OCRRouter

dots.ocr vs SmolDocling

Pick dots.ocr for multilingual business and academic PDFs where you need Markdown with layout, HTML tables and LaTeX formulas, and you have a CUDA GPU. Pick SmolDocling only for English experiments on small hardware. Its authors call it a preview and point users to Granite-Docling for support.

dots.ocrrednote/dots-ocrSmolDoclinghuggingface/smoldocling
TypeOpen sourceOpen source
Price2 credits / page1 credit / page
Per 1,000 pages$2.00$1.00
Typical speed1.5 s / page1.0 s / page
Best forMultilingual document parsingVery cheap page conversion
InputsPDF, imagesPDF, images
Languages100+ languagesMainly 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 documents are not English-only, and you need multilingual PDFs turned into Markdown with layout.
  • You need one pass to return layout boxes, categories and text, with tables as HTML and formulas as LaTeX.
  • You want MIT-licensed code and a published olmOCR-bench score (79.1 for dots.ocr, 83.9 for the dots.mocr 1.5 release).

Choose SmolDocling if

  • Your pages are English and you want a 256M model that runs on modest hardware.
  • You want speed: about 0.35 seconds per page on an A100 with vLLM.
  • You want DocTags output for Docling, or Apache-2.0 licensing, for a low-budget experiment.

Key differences

dots.ocrSmolDocling
Size3B parameters (1.7B LLM backbone)256M parameters
LicenceMIT, with a separate dots.ocr license agreement file in the repositoryApache-2.0 per the model summary (metadata tag says CDLA-Permissive-2.0)
How you run itSelf-hosted. Single CUDA GPU with vLLM is the documented path. No VRAM figure stated.Self-hosted. Small enough for modest hardware.
LanguagesMultilingualEnglish only
Output formatsJSON, Markdown, HTML, LaTeX, bounding boxes, SVGDocTags, Markdown, HTML, JSON, LaTeX, bounding boxes
Benchmarks (reported by authors)olmOCR-bench 79.1 (dots.ocr), 83.9 (dots.mocr 1.5)FinTabNet table TEDS structure 0.82, full-page OCR F1 0.80, OCRBench 338
Known failureRuns of ellipses or underscores can make output repeat endlessly. Complex tables and formulas stay difficult. A table above 1000 grid cells raises an error and drops the whole page (Docling issue #5146).Endless token repetition and missing closing tags are typical defects. Output can loop on some images and make generation very slow (Docling issue #1515). Preview release.

Sources

What each one is

dots.ocr

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

dots.ocr details

SmolDocling

Tiny 256M vision model from Hugging Face and IBM that outputs DocTags. Predecessor of Granite Docling.

SmolDocling 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": ["huggingface/smoldocling"],"file_url": "https://example.com/document.pdf"}

Questions

More comparisons

dots.ocr alternativesSmolDocling alternativesBest OCR models in 2026