Documents and knowledge · Edition No. 9 · 3 Sep 2026

datalab-to/marker

The recognised open state-of-the-art PDF-to-Markdown converter — Docling's head-to-head, on a corpus of your users' actual uploads.

← Documents and knowledgeRead the whole edition →

39.6k stars · Apache-2.0 · v2.0.0 (2026-07-20)

The recognised open state-of-the-art PDF-to-Markdown converter — Docling's head-to-head, on a corpus of your users' actual uploads.

Repo detailsthe review · specs · pros & cons · install

What it is

A Python PDF-to-Markdown-and-JSON converter built on layout models by the datalab.to team. Apache-2.0. Ships as CLI, Python library, and a hosted service.

Why it matters

Docling (Ed.3 #9) is IBM's answer; Marker is the community's. Both are alive and neither has one obvious win. This is the same head-to-head Ed.6 asked for on BERTopic vs turftopic vs LightRAG: two engines with overlapping strengths, and the only fair test is on real inputs. Marker tends to handle equations and tables better; Docling handles more input types. Run both on ten real PDFs and keep the winner. Both are Apache/MIT so no licence trap either way.

Stars39.6k
LicenceApache-2.0
Latestv2.0.0 (2026-07-20)checked 7 Sep 2026
Written inPython
NeedsPython 3.10+, PyTorch, and about 1 GB of model weights. A GPU is recommended.
RunsLocally, offline once the models are downloaded
Setup effort
3 / 5
Payoff
3 / 5
Good
  • Legitimate head-to-head option: no single PDF converter is best at everything
  • Apache-2.0 — safe inside a paid Grasppy feature
  • Fast, layout-aware, actively developed by a real team
Watch for
  • Model weights are large (~1 GB) — noticeable RAM footprint on the VPS
  • Better with a GPU, though v2.0 improved CPU support
  • Pick ONE (this or Docling) as the default — running both in production wastes resources
Replaces

Paid PDF-extraction APIs

Works well with

A chunker downstream, and a background job queue to run it out of the request path

How to get it

Python library called from a Procrastinate job (Ed.5 #10).

Get the next edition in your inbox

A dozen repositories, opened and checked. The licence read, the last release dated, and the ones that did not make it named with the reason. It is the half most lists leave out.

No tracking pixels. One click to leave. The archive stays free either way.

We use your address to send the edition and nothing else. Confirm by email, leave in one click. How we handle it.