Nanonets-OCR vs Florence-2
Pick Nanonets-OCR for printed business documents that you want as Markdown, HTML or LaTeX, especially with signatures, checkboxes or flowcharts. Pick Florence-2 for English scene text and short labels where you need bounding boxes, or when you also need detection in one small MIT-licensed model. Neither suits handwriting.
Nanonets-OCRnanonets/nanonets-ocr | Florence-2microsoft/florence-2 | |
|---|---|---|
| Type | Open source | Open source |
| Price | 2 credits / page | 1 credit / page |
| Per 1,000 pages | $2.00 | $1.00 |
| Typical speed | 1.8 s / page | 0.8 s / page |
| Best for | Markdown with semantic tags | Lightweight OCR with boxes |
| Inputs | PDF, images | images |
| Languages | Multilingual | English |
| Max pages per request | 500 | 500 |
| Tables | ✓ Yes | No |
| Handwriting | No | No |
| Math | ✓ Yes | No |
| Bounding boxes | No | ✓ Yes |
| Confidence | No | No |
| Forms | ✓ Yes | 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 Nanonets-OCR if
- You need structured output: Markdown, HTML, LaTeX or Mermaid, not just raw text.
- Your documents have signatures, watermarks, checkboxes or flowcharts, which it tags, converts to unicode symbols or exports as Mermaid.
- You want printed PDFs turned into Markdown for LLMs, with a published DocVQA score of 89.43 for the OCR2-3B model.
Choose Florence-2 if
- You need a very small model: 0.23B (base) or 0.77B (large) parameters, for lightweight English OCR at the edge.
- You need per-line quadrilateral boxes from OCR_WITH_REGION, or detection and grounding in the same model.
- You want a clear MIT licence, since the Nanonets-OCR2-3B model card states no licence.
Key differences
| Nanonets-OCR | Florence-2 | |
|---|---|---|
| Output formats | Markdown, HTML, LaTeX, Mermaid | Plain text and bounding boxes. No Markdown, tables or LaTeX. |
| Model size | 3B (Hugging Face lists about 4B params), fine-tuned from Qwen2.5-VL-3B | 0.23B (base), 0.77B (large) |
| Licence | Not stated on the Nanonets-OCR2-3B model card (base model: Qwen2.5-VL-3B) | MIT |
| Price and deployment | Free, self-hosted from Hugging Face. You pay for compute. | Free, self-hosted via Transformers. You pay for compute. |
| Benchmarks | olmOCR-bench overall 69.5, DocVQA 89.43, ChartQA 78.56 (OCR2-3B, reported by the authors) | TextCaps CIDEr 72.8 and COCO detection mAP 37.5, zero-shot, large (reported by the authors). No document OCR benchmark published. |
| Handwriting | Not trained on handwriting and can hallucinate. One handwriting test saw generation loops with CER up to 2100% and 69.1s per image on an L4. | A Hugging Face blog test found spelling errors and missing spaces on cursive handwriting. |
| Known failure | Trained on a fixed prompt. The team says hallucination with other prompts is expected. | Chinese output is mostly wrong because the vocabulary lacks CJK characters. The model card notes the continued-pretrained checkpoint used only 0.1B samples. |
Sources
- Nanonets-OCR2-3B model card
- Nanonets-OCR-s announcement
- HF discussion fixed prompt
- Florence-2-large model card
- HF blog OCR test
- HF discussion Chinese very bad
What each one is
Nanonets-OCR
Open image-to-Markdown model that tags tables, equations, signatures, watermarks and checkboxes.
Nanonets-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": "nanonets/nanonets-ocr","fallbacks": ["microsoft/florence-2"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Nanonets-OCR alternativesFlorence-2 alternativesBest OCR models in 2026