Marker vs Amazon Textract
Pick Marker if you want PDFs turned into Markdown, HTML or RAG chunks on your own hardware, especially books and papers with equations, and your company is under the $5M weight-licence line. Pick Amazon Textract if you already run on AWS and need English forms, invoices or big scanned batches through a managed API. The main split is self-hosted Markdown output versus a hosted, pay-per-page service that returns JSON with geometry.
Markerdatalab/marker | Amazon Textractaws/textract | |
|---|---|---|
| Type | OCR API | OCR API |
| Price | 14 credits / page | 25 credits / page |
| Per 1,000 pages | $14.00 | $25.00 |
| Typical speed | 2.5 s / page | 1.2 s / page |
| Best for | Books and research papers | Forms and key-value extraction |
| Inputs | PDF, images, DOCX | PDF, images |
| Languages | 90+ languages | 6 languages |
| Max pages per request | 1,000 | 3,000 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | ✓ Yes |
| Math | ✓ Yes | No |
| Bounding boxes | ✓ Yes | ✓ Yes |
| Confidence | No | ✓ Yes |
| 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 Markdown, HTML, JSON block trees or RAG chunks straight from the converter.
- You need to run locally on GPU, CPU or Apple Silicon, with a fast mode that does 23.6 pages per second on CPU.
- You work with books and papers with equations and want the code under Apache-2.0 and free self-hosting within the weight licence.
Choose Amazon Textract if
- You need English forms, invoices and queries extracted through one AWS API.
- You have large scanned PDF batches: the async API handles files up to 500 MB and 3,000 pages.
- You want word and line geometry with confidence scores, and a flat per-page price such as $1.50 per 1,000 pages for text detection.
Key differences
| Marker | Amazon Textract | |
|---|---|---|
| Licence | Code Apache-2.0; weights under a modified AI Pubs Open Rail-M licence, free for research, personal use and startups under $5M. | Proprietary. |
| How you run it | Local on GPU, CPU or Apple Silicon, or Datalab's hosted API. vllm mode needs an NVIDIA GPU. | Hosted AWS API only. The sync API takes one page; the async API takes PDFs up to 500 MB and 3,000 pages. |
| Price | Free self-hosted (you pay for compute) within the weight licence. Hosted API has about 5,000 free pages per month, then paid plans. | US West (Oregon), first 1M pages a month: text detection $0.0015/page, Tables $0.015, Queries $0.015, Forms $0.05. |
| Output formats | Markdown, HTML, JSON, chunks, LaTeX, images. | JSON, plain text, bounding boxes. Markdown for RAG needs post-processing. |
| Benchmarks | 83.7 on olmOCR-bench in vllm mode, 66.6 in CPU fast mode. These are Datalab's own numbers. | 80.9% table accuracy on RD-TableBench, reported by an independent source. The two scores are different tests and do not compare directly. |
| Languages and handwriting | Not published | Text detection supports English, French, German, Italian, Portuguese and Spanish. Handwriting and Queries are English only. |
| Known failure | Weight licence blocks free commercial use for companies above $5M. CPU fast mode scores lower than vllm mode. | Vertical Japanese and Chinese text is not supported. Not suited to non-Latin languages. |
Sources
- Marker GitHub README
- Datalab pricing
- Textract pricing
- Textract quotas
- Textract hard limits
- Reducto RD-TableBench
What each one is
Marker
Hosted version of the open-source Marker and Surya models. Strong on books and papers.
Marker detailsAmazon Textract
AWS OCR with form key-value pairs, tables and per-word confidence scores.
Amazon Textract 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": ["aws/textract"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Marker alternativesAmazon Textract alternativesBest OCR models in 2026