Mistral OCR vs Donut
Pick Mistral OCR when you need full-page PDF to Markdown for scans, forms and tables with no training, and you can send pages to a hosted API. Pick Donut when you have a narrow, fixed-layout document type such as receipts, can fine-tune a model, and must self-host under an MIT licence. The main split is a ready-made paid API versus a base model you must train yourself.
Mistral OCRmistral/mistral-ocr | Donutnaver/donut | |
|---|---|---|
| Type | OCR API | Open source |
| Price | 25 credits / page | 1 credit / page |
| Per 1,000 pages | $25.00 | $1.00 |
| Typical speed | 0.9 s / page | 1.0 s / page |
| Best for | High-volume PDF to Markdown | Receipt and form fields after fine-tuning |
| Inputs | PDF, images | images |
| Languages | 100+ languages | English, Chinese, Japanese, Korean |
| Max pages per request | 1,000 | 100 |
| Tables | ✓ Yes | No |
| Handwriting | No | No |
| Math | ✓ Yes | No |
| Bounding boxes | ✓ Yes | No |
| Confidence | No | No |
| Forms | No | ✓ 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 Mistral OCR if
- It works out of the box through the Mistral API, while Donut's base checkpoint is pretrain-only and must be fine-tuned before use.
- It returns Markdown, HTML, JSON and images, and OCR 3 rebuilds tables as HTML with rowspan and colspan.
- It costs $2 per 1,000 pages, or $1 per 1,000 through the Batch API, which suits high-volume PDF to Markdown for RAG.
Choose Donut if
- It is MIT licensed and self-hosted, so you pay only for compute and no pages go to a third-party API.
- It maps document images straight to structured JSON with no separate OCR step, which fits receipt and form field extraction after fine-tuning.
- One HN practitioner reports that about 200 documents of fine-tuning can reach human-level accuracy on narrow data.
Key differences
| Mistral OCR | Donut | |
|---|---|---|
| Licence | Proprietary | MIT |
| How you run it | Hosted API (mistral-ocr-2512, OCR 3) | Self-host; available in Hugging Face Transformers |
| Price | $2 per 1,000 pages; $3 with structured annotations; $1 via Batch API | Free; you pay for compute |
| Output formats | Markdown, HTML, JSON, images | JSON, plain text |
| Setup before use | None; call the API | Base checkpoint must be fine-tuned before use |
| Benchmark | OmniDocBench v1.6 overall: 85.66 (v1.5 score of the first version, OCR 2503: 78.83) | Not published |
| Known failure | Hallucinated text on unreadable handwriting; changed checkbox states and misread characters on a dense financial table page | Misses tiny text in large images because of input resolution limits; Transformers v5 dropped image-to-text pipeline support, so you need direct loading or v4 |
Sources
- Mistral OCR 3 announcement
- OmniDocBench v1.6 leaderboard
- Reducto test of Mistral OCR
- donut-base model card
- Donut ECCV paper
- Hacker News thread
What each one is
Mistral OCR
Fast document OCR that returns clean Markdown with tables and embedded images.
Mistral 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.
Use one, fall back to the other
{"model": "mistral/mistral-ocr","fallbacks": ["naver/donut"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Mistral OCR alternativesDonut alternativesBest OCR models in 2026