scans via OCR; marklipi deliberately returns a 422 for scans so you can route them to an OCR tool. Many pipelines use exactly that split: marklipi for digital-native documents, an OCR tool for the rest.

Which is more accurate?

On PDFs with a text layer, deterministic extraction reads the exact characters in the file, so there is nothing to misrecognize. On scans, Marker is the only option of the two — marklipi doesn't attempt them.

Which is cheaper?

marklipi's free tier is 100 conversions/day with no signup. Marker is free software, but you pay in setup time and compute (a GPU helps), and its license has commercial terms worth checking for company use.

This page is maintained by marklipi and aims to be factually fair — Marker is a good tool, and its capabilities evolve; verify details on their site. Spot an inaccuracy? Email hello@dovey.com.au.

Judge it on your own PDF.

One curl, no signup — or drag a file into the free converter.

Try the converter Read the docs