OCRRouter

Docling vs Surya

Pick Docling to ingest born-digital PDFs and Office files into RAG pipelines, because it parses many formats and exports lossless JSON with layout and table structure. Pick Surya for OCR on scans, with layout, reading order and bounding boxes in 90+ languages, running locally on a laptop or GPU. Check the licence first: Surya's weights need a commercial licence above $5M funding or revenue, while Docling is MIT.

Doclingibm/doclingSuryadatalab/surya
TypeOpen sourceOpen source
Price2 credits / page2 credits / page
Per 1,000 pages$2.00$2.00
Typical speed2.2 s / page1.5 s / page
Best forStructured export for RAGLayout and line-level OCR
InputsPDF, images, DOCXPDF, images
LanguagesLatin scripts, CJK90+ languages
Max pages per request500500
Tables✓ Yes✓ Yes
HandwritingNoNo
Math✓ YesNo
Bounding boxes✓ Yes✓ Yes
ConfidenceNoNo
FormsNoNo
MultilingualNo✓ 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 Docling if

  • You need one API for PDF, DOCX, PPTX, XLSX, HTML, EPUB, images, emails and XBRL.
  • You want lossless JSON export that keeps layout, reading order and table structure for downstream tools.
  • You want an MIT-licensed library with LangChain or LlamaIndex integrations and a choice of OCR backends (EasyOCR, Tesseract, RapidOCR) or the 258M GraniteDocling VLM.

Choose Surya if

  • You need OCR with layout labels, reading order and per-block bounding boxes for 90+ languages from one 689M model.
  • You want local OCR on a laptop: it runs on CPU or Apple Silicon via llama.cpp, with peak memory of about 4-6 GB on an M4 Max.
  • You want a published score: 83.7% on olmOCR-bench (Surya 2.5, reported by the authors), and 4.52 pages/s at 192 DPI on one H100 with vLLM.

Key differences

DoclingSurya
LicenceMIT (individual models have their own licences)Code Apache-2.0; weights under a modified AI Pubs Open Rail-M. Free for research, personal use and companies under $5M funding or revenue
How you run itPython library, runs locally on macOS, Linux and Windows. CPU works, GPU speeds up model stagesLightweight client that needs a separately running vLLM or llama.cpp inference server. NVIDIA GPU, CPU or Apple Silicon
Input typesPDF, DOCX, PPTX, XLSX, HTML, EPUB, images, emails and XBRLDocument images and scans. Not for photos or natural scenes
Output formatsMarkdown, HTML, JSON, DocTags, plain textJSON, HTML, LaTeX, bounding boxes, plain text
BenchmarksNot publishedolmOCR-bench 83.7%, OmniParseBench 85.9, 91-language pass rate 74.9% (Surya 2.5, all reported by the authors)
Scan accuracyDepends on the OCR engine you plug in, not Docling itselfOCR is built into the model. The README sends hard documents (handwriting, complex tables, math) to the hosted Datalab API
Known failureVersion 2.143.1 drops words and numbers with full-page Tesseract OCR (issue #5254). Reading order can merge neighbouring columns (issue #5132)Surya 2.5 table output lacks row and column geometry. Multilingual tables score 58.1, and only 21 of 91 languages reach a 90% pass rate

Sources

What each one is

Docling

IBM's open-source document converter with a structured document model.

Docling details

Surya

Open-source OCR toolkit with line detection, layout analysis, reading order and tables.

Surya 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": "ibm/docling","fallbacks": ["datalab/surya"],"file_url": "https://example.com/document.pdf"}

Questions

More comparisons

Docling alternativesSurya alternativesBest OCR models in 2026