OCRRouter

Kraken vs Surya

Pick Kraken for historical manuscripts, early prints and right-to-left scripts where you can train on your own material and need ALTO or PageXML output. Pick Surya for local OCR, layout detection and reading order on modern documents in 90+ languages, with no training step. The main split is purpose: Kraken is a trainable toolkit for historical and non-Latin material, while Surya is a ready-made model for layout and text.

Krakenkraken/krakenSuryadatalab/surya
TypeOpen sourceOpen source
Price1 credit / page2 credits / page
Per 1,000 pages$1.00$2.00
Typical speed1.5 s / page1.5 s / page
Best forHistorical documents and manuscriptsLayout and line-level OCR
InputsimagesPDF, images
LanguagesMany scripts, incl. right-to-left90+ languages
Max pages per request500500
TablesNo✓ Yes
Handwriting✓ YesNo
MathNoNo
Bounding boxes✓ Yes✓ Yes
ConfidenceNoNo
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 Kraken if

  • Your material is historical manuscripts or early prints, which Kraken is built for.
  • You work with Arabic, Hebrew or other RTL scripts and need BiDi and top-to-bottom support.
  • You need ALTO, PageXML, abbyyXML or hOCR output for digital humanities tools, and can train layout analysis, reading order and recognition on your own data.

Choose Surya if

  • You want layout labels, reading order, OCR, HTML tables and KaTeX math with per-block bounding boxes from one model.
  • You need local OCR on a laptop: it runs on CPU or Apple Silicon through llama.cpp, with peak RAM of about 4 to 6 GB on an M4 Max.
  • You want speed on a GPU: 4.52 pages/s at 192 DPI on one H100 with vLLM, from a 689M parameter model.

Key differences

KrakenSurya
LicenceApache-2.0Code Apache-2.0; weights under a modified AI Pubs Open Rail-M licence, free for research, personal use and companies under $5M funding or revenue
How you run itPython library installed with pip on Linux or macOS (x64 or ARM, Python 3.10 to 3.13). No Windows support.Lightweight client against a vLLM server (NVIDIA GPU) or llama.cpp (CPU or Apple Silicon). Needs a separately running inference backend.
PriceFree, self-hosted. You pay for compute.Free, self-hosted. You pay for compute. Commercial license needed above $5M revenue or funding.
Output formatsALTO, PageXML, hOCR, plain text, bounding boxesJSON, HTML, LaTeX, bounding boxes, plain text
BenchmarksNot publishedSurya 2.5: 83.7% on olmOCR-bench, 85.9 on OmniParseBench, 74.9% pass rate on 91 languages (author-reported)
Accuracy out of the boxUsually needs training on your own material for good accuracyNo training step. Only 21 of 91 languages reach a 90% pass rate, and multilingual tables score 58.1
Known failureInstall dependency conflict (issue #816). Learned line order is lost in segmentation and text output (issue #820).Surya 2.5 tables lack row and column geometry, and simple table mode errors, so you need the older Surya 2 checkpoint. Not for handwriting or photos of natural scenes.

Sources

What each one is

Kraken

OCR engine for historical documents and manuscripts, with trainable models for unusual scripts.

Kraken 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": "kraken/kraken","fallbacks": ["datalab/surya"],"file_url": "https://example.com/document.pdf"}

Questions

More comparisons

Kraken alternativesSurya alternativesBest OCR models in 2026