OCRRouter

Hunyuan OCR vs SmolDocling

Pick Hunyuan OCR for photo and scene text, image translation, and low-resource or historical scripts, if you can run it under its licence. Pick SmolDocling only for small English experiments on modest hardware, because it is a preview release and its authors point users to Granite-Docling for support.

Hunyuan OCRtencent/hunyuan-ocrSmolDoclinghuggingface/smoldocling
TypeOpen sourceOpen source
Price2 credits / page1 credit / page
Per 1,000 pages$2.00$1.00
Typical speed1.5 s / page1.0 s / page
Best forChinese documents and photosVery cheap page conversion
InputsPDF, imagesPDF, images
Languages100+ languagesMainly English
Max pages per request500500
Tables✓ Yes✓ Yes
HandwritingNoNo
Math✓ Yes✓ Yes
Bounding boxes✓ Yes✓ Yes
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 Hunyuan OCR if

  • One ~1B model covers document parsing, text spotting, information extraction and image translation.
  • It handles images up to 4K and a 128K context window in v1.5, and exposes 12 task types including low-resource and ancient-script OCR.
  • It can run on a laptop through the llama.cpp GGUF path, so you are not limited to a CUDA server.

Choose SmolDocling if

  • At 256M parameters it fits modest hardware, and it takes about 0.35 seconds per page on an A100 with vLLM.
  • The licence is Apache-2.0 per the model summary, with no regional exclusions. The metadata tag says CDLA-Permissive-2.0, so check which applies.
  • It outputs DocTags for Docling, and one model handles text, tables (OTSL), code, formulas and charts with location tags.

Key differences

Hunyuan OCRSmolDocling
LicenceTencent Hunyuan Community License. It excludes the EU, UK and South Korea, and needs a separate licence above 100M monthly active users.Apache-2.0 per the model summary (metadata tag says CDLA-Permissive-2.0).
SizeAbout 1B parameters.256M parameters.
How you run itSelf-hosted via vLLM, Transformers or llama.cpp. The unified vLLM environment recommends CUDA 13 and needs vLLM 0.25.1+. A llama.cpp GGUF path covers CPU, consumer GPU or laptop.Self-hosted, small enough for modest hardware. About 0.35 seconds per page on an A100 with vLLM.
Output formatsMarkdown, HTML, LaTeX, JSON, plain text, bounding boxes.DocTags, Markdown, HTML, JSON, LaTeX, bounding boxes.
LanguagesIncludes low-resource and ancient-script OCR, and image translation.English only.
BenchmarksParseBench-2 overall 67.11 and ParseBench mean 63.86 (v1.5), both independent. Text score is 62.88 in a public CI replay.Table TEDS structure on FinTabNet 150dpi 0.82, full-page OCR F1 0.80, OCRBench 338, all reported by the authors.
Known failureAn archived 1.0 note said Transformers and vLLM outputs differed in accuracy.Output can loop endlessly or miss closing tags. Docling issue #1515 reports very slow generation on some images, even sparse pages.

Sources

What each one is

Hunyuan OCR

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

Hunyuan OCR details

SmolDocling

Tiny 256M vision model from Hugging Face and IBM that outputs DocTags. Predecessor of Granite Docling.

SmolDocling 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": "tencent/hunyuan-ocr","fallbacks": ["huggingface/smoldocling"],"file_url": "https://example.com/document.pdf"}

Questions

More comparisons

Hunyuan OCR alternativesSmolDocling alternativesBest OCR models in 2026