SmolDocling vs MonkeyOCR
Pick MonkeyOCR for English and Simplified Chinese PDFs such as academic papers with formulas and tables, as long as you only need it for research or have a commercial licence. Pick SmolDocling only for quick English experiments on small hardware, because it is a preview release that its authors have replaced with Granite-Docling.
SmolDoclinghuggingface/smoldocling | MonkeyOCRmonkeyocr/monkeyocr | |
|---|---|---|
| Type | Open source | Open source |
| Price | 1 credit / page | 2 credits / page |
| Per 1,000 pages | $1.00 | $2.00 |
| Typical speed | 1.0 s / page | 1.5 s / page |
| Best for | Very cheap page conversion | Chinese and English PDFs |
| Inputs | PDF, images | PDF, images |
| Languages | Mainly English | Chinese, English |
| Max pages per request | 500 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | No |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | ✓ Yes | ✓ Yes |
| Confidence | No | No |
| Forms | No | No |
| Multilingual | No | 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 SmolDocling if
- It is a 256M-parameter model, small enough for modest hardware.
- It runs at about 0.35 seconds per page on an A100 with vLLM.
- Its Apache-2.0 licence (the metadata tag says CDLA-Permissive-2.0) is not stated to carry a non-commercial restriction, unlike MonkeyOCR's v1 weights.
Choose MonkeyOCR if
- Pro-3B reports the best overall OmniDocBench edit distance among the models its authors compared, in English and Chinese (0.138 / 0.206).
- It covers Simplified Chinese as well as English, while SmolDocling is English only.
- The 1.2B size is about 36% faster than pro-3B for about 1.6% lower accuracy, and quantized weights run on an 8GB RTX 4060.
Key differences
| SmolDocling | MonkeyOCR | |
|---|---|---|
| Licence | Apache-2.0 per the model summary; the metadata tag says CDLA-Permissive-2.0 | Code Apache-2.0; v1 weights are research and non-commercial only, and commercial use needs a written licence |
| Model size | 256M parameters | 3B and 1.2B |
| Languages | English only | English and Simplified Chinese; Traditional Chinese and multilingual text are not fully supported |
| Speed | About 0.35 seconds per page on an A100 with vLLM | Pro-3B about 0.49 pages/s on an RTX 3090; pro-1.2B about 36% faster |
| Table accuracy | TEDS structure 0.82 on FinTabNet 150dpi | TEDS 81.5 (EN) / 87.5 (ZH) on OmniDocBench, pro-3B |
| Output formats | DocTags, Markdown, HTML, JSON, LaTeX, bounding boxes | Markdown, JSON, LaTeX, HTML, bounding boxes, plus a layout PDF |
| Known failure | Endless token repetition and missing closing tags; Docling issue #1515 reports output loops that make generation very slow | Handwriting and photographed text are not fully supported; Docker on 20/30/40-series GPUs and V100 needs a patched image to avoid an LMDeploy shared-memory error |
Sources
- SmolDocling model card
- SmolDocling paper
- Docling issue #1515
- Granite-Docling comparison table
- MonkeyOCR GitHub README
What each one is
SmolDocling
Tiny 256M vision model from Hugging Face and IBM that outputs DocTags. Predecessor of Granite Docling.
SmolDocling detailsMonkeyOCR
Open-source document parser that splits work into structure, recognition and relation steps.
MonkeyOCR 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": "huggingface/smoldocling","fallbacks": ["monkeyocr/monkeyocr"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
SmolDocling alternativesMonkeyOCR alternativesBest OCR models in 2026