OCRRouter

Donut vs Pixtral

Pick Pixtral only if you already run on the Mistral API and need question answering over documents or charts. Pick Donut for a narrow, fixed-layout document type such as receipts or forms where you can fine-tune. Pixtral 12B is marked deprecated by Mistral, so avoid it for new production OCR.

Donutnaver/donutPixtralmistral/pixtral
TypeOpen sourceVision model
Price1 credit / page2 credits / page
Per 1,000 pages$1.00$2.00
Typical speed1.0 s / page3.0 s / page
Best forReceipt and form fields after fine-tuningQuestions about document images
Inputsimagesimages
LanguagesEnglish, Chinese, Japanese, KoreanMultilingual
Max pages per request100500
TablesNoNo
HandwritingNoNo
MathNoNo
Bounding boxesNoNo
ConfidenceNoNo
Forms✓ YesNo
MultilingualNo✓ Yes
On OCRRouterAvailableAvailable

Prices are what you pay on OCRRouter. Speed is the provider's typical figure, not a benchmark we ran.

Which to choose

Choose Donut if

  • You can fine-tune it on one fixed-layout document type, and a practitioner reports about 200 documents can reach human-level accuracy on narrow data.
  • It maps the page image straight to JSON with no separate OCR step, which suits receipt and form field extraction.
  • It is MIT licensed, runs only self-hosted, and costs nothing beyond your compute.

Choose Pixtral if

  • You already have an integration on the Mistral API and want to keep it.
  • You need question answering over charts and documents, where Mistral reports 90.7 ANLS on DocVQA and 81.8 on ChartQA.
  • You want a 128K token context so one request can carry many images.

Key differences

DonutPixtral
Maintenance statusPublished at ECCV 2022 as a base model to fine-tune; no deprecation noted.Marked deprecated and no longer maintained by Mistral.
LicenceMITApache-2.0 open weights
How you run itSelf-host onlyMistral API (pixtral-12b-2409) or self-host
PriceFree, you pay for compute$0.15 per million input and output tokens at API launch; free to self-host
Ready to useNo. The base checkpoint is pretrain-only and needs fine-tuning.General vision language model that answers prompts without training
Output formatsJSON, plain textPlain text, Markdown, JSON
Known failureMisses tiny text in large images because of input resolution limits; Transformers v5 dropped its image-to-text pipeline support.Does not always get OCR right; accuracy drops reported on handwriting, small fonts and non-Latin scripts; Q4 quantization degrades small text.

Sources

What each one is

Donut

OCR-free document understanding model. Fine-tuned for receipts, forms and document Q&A.

Donut details

Pixtral

Mistral's open-weight vision models, Pixtral 12B and Pixtral Large. Mistral's API no longer serves them, so they run self-hosted.

Pixtral details

Try 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": "naver/donut","fallbacks": ["mistral/pixtral"],"file_url": "https://example.com/document.pdf"}

Questions

More comparisons

Donut alternativesPixtral alternativesBest OCR models in 2026