SmolDocling vs Florence-2
Pick SmolDocling for English page conversion into structured output such as DocTags, Markdown, HTML or JSON. Pick Florence-2 for scene text and short labels where you need text with line boxes, or when the same model must also do captioning or detection. SmolDocling gives document structure, but it is a preview release and can loop on some images. Florence-2 gives only text and boxes.
SmolDoclinghuggingface/smoldocling | Florence-2microsoft/florence-2 | |
|---|---|---|
| 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 | 0.8 s / page |
| Best for | Very cheap page conversion | Lightweight OCR with boxes |
| Inputs | PDF, images | images |
| Languages | Mainly English | English |
| Max pages per request | 500 | 500 |
| Tables | ✓ Yes | No |
| Handwriting | No | No |
| Math | ✓ Yes | No |
| 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 converts whole pages and outputs DocTags, Markdown, HTML, JSON, LaTeX and bounding boxes, while Florence-2 returns only text and boxes.
- One 256M model handles text, tables (OTSL), code, formulas and charts with location tags.
- It runs at about 0.35 seconds per page on an A100 with vLLM, and the authors report a table structure TEDS of 0.82 on FinTabNet.
Choose Florence-2 if
- It is MIT licensed, and the SmolDocling licence is stated inconsistently (Apache-2.0 in the summary, CDLA-Permissive-2.0 in the metadata tag).
- OCR_WITH_REGION returns quadrilateral boxes per text line, and the same model also does captioning, detection and grounding.
- It comes in 0.23B base and 0.77B large sizes, so you can pick the smaller one for lightweight edge OCR in English.
Key differences
| SmolDocling | Florence-2 | |
|---|---|---|
| Licence | Apache-2.0 per the model summary; the metadata tag says CDLA-Permissive-2.0 | MIT |
| How you run it | Self-hosted, free, you pay for compute | Self-hosted via Transformers, free, you pay for compute |
| Model size | 256M parameters | 0.23B (base), 0.77B (large) |
| Output formats | DocTags, Markdown, HTML, JSON, LaTeX, bounding boxes | Plain text and bounding boxes only |
| Tables and document structure | Tables via OTSL; TEDS 0.82 on FinTabNet 150dpi | No tables, Markdown or LaTeX output |
| Languages | English only | Chinese output mostly wrong because the vocabulary lacks CJK characters; spelling errors and missing spaces on French text |
| Known failure | Endless token repetition and missing closing tags; output loops on some images (Docling issue #1515); preview release superseded by Granite-Docling | Spelling errors and missing spaces on cursive handwriting; the model card notes the continued-pretrained checkpoint might not be well trained |
Sources
- SmolDocling model card
- Docling issue #1515
- Granite-Docling comparison table
- Florence-2-large model card
- HF blog OCR test
- HF discussion Chinese very bad
What each one is
SmolDocling
Tiny 256M vision model from Hugging Face and IBM that outputs DocTags. Predecessor of Granite Docling.
SmolDocling detailsFlorence-2
Small Microsoft vision model (0.23B and 0.77B) with OCR and OCR-with-regions tasks that return text with boxes.
Florence-2 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": ["microsoft/florence-2"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
SmolDocling alternativesFlorence-2 alternativesBest OCR models in 2026