OmniDoc Document Extraction

OmniDoc Document Extraction / 2026 comparison

Free document extraction tools compared for 2026: PDF to Markdown and JSON

Turning PDFs, scans and Office files into Markdown and JSON with headings, tables and images can mean self-hosting an open-source model or calling an API billed by the page. This page puts the free allowance, inputs, outputs and setup of seven common tools side by side, with the official source for every number.

This comparison is written by the OmniDoc team, and OmniDoc is our own product. Every number about another tool comes from its official page or repository, checked on 14 November 2026; their pages win if anything differs.

Extract a file free with OmniDoc

At a glance: free allowance, inputs, outputs

ToolFree allowanceInputsOutput / setup
OmniDoc free document extraction oursUp to 501 pages per PDF, no monthly cap; free jobs wait in a queue; free OmniDoc account requiredPDF, scanned PDF, PNG/JPG/WEBP/BMP images, Word, PowerPoint, Excel, CSV, EPUB, HTML, Markdown, TXT, SRT/ASSWeb page, nothing to install; a ZIP with document.md, document.json, images/ and manifest.json
MinerUOpen source, self-hosted; online API: 2,000 pages a day per account at top priority, lower priority after thatPDF, images, DOC/DOCX, PPT/PPTX, XLS/XLSX, RTF, ODT/ODS/ODP, EPUB, OFD, HTML/MHTML, CSV/TSVMarkdown, HTML, LaTeX, DOCX and more; self-hosted, online API, or demos on Hugging Face and ModelScope
DoclingOpen source (code under MIT), runs on your own machine, no allowance to countPDF, DOCX, PPTX, XLSX, HTML, EPUB, images and moreMarkdown, HTML, DocTags, lossless JSON; runs locally, OCR for scans
Marker (Datalab)Code open source (Apache 2.0); hosted API allowance of $20 a month with a work email, $10 with a personal onePDF, images, PPTX, DOCX, XLSX, HTML, EPUBMarkdown, JSON, HTML, chunks; equations as LaTeX, tables as HTML
LlamaParse (LlamaIndex)10,000 free credits a month for new users; basic parsing from 1 credit per pageSee the official siteOnline API (LlamaCloud), account needed
MathpixFree: 10 PDF pages and 10 images a month (2x with an .edu email); Pro $4.99/month for 1,000 pagesPDF, imagesApp and API; strong on formulas
Azure AI Document IntelligenceFree tier F0: 500 pages a month, but each analysis covers only the first 2 pages and files up to 4 MBPDF, JPEG/PNG/BMP/TIFF/HEIF images, DOCX, PPTX, XLSCloud API, Azure account needed; the layout model returns paragraphs, tables and structure

Tool by tool: who it suits and what to watch

OmniDoc free document extraction ours

Headings, paragraphs and lists keep the hierarchy of the original; two or three columns are joined in reading order. Tables come out as Markdown tables and as rows and columns in the JSON; pictures are saved in images/ with their positions marked. Scans go through text recognition (OCR) and are then organised by their layout. Only you can see the files you upload and their results.

Good for: Markdown and JSON without setting anything up or writing code; preparing documents for a knowledge base or RAG.

MinerU

Licensed under the MinerU Open Source License (based on Apache 2.0 with additional conditions). The Precision Extract API needs a token and takes files up to 200 MB and 600 pages; the Agent Lightweight API needs no login and takes files up to 10 MB and 20 pages, rate-limited per IP.

Good for: developers with a GPU who want to self-host, or who batch Chinese documents through an API.

Source: MinerU — GitHub README, MinerU — API documentation (limits)

Docling

Its README lists page layout, reading order and table structure, OCR for scanned PDFs and images, and local execution for sensitive data and air-gapped environments. Model weights carry their own licences, to be checked separately.

Good for: teams that must stay offline or inside their own network and are comfortable with Python.

Source: Docling — GitHub README

Marker (Datalab)

The model weights use a modified OpenRAIL-M licence, free for research, personal use and startups under $5M in funding or revenue; commercial use beyond that is paid. The hosted API is priced per 1,000 pages: $1 fast, $3 balanced, $7.50 accurate; a card is needed only after the free allowance.

Good for: developers who want LaTeX equations and HTML tables and can work with the weights licence.

Source: Marker — GitHub README, Datalab — Pricing

LlamaParse (LlamaIndex)

LlamaIndex’s pricing page lists a $0 pay-as-you-go plan with 10K free credits a month; each parsing tier uses a different number of credits per page, from 1 for basic parsing. It fits neatly into the LlamaIndex RAG stack.

Good for: developers already building RAG with LlamaIndex.

Source: LlamaIndex — Pricing

Mathpix

Mathpix’s pricing lists 10 PDF pages and 10 images a month on the free plan, doubled with an .edu email; Pro is $4.99 a month for 1,000 PDF pages and 5,000 images. Formula recognition is its strength; the free allowance is small.

Good for: the occasional page of formulas or a photo of a maths problem.

Source: Mathpix — Snip pricing

Which one for which job

Your situationTry first
Hundreds of pages of PDF or scans into Markdown and JSON, without codeOmniDoc (up to 501 pages per PDF, in the browser)
Data must stay on your own serversDocling or MinerU, self-hosted
Chinese documents in daily batchesMinerU online API (2,000 top-priority pages a day), or OmniDoc
Equations from papers as LaTeXMarker, or Mathpix (small free allowance)
Already building RAG with LlamaIndexLlamaParse
Already building on AzureAzure Document Intelligence (the free tier reads 2 pages per request)
Word, PowerPoint, Excel, EPUB or HTML to MarkdownOmniDoc, MinerU or Docling

Questions

Is there a free PDF to Markdown converter in 2026?

Yes. MinerU, Docling and Marker are open source and can be self-hosted; the OmniDoc free document extraction runs in the browser with no monthly cap and up to 501 pages per PDF, and its ZIP holds document.md and document.json; MinerU’s online API gives 2,000 top-priority pages a day per account. Figures from the official pages, checked November 2026.

How is structured extraction different from PDF to text?

PDF to text gives you a run of plain text. Structured extraction keeps heading levels, paragraphs, tables as rows and columns, image positions and reading order, as Markdown and JSON that programs and AI tools can use directly.

I cannot self-host. What are my options?

Use an online tool: the OmniDoc free document extraction takes an upload in the browser; MinerU, Datalab, LlamaParse and Azure offer online APIs, which need an account, a token and some code.

Can these tools handle scanned PDFs and tables?

Yes. OmniDoc, Docling and MinerU recognise text in scans and the structure of tables. OmniDoc writes tables as Markdown tables and as rows and columns in the JSON; check very complex tables against the original.

Who wrote this comparison?

The OmniDoc team; OmniDoc is our own product. Every figure about another tool links to its official page or repository, and that page wins if anything has changed.

Sources (checked 14 November 2026)

  1. MinerU — GitHub READMEhttps://github.com/opendatalab/MinerU
  2. MinerU — API documentation (limits)https://mineru.net/doc/docs/index_en/
  3. Docling — GitHub READMEhttps://github.com/docling-project/docling
  4. Marker — GitHub READMEhttps://github.com/datalab-to/marker
  5. Datalab — Pricinghttps://www.datalab.to/pricing
  6. LlamaIndex — Pricinghttps://www.llamaindex.ai/pricing
  7. Mathpix — Snip pricinghttps://mathpix.com/pricing/snip
  8. Microsoft Learn — Document Intelligence quotas and limitshttps://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/service-limits
  9. Azure — Document Intelligence pricinghttps://azure.microsoft.com/en-us/pricing/details/ai-document-intelligence/

2026 guide: 8 things to check in a document extraction tool · OmniDoc Document Extraction · Document translation · Open the tool · Get the desktop app