OCRRouter

DeepSeek-OCR vs dots.ocr

Pick dots.ocr when you need structured output from multilingual PDFs: layout boxes, tables as HTML, formulas as LaTeX, and a published olmOCR-bench score. Pick DeepSeek-OCR for cheap self-hosted bulk text extraction where token count and throughput matter. Both are MIT-licensed and self-hosted, so the main split is structured parsing versus token-efficient throughput.

DeepSeek-OCRdeepseek/deepseek-ocrdots.ocrrednote/dots-ocr
TypeOpen sourceOpen source
Price1 credit / page2 credits / page
Per 1,000 pages$1.00$2.00
Typical speed1.1 s / page1.5 s / page
Best forCheap bulk conversionMultilingual document parsing
InputsPDF, imagesPDF, images
Languages100+ languages100+ languages
Max pages per request200500
Tables✓ Yes✓ Yes
HandwritingNoNo
Math✓ Yes✓ 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 DeepSeek-OCR if

  • It offers resolution modes from 64 to 400 vision tokens per page, so you can trade detail for cost.
  • The authors quote about 2,500 tokens per second of PDF throughput on one A100-40G, which suits bulk jobs.
  • It is MIT-licensed with no extra licence agreement file, so commercial use is simple to check.

Choose dots.ocr if

  • One model returns layout boxes, categories and text, with tables as HTML and formulas as LaTeX.
  • It has authors' olmOCR-bench scores of 79.1, and 83.9 for the dots.mocr (1.5) version, which gives you a number to compare against.
  • The dots.mocr-svg variant converts charts and diagrams to SVG, and the README documents multilingual PDF to Markdown.

Key differences

DeepSeek-OCRdots.ocr
LicenceMITMIT, plus a separate dots.ocr license agreement file in the repository
How you run itSelf-host with Transformers or vLLM on a CUDA GPU (CUDA 11.8, PyTorch 2.6, bfloat16 reference setup)Self-host with vLLM or Hugging Face Transformers; single CUDA GPU with vLLM is the documented path, no VRAM figure stated
PriceFree, you pay for computeFree, you pay for compute
Output formatsMarkdown, plain text, HTML, bounding boxesJSON, Markdown, HTML, LaTeX, bounding boxes, SVG
BenchmarksNo benchmark scores; throughput of about 2,500 tokens per second on an A100-40G is from the authorsolmOCR-bench 79.1 (dots.ocr) and 83.9 (dots.mocr), both from the authors
Known failure: repetitionRepeated tokens and looping over table headers (GitHub issues #12 and #16); a blank image took 30 seconds and produced repeated garbage in a LearnOpenCV testRuns of ellipses or underscores can make output repeat endlessly, per the authors
Known failure: tables and exact textIt corrects nonsense words instead of copying them, so unusual words or codes may not be transcribed exactly (arXiv 2601.03714)Complex tables and formulas remain difficult; a table above 1000 grid cells raises an error and drops the whole page (Docling issue #5146)

Sources

What each one is

DeepSeek-OCR

Open-source vision model that compresses pages into few tokens.

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

Questions

More comparisons

DeepSeek-OCR alternativesdots.ocr alternativesBest OCR models in 2026