Granite Docling vs Florence-2
Pick Granite Docling to convert English PDFs with tables, code and equations into structured output inside a Docling pipeline. Pick Florence-2 for scene text and short labels where you need line boxes, or when the same model must also do captioning or detection. The main split is structure: Granite Docling outputs DocTags, Markdown, HTML and LaTeX, while Florence-2 returns only plain text and boxes.
Granite Doclingibm/granite-docling | 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 | 1.5 s / page | 0.8 s / page |
| Best for | Small-model document conversion | Lightweight OCR with boxes |
| Inputs | PDF, images | images |
| Languages | English, experimental CJK and Arabic | 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 Granite Docling if
- It converts full pages to DocTags, Markdown, HTML or LaTeX with bounding boxes, so tables and equations keep their structure.
- The model card reports table structure TEDS of 0.97 on FinTabNet, equation F1 of 0.968 and code recognition F1 of 0.988.
- At 258M parameters and Apache-2.0, it is a low-cost fit for CPU or small GPU conversion inside Docling.
Choose Florence-2 if
- OCR_WITH_REGION returns quadrilateral boxes for each text line, which suits scene text and short labels.
- The same model also does captioning, detection and grounding, so one model can cover a pipeline that needs more than OCR.
- It is MIT licensed and comes in 0.23B base and 0.77B large sizes, so it can run as lightweight edge OCR in English.
Key differences
| Granite Docling | Florence-2 | |
|---|---|---|
| Output formats | DocTags, Markdown, HTML, LaTeX, bounding boxes | Plain text and bounding boxes only |
| Licence | Apache-2.0 | MIT |
| Size | 258M parameters | 0.23B (base), 0.77B (large) |
| Tables and equations | Table structure TEDS 0.97 on FinTabNet, equation F1 0.968 | No table or LaTeX output |
| Languages | English primary; Japanese, Arabic and Chinese marked experimental | Chinese output mostly wrong because the vocabulary lacks CJK characters |
| Handwriting | Not published | Spelling errors and missing spaces on cursive handwriting and French text in a Hugging Face blog test |
| Known failure | Repetitive loops on page headings and invoices; may print only exclamation marks on GPUs without bfloat16 such as T4 unless run in float32 | Model card notes the continued-pretrained checkpoint used only 0.1B samples and might not be well trained |
Sources
- Granite-Docling model card
- HF discussion #24 repetition
- Florence-2-large model card
- HF blog OCR test
- HF discussion Chinese very bad
What each one is
Granite Docling
Compact IBM vision model that converts pages to DocTags for the Docling pipeline.
Granite Docling 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": "ibm/granite-docling","fallbacks": ["microsoft/florence-2"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Granite Docling alternativesFlorence-2 alternativesBest OCR models in 2026