OCRRouter

Marker vs PaddleOCR (PP-OCR)

Pick Marker when you need whole PDFs, books or papers turned into Markdown, HTML or JSON, with equations and RAG chunks. Pick PaddleOCR (PP-OCR) when you need fast text lines and boxes from printed Chinese, English, receipts, signs or screenshots, or need OCR on edge devices and in the browser. The main split is scope: Marker converts documents, PP-OCR only detects and recognises text, and Marker's weights carry a commercial limit that PP-OCR's Apache-2.0 licence does not.

Markerdatalab/markerPaddleOCR (PP-OCR)paddle/paddleocr
TypeOCR APIOpen source
Price14 credits / page1 credit / page
Per 1,000 pages$14.00$1.00
Typical speed2.5 s / page0.7 s / page
Best forBooks and research papersFast printed text, Chinese and English
InputsPDF, images, DOCXPDF, images
Languages90+ languages80+ languages
Max pages per request1,000500
Tables✓ YesNo
HandwritingNoNo
Math✓ YesNo
Bounding boxes✓ Yes✓ Yes
ConfidenceNo✓ Yes
FormsNoNo
Multilingual✓ Yes✓ Yes
On OCRRouterAvailableAvailable

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 PDF to Markdown, HTML, JSON block trees or RAG chunks, not just text lines.
  • Your documents are books or papers with equations, and Datalab reports 83.7 on olmOCR-bench in vllm mode.
  • You want a hosted option too: Datalab's API includes about 5,000 free pages per month.

Choose PaddleOCR (PP-OCR) if

  • You need a licence with no commercial limit: the whole project is Apache-2.0, while Marker's weights are not free for companies above $5M.
  • You run on edge, mobile or in the browser: PP-OCRv6 medium is 34.5M parameters and the PaddleOCR.js SDK runs it in a browser.
  • You read Chinese, English, Japanese or Latin-script text: one PP-OCRv6 model covers 50 languages.

Key differences

MarkerPaddleOCR (PP-OCR)
What it doesPDF to Markdown converter built on the Surya OCR modelsText detection and recognition pipeline; returns text lines and boxes only
LicenceCode Apache-2.0; weights under a modified AI Pubs Open Rail-M licence (free for research, personal use and startups under $5M)Apache-2.0
How you run itLocal on GPU, CPU or Apple Silicon, or Datalab's hosted APILocal via the paddleocr Python package, C++ or the PaddleOCR.js browser SDK; no hosted API listed
Output formatsMarkdown, HTML, JSON, chunks, LaTeX, imagesPlain text, bounding boxes, JSON
Benchmarks83.7 on olmOCR-bench and 85.9 on OmniParseBench in vllm mode (Datalab's own numbers); 66.6 on olmOCR-bench in CPU fast modeNot published
Speed and hardwareFast mode runs on CPU at 23.6 pages per second with about 18 MB of ONNX weights; vllm mode needs an NVIDIA GPU0.13s on A100 and 5.2x CPU speedup with OpenVINO, per the README; runs on CPU, NVIDIA GPU, Kunlunxin XPU
Known failureFast mode drops to 66.6 on olmOCR-bench; all published benchmark numbers come from DatalabTables and formulas need the separate PP-Structure or VL pipelines; PP-OCRv6 weights have denormals that slow CPU inference 4 to 11x (RapidOCR issue #743)

Sources

What each one is

Marker

Hosted version of the open-source Marker and Surya models. Strong on books and papers.

Marker details

PaddleOCR (PP-OCR)

Baidu's classic open-source text detection and recognition pipeline. Light enough to run on CPU.

PaddleOCR (PP-OCR) details

Try 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": "datalab/marker","fallbacks": ["paddle/paddleocr"],"file_url": "https://example.com/document.pdf"}

Questions

More comparisons

Marker alternativesPaddleOCR (PP-OCR) alternativesBest OCR models in 2026