OCRRouter

dots.ocr vs MonkeyOCR

Pick dots.ocr for multilingual PDFs, commercial use and anything that needs charts converted to SVG, since its code is MIT-licensed and it covers more languages. Pick MonkeyOCR for English and Simplified Chinese academic papers, or for research on a small GPU. Its v1 weights are non-commercial, so a SaaS product needs a written license first.

dots.ocrrednote/dots-ocrMonkeyOCRmonkeyocr/monkeyocr
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 forMultilingual document parsingChinese and English PDFs
InputsPDF, imagesPDF, images
Languages100+ languagesChinese, English
Max pages per request500500
Tables✓ Yes✓ Yes
HandwritingNoNo
Math✓ Yes✓ Yes
Bounding boxes✓ Yes✓ Yes
ConfidenceNoNo
FormsNoNo
Multilingual✓ YesNo
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 dots.ocr if

  • Your PDFs are multilingual, because MonkeyOCR does not fully support Traditional Chinese or other languages.
  • You ship a commercial product, since the code is MIT-licensed while MonkeyOCR v1 weights need a written license for commercial use.
  • You want charts and diagrams turned into SVG, which the dots.mocr-svg variant does.

Choose MonkeyOCR if

  • Your documents are English or Simplified Chinese academic papers with formulas and tables, where Pro-3B reports the best overall OmniDocBench edit distance among the models the authors compared.
  • You have a small GPU, as quantized weights run on an 8GB RTX 4060.
  • You want a faster option, since Pro-1.2B is about 36% faster than Pro-3B for about 1.6% lower accuracy.

Key differences

dots.ocrMonkeyOCR
LicenceMIT (the repository also includes a separate dots.ocr license agreement file)Code Apache-2.0; v1 weights research and non-commercial only, commercial use needs a written license
Model size3B (1.7B LLM backbone)3B and 1.2B
HardwareSingle CUDA GPU with vLLM is the documented path; no VRAM figure statedTested on RTX 3090/4090, A6000, H800, A100; RTX 4060 8GB works with quantized models
How you run itSelf-hosted via vLLM or Hugging Face TransformersSelf-hosted via LMDeploy or Docker; 20/30/40-series and V100 need a patched Docker image to avoid a shared-memory error
Output formatsJSON, Markdown, HTML, LaTeX, bounding boxes, SVGMarkdown, JSON, LaTeX, HTML, bounding boxes
Benchmark (authors' figures)olmOCR-bench 79.1 for dots.ocr, 83.9 for dots.mocr (v1.5)OmniDocBench overall edit distance 0.138 EN / 0.206 ZH and table TEDS 81.5 / 87.5 (pro-3B)
Known failureRuns of ellipses or underscores can make output repeat endlessly; a table above 1000 grid cells raised an error and dropped the whole page in Docling issue #5146Photographed text, handwriting, Traditional Chinese and multilingual text are not fully supported; Pro-3B throughput is about 0.49 pages/s on a 3090

Sources

What each one is

dots.ocr

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

dots.ocr details

MonkeyOCR

Open-source document parser that splits work into structure, recognition and relation steps.

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

Questions

More comparisons

dots.ocr alternativesMonkeyOCR alternativesBest OCR models in 2026