OCRRouter

Surya vs Hunyuan OCR

Pick Surya if you need layout detection, reading order and bounding boxes for 90+ languages, run locally on a laptop with about 4-6 GB of RAM. Pick Hunyuan OCR if you need photo and scene text spotting, image translation, or low-resource and historical scripts, and you are outside the EU, UK and South Korea. The main split is scope: Surya is a document layout and OCR toolkit, Hunyuan OCR is one end-to-end model for parsing, spotting, extraction and translation.

Suryadatalab/suryaHunyuan OCRtencent/hunyuan-ocr
TypeOpen sourceOpen source
Price2 credits / page2 credits / page
Per 1,000 pages$2.00$2.00
Typical speed1.5 s / page1.5 s / page
Best forLayout and line-level OCRChinese documents and photos
InputsPDF, imagesPDF, images
Languages90+ languages100+ languages
Max pages per request500500
Tables✓ Yes✓ Yes
HandwritingNoNo
MathNo✓ Yes
Bounding boxes✓ Yes✓ Yes
ConfidenceNoNo
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 Surya if

  • You need layout labels, reading order, HTML tables, KaTeX math and per-block bounding boxes from one model.
  • You want a small 689M model that scores 83.7% on olmOCR-bench and reaches 4.52 pages/s at 192 DPI on one H100 with vLLM.
  • You are in the EU, UK or South Korea, or your company is under $5M in funding and revenue, where the weights are free to use.

Choose Hunyuan OCR if

  • You need to read text in photos and natural scenes, or translate text inside images.
  • You work with low-resource or ancient scripts, or need images up to 4K and a 128K context window in v1.5.
  • You want one ~1B model with 12 task types via the vLLM client, covering parsing, spotting, extraction and translation.

Key differences

SuryaHunyuan OCR
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.Tencent Hunyuan Community License. Excludes the EU, UK and South Korea. Separate license needed above 100M monthly active users.
Model size689M parametersAbout 1B parameters
How you run itLight client against a separately running vLLM or llama.cpp server. vLLM on NVIDIA GPU in Docker or Podman, llama.cpp for CPU and Apple Silicon.vLLM, Transformers or llama.cpp. The unified vLLM environment recommends CUDA 13 and vLLM 0.25.1+. A llama.cpp GGUF path covers CPU, consumer GPU or laptop.
Output formatsJSON, HTML, LaTeX, bounding boxes, plain textMarkdown, HTML, LaTeX, JSON, plain text, bounding boxes
Benchmark83.7% on olmOCR-bench and 85.9 on OmniParseBench headline (Surya 2.5), both reported by the authors67.11 on ParseBench-2 overall and 63.86 ParseBench mean (v1.5), independent. Not published for olmOCR-bench.
Tables and hard documentsSurya 2.5 table output lacks row and column geometry. Multilingual tables score 58.1. The README sends handwriting, complex tables and math to the hosted Datalab API.Not published
Known failureApple M4 Max throughput is 0.198 pages/s at 96 DPI with llama.cpp Q8_0. Only 21 of 91 languages reach a 90% pass rate. Not for photos or natural scenes.Archived 1.0 note said Transformers and vLLM outputs differed in accuracy. ParseBench-2 text score is 62.88 in public CI replay.

Sources

What each one is

Surya

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

Surya details

Hunyuan OCR

Small end-to-end OCR model from Tencent covering parsing, extraction and photo translation.

Hunyuan 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/surya","fallbacks": ["tencent/hunyuan-ocr"],"file_url": "https://example.com/document.pdf"}

Questions

More comparisons

Surya alternativesHunyuan OCR alternativesBest OCR models in 2026