OCRRouter

Surya vs Phi-4-multimodal

Pick Surya when you need layout, reading order and bounding boxes, or when you want OCR on a laptop or CPU across 90+ languages. Pick Phi-4-multimodal for English document Q&A on a modest GPU, or when one model must handle both audio and images. The main split is that Surya returns structure and coordinates, while Phi-4-multimodal answers questions about English documents and gives no coordinates.

Suryadatalab/suryaPhi-4-multimodalmicrosoft/phi-4-multimodal
TypeOpen sourceVision model
Price2 credits / page2 credits / page
Per 1,000 pages$2.00$2.00
Typical speed1.5 s / page2.5 s / page
Best forLayout and line-level OCRSmall model for OCR and questions
InputsPDF, imagesimages
Languages90+ languagesEnglish for vision
Max pages per request500500
Tables✓ Yes✓ Yes
HandwritingNoNo
MathNo✓ 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 Surya if

  • It returns layout labels, reading order, OCR, HTML tables and KaTeX math with per-block bounding boxes from one model.
  • It covers 90+ languages, while Phi-4-multimodal supports vision in English only.
  • It runs on CPU and Apple Silicon through llama.cpp, with peak RAM of about 4-6 GB on an M4 Max.

Choose Phi-4-multimodal if

  • It scores 93.2 on DocVQA at 5.6B parameters, which suits English document question answering.
  • It uses the MIT license, so there is no revenue or funding threshold like Surya's $5M limit.
  • It takes audio as well as images, and has a 128K token context, so mixed pipelines need only one model.

Key differences

SuryaPhi-4-multimodal
LicenceCode Apache-2.0; weights under a modified AI Pubs Open Rail-M. Free for research, personal use and companies under $5M funding or revenue.MIT
How you run itSelf-host only. A lightweight client talks to a separate vLLM or llama.cpp inference server.Self-host, or use it through Azure AI Foundry.
Size and hardware689M parameters. NVIDIA GPU via vLLM, or CPU and Apple Silicon via llama.cpp.5.6B parameters. Flash attention on A100, A6000 and H100. V100 and older need eager attention.
OutputJSON, HTML, LaTeX, bounding boxes, plain textPlain text, Markdown, JSON. Not suited to jobs needing coordinates or layout.
BenchmarksolmOCR-bench 83.7%, OmniParseBench 85.9, 91-language pass rate 74.9% (author-reported)DocVQA 93.2, OCRBench 84.4, ChartQA 81.4, InfoVQA 72.7 (author-reported)
Languages90+ languages. Multilingual tables score only 58.1, and 21 of 91 languages reach a 90% pass rate.Vision is English only, and performance drops in other languages.
Known failureSurya 2.5 table output lacks row and column geometry. Handwriting and complex tables are better sent to the hosted Datalab API. Not for photos or natural scenes.The model card warns of repetitive or inconsistent output in long chats.

Sources

What each one is

Surya

Open-source OCR toolkit with line detection, layout analysis, reading order and tables.

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

Questions

More comparisons

Surya alternativesPhi-4-multimodal alternativesBest OCR models in 2026