OCRRouter

Surya vs dots.ocr

Pick Surya for local OCR and layout work on a laptop or CPU box, where you need bounding boxes and reading order for 90+ languages. Pick dots.ocr for multilingual PDFs that must become Markdown with HTML tables and LaTeX formulas, on a CUDA GPU. The main split is licence and hardware: Surya has a revenue cap and runs on CPU, dots.ocr is MIT code and is documented for GPU only.

Suryadatalab/suryadots.ocrrednote/dots-ocr
TypeOpen sourceOpen source
Price2 credits / page2 credits / page
Per 1,000 pages$2.00$2.00
Typical speed1.5 s / page1.5 s / page
Best forLayout and line-level OCRMultilingual document parsing
InputsPDF, imagesPDF, images
Languages90+ languages100+ languages
Max pages per request500500
Tables✓ Yes✓ Yes
HandwritingNoNo
MathNo✓ Yes
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 runs on CPU and Apple Silicon through llama.cpp, with peak memory of about 4.3-5.9 GB on an M4 Max.
  • At 689M parameters it scores 83.7% on olmOCR-bench (authors' figure) and reaches 4.52 pages/s at 192 DPI on one H100 with vLLM.
  • One model returns layout labels, reading order, OCR, HTML tables and KaTeX math with per-block bounding boxes.

Choose dots.ocr if

  • The code is MIT-licensed, while Surya's weights are free only for research, personal use and companies under $5M funding or revenue.
  • It outputs Markdown directly, plus JSON, HTML, LaTeX and bounding boxes, which suits multilingual PDFs with mixed tables and text.
  • The dots.mocr-svg variant converts charts and diagrams to SVG, and dots.mocr (v1.5) scores 83.9 on olmOCR-bench.

Key differences

Suryadots.ocr
LicenceCode Apache-2.0; weights under a modified AI Pubs Open Rail-M, with a commercial licence needed above $5M revenue or fundingMIT, with a separate dots.ocr licence agreement file in the repository
Model size689M parameters3B parameters (1.7B LLM backbone)
HardwareNVIDIA GPU via vLLM, or CPU and Apple Silicon via llama.cpp; needs a separately running inference serverSingle CUDA GPU with vLLM is the documented path; a CPU guide is linked; no VRAM figure stated
Output formatsJSON, HTML, LaTeX, bounding boxes, plain textJSON, Markdown, HTML, LaTeX, bounding boxes, SVG
olmOCR-bench83.7% (Surya 2.5, authors' figure)79.1 for dots.ocr; 83.9 for dots.mocr 1.5 (authors' figures)
TablesSurya 2.5 table output lacks row and column geometry; multilingual tables score only 58.1Complex tables remain difficult; a table above 1000 grid cells raises an error and drops the whole page in Docling
Known failureHandwriting, complex tables, math and photos or natural scenes are weak; the README points hard documents to the hosted Datalab APIRuns of ellipses or underscores can make output repeat endlessly, so forms with long dot leaders are a risk

Sources

What each one is

Surya

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

Surya details

dots.ocr

Single vision model from rednote that does layout detection and content recognition together.

dots.ocr 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": ["rednote/dots-ocr"],"file_url": "https://example.com/document.pdf"}

Questions

More comparisons

Surya alternativesdots.ocr alternativesBest OCR models in 2026