DeepSeek-OCR vs Donut
Pick DeepSeek-OCR when you want full-page text out of mixed documents and can run a CUDA GPU. It gives Markdown, HTML and bounding boxes with no training. Pick Donut when your documents are a narrow, fixed-layout type such as receipts or forms and you can fine-tune it to return JSON fields. The main split is that DeepSeek-OCR is a page reader, while Donut is a base model you must train for one job.
DeepSeek-OCRdeepseek/deepseek-ocr | Donutnaver/donut | |
|---|---|---|
| Type | Open source | Open source |
| Price | 1 credit / page | 1 credit / page |
| Per 1,000 pages | $1.00 | $1.00 |
| Typical speed | 1.1 s / page | 1.0 s / page |
| Best for | Cheap bulk conversion | Receipt and form fields after fine-tuning |
| Inputs | PDF, images | images |
| Languages | 100+ languages | English, Chinese, Japanese, Korean |
| Max pages per request | 200 | 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 DeepSeek-OCR if
- It works without fine-tuning and returns Markdown, plain text, HTML or bounding boxes.
- The authors quote about 2,500 tokens per second of PDF throughput on one A100-40G, which suits cheap bulk OCR.
- You can set resolution modes from 64 to 400 vision tokens per page to trade cost against detail.
Choose Donut if
- It maps a document image straight to JSON, so you skip a separate OCR step and field-parsing code.
- A practitioner on Hacker News reports that about 200 fine-tuning documents can reach human-level accuracy on narrow data.
- It suits receipt and form field extraction and document classification once fine-tuned.
Key differences
| DeepSeek-OCR | Donut | |
|---|---|---|
| Out-of-the-box use | Ready to run for page OCR with Transformers or vLLM. | Base checkpoint is pretrain-only and must be fine-tuned before use. |
| Output formats | Markdown, plain text, HTML, bounding boxes. | JSON, plain text. |
| Licence and price | MIT, free to self-host; you pay for compute. | MIT, free to self-host; you pay for compute. |
| Hardware and setup | CUDA GPU; reference setup is CUDA 11.8 and PyTorch 2.6 in bfloat16. | Available in Hugging Face Transformers, but v5 dropped image-to-text pipeline support, so you need direct loading or v4. |
| Best document types | Bulk page OCR and research on token-efficient OCR. | Receipts, forms and other narrow, fixed-layout documents. |
| Large pages and small text | Not published | Misses tiny text in large images because of input resolution limits. |
| Known failure | Repeated tokens and looping over table headers (GitHub issues #12 and #16); it corrects nonsense words instead of copying them. | One HN user found results on personal documents did not match the promised scores. |
Sources
- DeepSeek-OCR GitHub
- Visual Merit or Linguistic Crutch paper
- 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": "deepseek/deepseek-ocr","fallbacks": ["naver/donut"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
DeepSeek-OCR vs GOT-OCR 2.0DeepSeek-OCR vs NougatJina OCR vs DonutGoogle Document AI vs DonutLlamaParse vs DonutDeepSeek-OCR vs OpenOCRHunyuan OCR vs DonutTrOCR vs Donut
DeepSeek-OCR alternativesDonut alternativesBest OCR models in 2026