PaddleOCR (PP-OCR) vs Donut
Pick PaddleOCR (PP-OCR) when you need text lines and bounding boxes from printed Chinese, English or other Latin-script text today, with no training step. Pick Donut when your documents have a fixed layout, such as receipts or forms, and you can fine-tune it to return JSON fields directly. The main split: PP-OCR works out of the box as a detector plus recognizer, while Donut is a base model that must be fine-tuned before it does anything useful.
PaddleOCR (PP-OCR)paddle/paddleocr | Donutnaver/donut | |
|---|---|---|
| Type | Open source | Open source |
| Price | 1 credit / page | 1 credit / page |
| Per 1,000 pages | $1.00 | $1.00 |
| Typical speed | 0.7 s / page | 1.0 s / page |
| Best for | Fast printed text, Chinese and English | Receipt and form fields after fine-tuning |
| Inputs | PDF, images | images |
| Languages | 80+ languages | English, Chinese, Japanese, Korean |
| Max pages per request | 500 | 100 |
| Tables | No | No |
| Handwriting | No | No |
| Math | No | No |
| Bounding boxes | ✓ Yes | No |
| Confidence | ✓ Yes | 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 PaddleOCR (PP-OCR) if
- It works without training: one PP-OCRv6 model covers 50 languages, including Chinese, English, Japanese and 46 Latin-script languages.
- It runs on CPU, NVIDIA GPU and other accelerators, and in the browser through the PaddleOCR.js SDK, which suits edge and mobile OCR.
- PP-OCRv6 medium is 34.5M parameters and reports 0.13s on an A100 and a 5.2x CPU speedup with OpenVINO.
Choose Donut if
- It maps a document image straight to JSON with no separate OCR step, so field extraction from receipts and forms is one model.
- After fine-tuning it fits narrow, fixed-layout document types and document classification; one HN practitioner reports about 200 documents can reach human-level accuracy on narrow data.
- It is MIT licensed and available in Hugging Face Transformers.
Key differences
| PaddleOCR (PP-OCR) | Donut | |
|---|---|---|
| What it outputs | Plain text, bounding boxes and JSON of text lines. Tables and formulas need the separate PP-Structure or VL pipelines. | JSON or plain text, mapped straight from the image with no external OCR step. |
| Training needed | None stated. Use the released PP-OCRv6 models directly. | Yes. The base checkpoint is pretrain-only and must be fine-tuned before use. |
| Licence | Apache-2.0 | MIT |
| How you run it | Library: the paddleocr Python package, C++, or the PaddleOCR.js browser SDK. | Self-hosted model, loaded through Hugging Face Transformers. |
| Price | Free, self-hosted. You pay for compute. | Free, self-hosted. You pay for compute. |
| Languages | 50 languages in one PP-OCRv6 model, including Chinese, English, Japanese and 46 Latin-script languages. | Not published |
| Known failure | PP-OCRv6 weights contain denormals that make CPU inference 4 to 11x slower (RapidOCR issue #743). Version bugs are open in issues #18162 and #18445. | 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
- PaddleOCR GitHub README
- PP-OCRv6 on Hugging Face
- RapidOCR issue #743 denormal weights
- donut-base model card
- Donut ECCV paper
- Hacker News thread
What each one is
PaddleOCR (PP-OCR)
Baidu's classic open-source text detection and recognition pipeline. Light enough to run on CPU.
PaddleOCR (PP-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": "paddle/paddleocr","fallbacks": ["naver/donut"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
PaddleOCR (PP-OCR) alternativesDonut alternativesBest OCR models in 2026