Jina OCR vs Florence-2
Pick Jina OCR for turning PDF pages into Markdown, HTML or LaTeX in research and other non-commercial work. Pick Florence-2 for small, MIT-licensed OCR on English scene text and short labels where you need bounding boxes. The main split is output: Jina OCR returns document structure, while Florence-2 returns only plain text and boxes.
Jina OCRjina/jina-ocr | 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 | Clean Markdown at low cost | Lightweight OCR with boxes |
| Inputs | PDF, images | images |
| Languages | About 100 languages | English |
| Max pages per request | 200 | 500 |
| Tables | ✓ Yes | No |
| Handwriting | No | No |
| Math | ✓ Yes | No |
| 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 converts full pages to Markdown, HTML or LaTeX, while Florence-2 gives no document structure.
- It scores 91.14 on OmniDocBench v1.6 and 83.4 on olmOCR-Bench, 7.4 points above DeepSeek-OCR (scores from the authors).
- You can self-host it or call Jina's hosted API, and it reaches 2.57 pages per second on an A100 with about 570M active parameters.
Choose Florence-2 if
- It is MIT licensed, so commercial use needs no separate license, unlike Jina OCR's CC BY-NC 4.0.
- It is much smaller at 0.23B (base) or 0.77B (large), which suits lightweight edge OCR in English.
- OCR_WITH_REGION returns quadrilateral boxes per text line, and the same model also does captioning, detection and grounding.
Key differences
| Jina OCR | Florence-2 | |
|---|---|---|
| Licence | CC-BY-NC-4.0; commercial use needs a license from Jina/Elastic | MIT |
| How you run it | Self-host or Jina's hosted API | Self-host via Transformers |
| Model size | 3B MoE, about 570M active | 0.23B (base), 0.77B (large) |
| Output formats | Markdown, HTML, LaTeX | Plain text and bounding boxes. No Markdown, tables or LaTeX |
| Published benchmarks | OmniDocBench v1.6 overall 91.14; olmOCR-Bench overall 83.4 (authors) | TextCaps CIDEr zero-shot (large) 72.8; COCO detection mAP zero-shot (large) 37.5 (authors). No OCR document benchmark published |
| Hardware | GPU recommended (tested on A100 and L4); falls back to CPU in Transformers | Not published |
| Known failure | Hosted API returns HTTP 503 on cold start; retry after 30 to 60 seconds. Reported scores use a stricter prompt than the recommended default | Spelling errors and missing spaces on cursive handwriting and French text; Chinese output mostly wrong because the vocabulary lacks CJK characters |
Sources
- jina-ocr-v1 model card
- Jina model page
- Florence-2-large model card
- HF blog OCR test
- HF discussion Chinese very bad
What each one is
Jina OCR
Compact page-to-Markdown model from Jina AI, built on the DeepSeek-OCR encoder.
Jina OCR 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": "jina/jina-ocr","fallbacks": ["microsoft/florence-2"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Jina OCR alternativesFlorence-2 alternativesBest OCR models in 2026