Marker vs Donut
Pick Marker when you need full-page PDF to Markdown, HTML or JSON from books, papers with equations, or RAG pipelines, and you can live with its weight licence. Pick Donut when you have a narrow, fixed-layout document type such as receipts or forms and are ready to fine-tune a model for field extraction. The main split is that Marker converts whole pages out of the box, while Donut is a pretrain-only base model that does nothing useful until you train it.
Markerdatalab/marker | Donutnaver/donut | |
|---|---|---|
| Type | OCR API | Open source |
| Price | 14 credits / page | 1 credit / page |
| Per 1,000 pages | $14.00 | $1.00 |
| Typical speed | 2.5 s / page | 1.0 s / page |
| Best for | Books and research papers | Receipt and form fields after fine-tuning |
| Inputs | PDF, images, DOCX | images |
| Languages | 90+ languages | English, Chinese, Japanese, Korean |
| Max pages per request | 1,000 | 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 Marker if
- You want full-page PDF to Markdown, with HTML, JSON block trees, RAG chunks or LaTeX as other output options.
- You work with books and papers that contain equations and want something that runs without any training.
- You need a CPU option: fast mode runs on CPU with about 18 MB of ONNX weights at 23.6 pages per second, though olmOCR-bench drops to 66.6.
Choose Donut if
- You process one narrow, fixed-layout document type, such as receipts or forms, and want structured JSON fields from it.
- You want an MIT licence with no revenue cap, and a model you can load through Hugging Face Transformers.
- You want to skip the separate OCR step, since Donut maps document images straight to structured output.
Key differences
| Marker | Donut | |
|---|---|---|
| Licence | Code Apache-2.0. Weights under a modified AI Pubs Open Rail-M licence, free for research, personal use and startups under $5M. | MIT. |
| How you run it | Self-host on GPU, CPU or Apple Silicon (MPS), or use Datalab's hosted API. vllm mode needs an NVIDIA GPU. | Self-host only. |
| Price | Free self-hosted within the weight licence, and you pay for compute. The hosted API includes about 5,000 free pages per month. | Free self-hosted, and you pay for compute. |
| Ready to use | Converts PDFs as is. | The base checkpoint is pretrain-only and must be fine-tuned before use. |
| Output formats | Markdown, HTML, JSON, chunks, LaTeX, images. | JSON, plain text. |
| Benchmarks | 83.7 on olmOCR-bench and 85.9 on OmniParseBench in vllm mode. These are Datalab's own numbers. | Not published. |
| Known failure | Commercial use is blocked above $5M without a paid licence. CPU fast mode scores 66.6 on olmOCR-bench. | 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
What each one is
Marker
Hosted version of the open-source Marker and Surya models. Strong on books and papers.
Marker 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": "datalab/marker","fallbacks": ["naver/donut"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Marker alternativesDonut alternativesBest OCR models in 2026