OCRRouter

Surya vs Unstructured

Pick Surya if you want a self-hosted model for layout, reading order and bounding boxes across 90+ languages, and you can run it on a laptop or a GPU. Pick Unstructured if you need to ingest mixed files (PDF, HTML, Word, PowerPoint, email, images) into a RAG pipeline, with a hosted API as an option. Neither is good for tables: Surya 2.5 table output has no row/column geometry, and the open-source Unstructured library scores 0.426 on table cell content.

Suryadatalab/suryaUnstructuredunstructured/unstructured
TypeOpen sourceOCR API
Price2 credits / page20 credits / page
Per 1,000 pages$2.00$20.00
Typical speed1.5 s / page3.0 s / page
Best forLayout and line-level OCRMixed file types into RAG
InputsPDF, imagesPDF, images, DOCX
Languages90+ languagesMultilingual
Max pages per request5001,000
Tables✓ Yes✓ Yes
HandwritingNoNo
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 Surya if

  • It returns layout labels, reading order, OCR, HTML tables and KaTeX math with per-block bounding boxes from one 689M-parameter model.
  • It scores 83.7% on olmOCR-bench (author-reported) and runs at 4.52 pages/s at 192 DPI on one H100 with vLLM.
  • It runs on CPU and Apple Silicon via llama.cpp, with peak RAM of about 4-6 GB on an M4 Max, so you can keep documents on your own machine.

Choose Unstructured if

  • It handles PDFs, HTML, Word, PowerPoint, emails and images in one library, so one pipeline covers mixed file types.
  • The hosted platform gives you 10,000 free pages, then $0.015 per page, with no servers to run.
  • The Auto, Fast, High-Res and VLM strategies let you use the fast strategy for born-digital PDFs and pay for quality only where you need it.

Key differences

SuryaUnstructured
LicenceCode Apache-2.0; weights under a modified AI Pubs Open Rail-M. Free for research, personal use and companies under $5M funding/revenue.Library Apache-2.0; hosted platform is proprietary.
How you run itSelf-host only. Needs a separate vLLM or llama.cpp inference server (NVIDIA GPU, CPU or Apple Silicon).Self-host the Python library or use the hosted API.
PriceFree software, you pay for compute. Commercial licence needed above $5M revenue or funding.Library free. Hosted: first 10,000 pages free, then $0.015 per page. Business plan is custom.
Input scopeDocument OCR. Not for photos or natural scenes.PDFs, HTML, Word, PowerPoint, emails and images.
TablesSurya 2.5 table output lacks row/column geometry. Multilingual tables score 58.1.Open-source table cell content scores 0.426 versus 0.866 hosted (vendor's own benchmark). hi_res tables can drop first letters at the edges.
Benchmark83.7% on olmOCR-bench; 74.9% pass rate on a 91-language test (author-reported).Not published
Languages90+ languages. Only 21 of 91 reach a 90% pass rate.Table OCR ignores language settings (issue #1694).

Sources

What each one is

Surya

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

Surya details

Unstructured

Document ETL for RAG. Partitions many file types into typed elements; open-source library too.

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

Questions

More comparisons

Surya alternativesUnstructured alternativesBest OCR models in 2026