Nanonets-OCR vs Donut
Pick Nanonets-OCR when you need printed business PDFs turned into Markdown with no training. Pick Donut when you have a narrow, fixed-layout document type such as receipts or forms and can fine-tune a model to output JSON fields. The main split is that Nanonets-OCR works as shipped, while the Donut base checkpoint must be fine-tuned first.
Nanonets-OCRnanonets/nanonets-ocr | Donutnaver/donut | |
|---|---|---|
| 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 | 1.0 s / page |
| Best for | Markdown with semantic tags | Receipt and form fields after fine-tuning |
| Inputs | PDF, images | images |
| Languages | Multilingual | English, Chinese, Japanese, Korean |
| Max pages per request | 500 | 100 |
| Tables | ✓ Yes | No |
| Handwriting | No | No |
| Math | ✓ Yes | No |
| Bounding boxes | No | No |
| Confidence | No | No |
| Forms | ✓ Yes | ✓ Yes |
| 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 want full-page Markdown, HTML, LaTeX or Mermaid output from printed documents.
- Your documents carry signatures, watermarks or checkboxes that you want tagged or converted to unicode symbols.
- You need flowcharts or org charts extracted as Mermaid code.
Choose Donut if
- You can fine-tune on your own data and want structured JSON for receipt or form fields.
- You need a clear licence: Donut is MIT, and it loads in Hugging Face Transformers.
- Your documents have a fixed layout and you want classification or field extraction without a separate OCR step.
Key differences
| Nanonets-OCR | Donut | |
|---|---|---|
| Licence | Not stated on the Nanonets-OCR2-3B model card; base model is Qwen2.5-VL-3B | MIT |
| Ready to use | Fine-tuned from Qwen2.5-VL-3B and used with a fixed prompt | Base checkpoint is pretrain-only and must be fine-tuned before use |
| Output formats | Markdown, HTML, LaTeX, Mermaid | JSON, plain text |
| Benchmarks | olmOCR-bench overall 69.5, DocVQA 89.43, ChartQA 78.56 (OCR2-3B, reported by the authors) | Not published |
| Handwriting | Not trained on handwriting; can hallucinate, and one test saw generation loops with CER up to 2100% | Not published |
| Known failure | Other prompts than the fixed one are expected to cause hallucination | Misses tiny text in large images because of input resolution limits |
| Cost and hosting | Free, self-hosted; you pay for compute | Free, self-hosted; you pay for compute. Transformers v5 dropped image-to-text pipeline support, so load it directly or use v4 |
Sources
- Nanonets-OCR2-3B model card
- Nanonets-OCR-s announcement
- HF discussion fixed prompt
- donut-base model card
- Donut ECCV paper
What each one is
Nanonets-OCR
Open image-to-Markdown model that tags tables, equations, signatures, watermarks and checkboxes.
Nanonets-OCR detailsDonut
OCR-free document understanding model. Fine-tuned for receipts, forms and document Q&A.
Donut detailsTry both on your own file
Run the file once with each model and compare the output.
PDF, PNG, JPG, WebP, HEIC. Up to 20 MB each, 10 files.No file?
Use one, fall back to the other
{"model": "nanonets/nanonets-ocr","fallbacks": ["naver/donut"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Nanonets-OCR vs GOT-OCR 2.0Nanonets-OCR vs NougatHunyuan OCR vs Donutdots.ocr vs DonutSurya vs DonutMarker vs DonutDonut vs RolmOCRUnlimited OCR vs Nanonets-OCR
Nanonets-OCR alternativesDonut alternativesBest OCR models in 2026