OCRRouter

Marker vs Phi-4-multimodal

Pick Marker to turn PDFs of books and papers into Markdown, HTML, JSON or RAG chunks on your own hardware, as long as your company is under the $5M weight-license limit. Pick Phi-4-multimodal for English document Q&A under an MIT licence, or when you also need audio in the same model. The main split is the job: Marker converts whole PDFs, Phi-4-multimodal answers questions about page images.

Markerdatalab/markerPhi-4-multimodalmicrosoft/phi-4-multimodal
TypeOCR APIVision model
Price14 credits / page2 credits / page
Per 1,000 pages$14.00$2.00
Typical speed2.5 s / page2.5 s / page
Best forBooks and research papersSmall model for OCR and questions
InputsPDF, images, DOCXimages
Languages90+ languagesEnglish for vision
Max pages per request1,000500
Tables✓ Yes✓ Yes
HandwritingNoNo
Math✓ Yes✓ Yes
Bounding boxes✓ YesNo
ConfidenceNoNo
FormsNoNo
Multilingual✓ YesNo
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 with equations, and Datalab reports 83.7 on olmOCR-bench in vllm mode.
  • You need RAG chunks, JSON block trees, HTML or LaTeX output, not only text.
  • You have no GPU: fast mode runs on CPU at 23.6 pages per second, though it drops to 66.6 on olmOCR-bench.

Choose Phi-4-multimodal if

  • You need permissive MIT licensing for commercial use with no revenue cap.
  • You want to ask questions about English documents, where Microsoft reports DocVQA 93.2 and OCRBench 84.4 at 5.6B parameters.
  • You run mixed audio and image pipelines, since one model takes both, with a 128K token context.

Key differences

MarkerPhi-4-multimodal
LicenceCode Apache-2.0; weights under a modified AI Pubs Open Rail-M licence, free for research, personal use and startups under $5MMIT
Main jobPDF to Markdown converter built on the Surya OCR modelsGeneral multimodal model taking text, image and audio input
Output formatsMarkdown, HTML, JSON, chunks, LaTeX, imagesPlain text, Markdown, JSON
HardwareGPU, CPU or Apple Silicon; vllm mode needs an NVIDIA GPU; fast mode uses about 18 MB of ONNX weights on CPUFlash attention, tested on A100, A6000 and H100; V100 and older need eager attention
Hosted option and priceFree self-hosted within the weight licence; Datalab hosted API with about 5,000 free pages per monthFree self-hosted (you pay for compute); also available via Azure AI Foundry
BenchmarksolmOCR-bench 83.7 and OmniParseBench 85.9 in vllm mode, all from DatalabDocVQA 93.2, OCRBench 84.4, ChartQA 81.4, InfoVQA 72.7, all from Microsoft
Known weaknessCommercial use blocked for companies above $5M without a paid licenceVision is English only, weaker in other languages, and long chats can give repetitive output

Sources

What each one is

Marker

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

Marker details

Phi-4-multimodal

Microsoft's 5.6B text, vision and audio model (February 2025). Scores 84.4 on OCRBench per its model card.

Phi-4-multimodal 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": ["microsoft/phi-4-multimodal"],"file_url": "https://example.com/document.pdf"}

Questions

More comparisons

Marker alternativesPhi-4-multimodal alternativesBest OCR models in 2026