Nougat vs Donut
Pick Nougat if you convert English scientific papers with heavy math into Markdown and LaTeX, and your use is non-commercial. Pick Donut if you need JSON field extraction from receipts or forms with a fixed layout, and you can fine-tune it. The main split is task and licence: Nougat's weights are CC-BY-NC and it parses whole pages, while Donut is MIT and must be trained on your own documents first.
Nougatmeta/nougat | Donutnaver/donut | |
|---|---|---|
| Type | Open source | Open source |
| Price | 2 credits / page | 1 credit / page |
| Per 1,000 pages | $2.00 | $1.00 |
| Typical speed | 3.0 s / page | 1.0 s / page |
| Best for | arXiv-style papers | Receipt and form fields after fine-tuning |
| Inputs | PDF, images | images |
| Languages | Mainly English | English, Chinese, Japanese, Korean |
| Max pages per request | 200 | 100 |
| Tables | ✓ Yes | No |
| Handwriting | No | No |
| Math | ✓ Yes | No |
| Bounding boxes | No | No |
| Confidence | No | No |
| Forms | No | ✓ Yes |
| 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 Nougat if
- You need whole scientific pages turned into Mathpix-style Markdown, with LaTeX for equations and tables.
- It was trained on arXiv and PubMed Central papers, so it works on that kind of document without extra training.
- It ships small and base checkpoints with a simple CLI and API, so you can run it right away.
Choose Donut if
- You need a commercial product, since the code and model are MIT licensed.
- You want structured JSON output for receipt or form fields, or document classification, without a separate OCR step.
- Your documents have a narrow, fixed layout, where a practitioner on Hacker News reports about 200 fine-tuning documents can reach human-level accuracy.
Key differences
| Nougat | Donut | |
|---|---|---|
| Licence | Code MIT; weights CC-BY-NC, so no commercial use | MIT |
| Output | Markdown with LaTeX | JSON or plain text |
| Ready to use | Yes, small and base checkpoints with a CLI and API | No, the base checkpoint is pretrain-only and must be fine-tuned |
| Best document types | English arXiv-style papers and math-heavy PDFs | Receipts, forms and other narrow, fixed-layout documents |
| Languages | Works best on English; Chinese, Russian and Japanese do not work | Not published |
| Known failure | Repetition loops on 1.5% of test pages in the paper, more on out-of-domain documents; looping pages can be dropped as MISSING_PAGE | Misses tiny text in large images because of input resolution limits |
| Tooling note | Self-hosted Python package | Transformers v5 dropped image-to-text pipeline support; load the model directly or use v4 |
Sources
- Nougat GitHub README
- Nougat paper
- Issue #11 repetitions
- donut-base model card
- Donut ECCV paper
- Hacker News thread
What each one is
Donut
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": "meta/nougat","fallbacks": ["naver/donut"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Nougat vs GOT-OCR 2.0MonkeyOCR vs DonutDonut vs ABBYY FineReaderAmazon Textract vs DonutNougat vs Typhoon OCRMathpix Convert vs DonutAzure Document Intelligence vs NougatOCRmyPDF vs Donut
Nougat alternativesDonut alternativesBest OCR models in 2026