OCRRouter

Donut vs Typhoon OCR

Pick Typhoon OCR if you need ready-to-use full-page conversion of Thai or English documents into Markdown, HTML tables or LaTeX. Pick Donut if you have a narrow, fixed-layout document type, such as receipts or forms, and can fine-tune a model for field extraction or classification. The main split is that Donut is a base model you must train before use, while Typhoon OCR is built for page-level reading out of the box.

Donutnaver/donutTyphoon OCRtyphoon/typhoon-ocr
TypeOpen sourceOpen source
Price1 credit / page2 credits / page
Per 1,000 pages$1.00$2.00
Typical speed1.0 s / page1.8 s / page
Best forReceipt and form fields after fine-tuningThai documents
InputsimagesPDF, images
LanguagesEnglish, Chinese, Japanese, KoreanThai, English
Max pages per request100500
TablesNo✓ Yes
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 want to fine-tune on a fixed-layout document type, and a practitioner on HN reports that about 200 documents can reach human-level accuracy on narrow data.
  • You need JSON field extraction or document classification straight from the image, with no separate OCR step.
  • You want a plain MIT licence with no extra terms, and loading through Hugging Face Transformers.

Choose Typhoon OCR if

  • You work with Thai or Thai-English documents such as government forms, financial reports or books; V1.5 scored BLEU 0.870 on Thai government forms, ahead of GPT-5 and Gemini 2.5 Pro.
  • You need full-page output with tables as HTML (merged cells kept), equations as LaTeX and checkboxes marked.
  • You want the choice of self-hosting the 2B model (or a 0.8B V2 variant) or calling it on the Typhoon API as typhoon-ocr-v2.

Key differences

DonutTyphoon OCR
Intended useBase model to fine-tune for parsing, classification or VQADocument OCR for Thai and English pages
How you run itSelf-host onlySelf-host, or the Typhoon API as typhoon-ocr-v2
Output formatsJSON, plain textMarkdown, HTML, LaTeX
Published benchmarksNot publishedolmOCR-bench 84.7, OmniDocBench v1.6 90.97, ThaiOCRBench 59.9 (V2 2B, authors' numbers)
PriceFree, self-hosted (you pay for compute)Free, self-hosted (you pay for compute). API pricing not published on the model card
Known failureMisses tiny text in large images because of input resolution limits; the base checkpoint is unusable until fine-tunedV2 can loop until the token limit on dense multi-column pages; key-information extraction is weakest at 21.0 on ThaiOCRBench

Sources

What each one is

Donut

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

Donut details

Typhoon OCR

Open vision model tuned for Thai and English documents, with Markdown output.

Typhoon OCR 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": ["typhoon/typhoon-ocr"],"file_url": "https://example.com/document.pdf"}

Questions

More comparisons

Donut alternativesTyphoon OCR alternativesBest OCR models in 2026