Jina OCR vs SmolDocling
Pick SmolDocling if you need a free, Apache-2.0 model for English pages on modest hardware, and only for experiments, since it is a preview release. Pick Jina OCR for non-commercial PDF-to-Markdown work where accuracy matters, but note that CC BY-NC 4.0 means commercial use needs a separate license. The main split is licence and maturity: Jina OCR is the stronger, hosted-or-self-hosted model with a non-commercial licence, while SmolDocling is small and permissive but a preview.
Jina OCRjina/jina-ocr | SmolDoclinghuggingface/smoldocling | |
|---|---|---|
| Type | Open source | Open source |
| Price | 1 credit / page | 1 credit / page |
| Per 1,000 pages | $1.00 | $1.00 |
| Typical speed | 1.5 s / page | 1.0 s / page |
| Best for | Clean Markdown at low cost | Very cheap page conversion |
| Inputs | PDF, images | PDF, images |
| Languages | About 100 languages | Mainly English |
| Max pages per request | 200 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | No |
| Math | ✓ Yes | ✓ 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 Jina OCR if
- It scores 91.14 on OmniDocBench v1.6 and 83.4 on olmOCR-Bench, 7.4 points above DeepSeek-OCR (scores reported by the authors).
- You can call Jina's hosted API instead of running a GPU yourself, or self-host the open weights.
- It has about 570M active parameters and runs at 2.57 pages per second on an A100, with Markdown, HTML and LaTeX output.
Choose SmolDocling if
- Its licence is Apache-2.0 per the model summary, so it is not restricted to non-commercial use like Jina OCR's CC BY-NC 4.0.
- At 256M parameters it is small enough for modest hardware, and it takes about 0.35 seconds per page on an A100 with vLLM.
- One model handles text, tables, code, formulas and charts, and outputs DocTags, Markdown, HTML, JSON, LaTeX and bounding boxes.
Key differences
| Jina OCR | SmolDocling | |
|---|---|---|
| Licence | CC-BY-NC-4.0; commercial use needs a license from Jina/Elastic | Apache-2.0 per model summary (metadata tag says CDLA-Permissive-2.0) |
| How you run it | Self-host or Jina's hosted API | Self-host only |
| Size | 3B MoE, about 570M active | 256M |
| Output formats | Markdown, HTML, LaTeX | DocTags, Markdown, HTML, JSON, LaTeX, bounding boxes |
| Speed (A100) | 2.57 pages per second | About 0.35 seconds per page with vLLM |
| Languages | Not published | English only |
| Known failure | Hosted API returns HTTP 503 on cold start; retry after 30 to 60 seconds | Endless token repetition and missing closing tags; output can loop on some images (Docling issue #1515) |
Sources
What each one is
Jina OCR
Compact page-to-Markdown model from Jina AI, built on the DeepSeek-OCR encoder.
Jina OCR 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": "jina/jina-ocr","fallbacks": ["huggingface/smoldocling"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Jina OCR alternativesSmolDocling alternativesBest OCR models in 2026