Marker vs Qianfan OCR
Pick Marker if you want self-hosted PDF to Markdown for books and papers with equations, or RAG chunks, and you can run on CPU or Apple Silicon. Pick Qianfan OCR if you need multilingual parsing, handwriting, key information extraction, or free commercial use under Apache-2.0. The main split is the licence and hardware: Marker's weights limit commercial use above $5M, while Qianfan OCR is Apache-2.0 but is benchmarked only on an A100 GPU.
Markerdatalab/marker | Qianfan OCRbaidu/qianfan-ocr | |
|---|---|---|
| Type | OCR API | OCR API |
| Price | 14 credits / page | 2 credits / page |
| Per 1,000 pages | $14.00 | $2.00 |
| Typical speed | 2.5 s / page | 2.0 s / page |
| Best for | Books and research papers | Chinese documents |
| Inputs | PDF, images, DOCX | PDF, images |
| Languages | 90+ languages | Chinese, English and more |
| Max pages per request | 1,000 | 200 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | ✓ Yes |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | ✓ Yes | No |
| Confidence | No | No |
| Forms | No | No |
| Multilingual | ✓ Yes | ✓ Yes |
| 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
- It runs on CPU, Apple Silicon or GPU, and fast mode runs on CPU at 23.6 pages per second with about 18 MB of ONNX weights.
- It outputs Markdown, HTML, JSON block trees and RAG chunks, which suits books and papers with equations.
- Datalab also sells a hosted API with about 5,000 free pages per month if you do not want to host it.
Choose Qianfan OCR if
- The Apache-2.0 licence allows commercial use with no revenue cap.
- It covers 192 languages, handwriting and scene text, and is listed as good for key information extraction, tables and formulas.
- It scores 93.12 on OmniDocBench v1.5, the top end-to-end score in Baidu's table.
Key differences
| Marker | Qianfan OCR | |
|---|---|---|
| Licence | Code Apache-2.0; weights under a modified AI Pubs Open Rail-M license, free for research, personal use and startups under $5M. | Apache-2.0, commercial use allowed. |
| Hardware | GPU, CPU or MPS; vllm mode needs an NVIDIA GPU; fast mode runs on CPU. | Benchmarked on one NVIDIA A100 with vLLM; no minimum VRAM stated; not suited to CPU-only deployment. |
| Hosted option | Datalab hosted API, about 5,000 free pages per month, then paid plans. | Baidu Qianfan platform. |
| Output formats | Markdown, HTML, JSON, chunks, LaTeX, images. | Markdown, JSON, HTML, LaTeX, bounding boxes. |
| olmOCR-bench | 83.7 in vllm mode, 66.6 in CPU fast mode (Datalab's own numbers). | 79.8 (reported by the authors). |
| Speed | 23.6 pages per second in CPU fast mode. | 1.024 pages per second on an A100 even with W8A8. |
| Known failure | CPU fast mode drops accuracy; all README benchmarks are Datalab's own. | Thinking mode can emit up to 16,384 tokens, adding latency; PaddleOCR-VL 1.5 scores higher at 94.50 on OmniDocBench v1.5. |
Sources
What each one is
Marker
Hosted version of the open-source Marker and Surya models. Strong on books and papers.
Marker detailsQianfan OCR
Baidu's OCR-tuned vision model. Strong on Chinese documents and tables.
Qianfan OCR 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": ["baidu/qianfan-ocr"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Marker alternativesQianfan OCR alternativesBest OCR models in 2026