OCRRouter

Docling vs SmolDocling

Pick Docling when you need a production pipeline that ingests PDFs, Office files, HTML and images into structured JSON, with an OCR engine you choose. Pick SmolDocling only to experiment with a tiny 256M document VLM on English pages. It is a preview model that its authors have superseded with Granite-Docling, and Docling itself can run the newer 258M GraniteDocling as a VLM option.

Doclingibm/doclingSmolDoclinghuggingface/smoldocling
TypeOpen sourceOpen source
Price2 credits / page1 credit / page
Per 1,000 pages$2.00$1.00
Typical speed2.2 s / page1.0 s / page
Best forStructured export for RAGVery cheap page conversion
InputsPDF, images, DOCXPDF, images
LanguagesLatin scripts, CJKMainly English
Max pages per request500500
Tables✓ Yes✓ Yes
HandwritingNoNo
Math✓ Yes✓ Yes
Bounding boxes✓ Yes✓ Yes
ConfidenceNoNo
FormsNoNo
MultilingualNoNo
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 Docling if

  • It reads PDF, DOCX, PPTX, XLSX, HTML, EPUB, images, emails and XBRL through one API, so one pipeline covers mixed inputs.
  • Its lossless JSON export keeps layout, reading order and table structure, which suits RAG ingestion and LangChain or LlamaIndex integrations.
  • It is MIT licensed and lets you swap OCR backends (EasyOCR, Tesseract, RapidOCR) or use the 258M GraniteDocling VLM instead of the model SmolDocling is replaced by.

Choose SmolDocling if

  • At 256M parameters it is small enough for modest hardware, and it runs at about 0.35 seconds per page on an A100 with vLLM.
  • One model handles text, tables (OTSL), code, formulas and charts with location tags, so you do not wire up separate OCR and layout stages.
  • It is free to self-host and suits cheap experiments converting English pages to DocTags, Markdown, HTML, JSON or LaTeX.

Key differences

DoclingSmolDocling
What it isPython library from IBM Research that converts documents into a DoclingDocument and plugs in OCR engines and VLMs256M document conversion VLM from Hugging Face and IBM that outputs DocTags for Docling
StatusMaintained library with about 70k GitHub stars and over 1,000 open issues as of November 2026Preview release, superseded by Granite-Docling; authors point users there for updates and support
LicenceMIT (individual models have their own licenses)Apache-2.0 per model summary (metadata tag says CDLA-Permissive-2.0)
Input formatsPDF, DOCX, PPTX, XLSX, HTML, EPUB, images, emails and XBRLPage images converted by the VLM; other formats Not published
Output formatsMarkdown, HTML, JSON, DocTags, plain textDocTags, Markdown, HTML, JSON, LaTeX, bounding boxes
LanguagesNot published; scanned-page accuracy depends on the OCR engine you plug inEnglish only
Known failureVersion 2.143.1 drops words and numbers with full-page Tesseract OCR (issue #5254); reading order can merge neighbouring columns (issue #5132)Output can loop with endless token repetition or missing closing tags, making generation very slow even on sparse pages (issue #1515)

Sources

What each one is

Docling

IBM's open-source document converter with a structured document model.

Docling 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": "ibm/docling","fallbacks": ["huggingface/smoldocling"],"file_url": "https://example.com/document.pdf"}

Questions

More comparisons

Docling alternativesSmolDocling alternativesBest OCR models in 2026