OCRmyPDF vs SmolDocling
Pick OCRmyPDF to make scanned PDF archives searchable and keep them as validated PDF/A. Pick SmolDocling only to experiment with a tiny document model that converts English pages to Markdown, HTML or JSON, and note that it is a preview release superseded by Granite-Docling. The main split is the output: OCRmyPDF adds a text layer to the PDF, while SmolDocling rebuilds page structure.
OCRmyPDFocrmypdf/ocrmypdf | SmolDoclinghuggingface/smoldocling | |
|---|---|---|
| Type | Open source | Open source |
| Price | 1 credit / page | 1 credit / page |
| Per 1,000 pages | $1.00 | $1.00 |
| Typical speed | 1.0 s / page | 1.0 s / page |
| Best for | Searchable PDFs from scans | Very cheap page conversion |
| Inputs | PDF, images | PDF, images |
| Languages | 100+ languages (via Tesseract) | Mainly English |
| Max pages per request | 1,000 | 500 |
| Tables | No | ✓ Yes |
| Handwriting | No | No |
| Math | No | ✓ Yes |
| Bounding boxes | No | ✓ Yes |
| Confidence | No | No |
| Forms | No | No |
| Multilingual | ✓ Yes | No |
| On OCRRouter | Available | Available |
Prices are what you pay on OCRRouter. Speed is the provider's typical figure, not a benchmark we ran.
Which to choose
Choose OCRmyPDF if
- You need searchable PDF/A output for archiving, validated by default.
- You need languages beyond English, since Tesseract covers 100+ languages through language packs.
- You want to run on CPU only, using all cores, with no GPU.
Choose SmolDocling if
- You need structured output such as Markdown, HTML, JSON, LaTeX or bounding boxes, which OCRmyPDF does not produce.
- You want one 256M-parameter model to handle text, tables, code, formulas and charts with location tags.
- You are testing small document models on English pages and want about 0.35 seconds per page on an A100 with vLLM.
Key differences
| OCRmyPDF | SmolDocling | |
|---|---|---|
| What it does | Adds a Tesseract OCR text layer to scanned PDFs | Converts pages to DocTags for Docling using a 256M vision-language model |
| Output formats | Searchable PDF (PDF/A by default) and plain text | DocTags, Markdown, HTML, JSON, LaTeX, bounding boxes |
| Licence | MPL-2.0 | Apache-2.0 per the model summary; the metadata tag says CDLA-Permissive-2.0 |
| Hardware | CPU only, spreads work across all cores | Speed figure given on an A100 with vLLM; small enough for modest hardware |
| Languages | 100+ through Tesseract language packs | English only |
| Benchmarks | Not published | Table TEDS 0.82 on FinTabNet at 150dpi, full-page OCR F1 0.80, OCRBench 338 (reported by the authors) |
| Known failure | Hangs on PDFs whose forms share one Resources dictionary (issue #1763); --skip-text with default PDF/A output removes form fields (issue #1790) | Output can loop on some images and become very slow (Docling issue #1515); endless token repetition and missing closing tags are listed as typical defects |
Sources
- OCRmyPDF GitHub README
- Issue #1763 hang
- Issue #1790 form fields
- SmolDocling model card
- Docling issue #1515
- Granite-Docling comparison table
What each one is
OCRmyPDF
Command-line tool that adds a searchable text layer to scanned PDFs, using Tesseract.
OCRmyPDF detailsSmolDocling
Tiny 256M vision model from Hugging Face and IBM that outputs DocTags. Predecessor of Granite Docling.
SmolDocling detailsTry both on your own file
Run the file once with each model and compare the output.
Use one, fall back to the other
{"model": "ocrmypdf/ocrmypdf","fallbacks": ["huggingface/smoldocling"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
OCRmyPDF alternativesSmolDocling alternativesBest OCR models in 2026