OCRRouter

Donut vs GOT-OCR 2.0

Pick GOT-OCR 2.0 if you want a model that reads text, formulas, tables, charts and sheet music without training first. Pick Donut if you have a narrow, fixed-layout document type such as receipts or forms and can fine-tune it to return structured JSON. The main split is that Donut is a pretrain-only base you must fine-tune, while GOT-OCR 2.0 is a general OCR model you can run directly.

Donutnaver/donutGOT-OCR 2.0stepfun/got-ocr-2
TypeOpen sourceOpen source
Price1 credit / page1 credit / page
Per 1,000 pages$1.00$1.00
Typical speed1.0 s / page1.5 s / page
Best forReceipt and form fields after fine-tuningFormatted text, formulas and charts
Inputsimagesimages
LanguagesEnglish, Chinese, Japanese, KoreanChinese, English
Max pages per request100100
TablesNo✓ Yes
HandwritingNoNo
MathNo✓ Yes
Bounding boxesNo✓ Yes
ConfidenceNoNo
Forms✓ YesNo
MultilingualNoNo
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 are extracting fields from receipts or forms with a fixed layout, and you can fine-tune on your own samples.
  • You need JSON output straight from the document image, with no separate OCR step.
  • You want a plain MIT licence with no data restrictions.

Choose GOT-OCR 2.0 if

  • You need to read formulas, sheet music or chemical structures, which GOT-OCR 2.0 lists as strengths.
  • You want OCR on a chosen region of a page, guided by box coordinates or color, or multi-page OCR with dynamic resolution.
  • You want a model that works without a fine-tuning step, with Markdown, LaTeX and plain text output.

Key differences

DonutGOT-OCR 2.0
LicenceMITCode Apache-2.0; training data CC BY-NC 4.0
How you run itSelf-hosted and free; you pay for compute. Available in Hugging Face Transformers. Transformers v5 dropped image-to-text pipeline support, so load it directly or use v4.Self-hosted and free; you pay for compute. 580M parameters, merged into Hugging Face Transformers with batched inference. README lists CUDA 11.8 and PyTorch 2.0.1; CPU users are pointed to a llama.cpp port.
Output formatsJSON, plain textPlain text, Markdown, LaTeX, TikZ, SMILES, kern
Best document typesReceipt and form field extraction, document classification, narrow fixed-layout documentsFormulas, sheet music, chemical structures, region-specific OCR, English and Chinese documents
Known failureMisses tiny text in large images because of input resolution limits. One HN user found results on personal documents did not match the promised scores.Issues report output looping characters midway, wrong text on images with no text instead of an empty string, and much worse results in format mode than in plain OCR mode.
BenchmarksNot publishedNot published

Sources

What each one is

Donut

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

Donut details

GOT-OCR 2.0

Unified OCR model that reads plain text, tables, formulas, sheet music and charts.

GOT-OCR 2.0 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": ["stepfun/got-ocr-2"],"file_url": "https://example.com/document.pdf"}

Questions

More comparisons

Donut alternativesGOT-OCR 2.0 alternativesBest OCR models in 2026