OCRRouter

PaddleOCR-VL vs OCRmyPDF

Pick PaddleOCR-VL when you need Markdown or JSON from scanned pages, including tables, formulas and charts, and you have a GPU. Pick OCRmyPDF when you need to make scanned PDFs searchable and keep them as PDF/A on CPU-only machines. The main split is output: structured page parsing versus a text layer added to the original PDF.

PaddleOCR-VLpaddle/paddleocr-vlOCRmyPDFocrmypdf/ocrmypdf
TypeOpen sourceOpen source
Price6 credits / page1 credit / page
Per 1,000 pages$6.00$1.00
Typical speed1.3 s / page1.0 s / page
Best forCJK documentsSearchable PDFs from scans
InputsPDF, imagesPDF, images
Languages109 languages100+ languages (via Tesseract)
Max pages per request5001,000
Tables✓ YesNo
HandwritingNoNo
Math✓ YesNo
Bounding boxes✓ YesNo
Confidence✓ YesNo
FormsNoNo
Multilingual✓ Yes✓ Yes
On OCRRouterAvailableAvailable

Prices are what you pay on OCRRouter. Speed is the provider's typical figure, not a benchmark we ran.

Which to choose

Choose PaddleOCR-VL if

  • You need Markdown or JSON output for full pages, not just a searchable PDF.
  • Your documents have tables, formulas or charts, such as academic papers and reports.
  • You scan in many scripts, since one model covers 109 languages including Cyrillic, Arabic, Devanagari and Thai.

Choose OCRmyPDF if

  • You need validated PDF/A output for archiving, with the OCR layer inserted without re-encoding other content where possible.
  • Your servers have no GPU, since it runs on CPU only and spreads work across all cores.
  • You run an office scanner pipeline and only need the scanned PDFs to become searchable.

Key differences

PaddleOCR-VLOCRmyPDF
What it producesMarkdown and JSON for full pages, covering text, tables, formulas and chartsSearchable PDF (PDF/A by default) and plain text, with no layout, table or Markdown output
How you run itSelf-host through PaddleOCR, vLLM or transformers; the transformers path only does element-level recognition, not full page parsingCommand-line tool and library that runs locally
HardwareNVIDIA GPU recommended; CPU inference possible but slow; no minimum VRAM statedCPU only; uses all CPU cores
Licence and priceApache-2.0; free, you pay for computeMPL-2.0; free, you pay for compute
Languages109 languages in one model100+ languages via Tesseract language packs
BenchmarksOmniDocBench v1.6 overall 96.33 (PaddleOCR-VL-1.6, authors); olmOCR-Bench overall 80.0 (independent)Not published
Known failuresVersion 1.6 regresses Russian and Kazakh OCR on born-digital PDFs, ships use_cache false (about 11x slower decoding), can miss table content, and hangs on blurry images--skip-text with default PDF/A output removes form fields, --deskew --clean-final can double or quadruple file size, and PDFs whose forms share one Resources dictionary can hang it

Sources

What each one is

PaddleOCR-VL

Open-source document model from Baidu. Excellent on Chinese, Japanese and Korean.

PaddleOCR-VL details

OCRmyPDF

Command-line tool that adds a searchable text layer to scanned PDFs, using Tesseract.

OCRmyPDF details

Try both on your own file

Run the file once with each model and compare the output.

PDF, PNG, JPG, WebP, HEIC. Up to 20 MB each, 10 files.No file?

Use one, fall back to the other

{"model": "paddle/paddleocr-vl","fallbacks": ["ocrmypdf/ocrmypdf"],"file_url": "https://example.com/document.pdf"}

Questions

More comparisons

PaddleOCR-VL alternativesOCRmyPDF alternativesBest OCR models in 2026