OCRRouter

PaddleOCR-VL vs Hunyuan OCR

Pick PaddleOCR-VL for multilingual scanned documents, academic papers with formulas, and tables and charts in reports. It covers 109 languages and scores 96.33 on OmniDocBench v1.6 (reported by the authors). Pick Hunyuan OCR for photo and scene text spotting, image translation, and low-resource or historical scripts, or if you want to run on a laptop through llama.cpp. Check its licence first, because it excludes the EU, UK and South Korea.

PaddleOCR-VLpaddle/paddleocr-vlHunyuan OCRtencent/hunyuan-ocr
TypeOpen sourceOpen source
Price6 credits / page2 credits / page
Per 1,000 pages$6.00$2.00
Typical speed1.3 s / page1.5 s / page
Best forCJK documentsChinese documents and photos
InputsPDF, imagesPDF, images
Languages109 languages100+ languages
Max pages per request500500
Tables✓ Yes✓ Yes
HandwritingNoNo
Math✓ Yes✓ Yes
Bounding boxes✓ Yes✓ Yes
Confidence✓ YesNo
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 PaddleOCR-VL if

  • It is Apache-2.0 licensed, so there are no regional exclusions or user-count limits.
  • It supports 109 languages, including Cyrillic, Arabic, Devanagari and Thai, and recognizes text, tables, formulas and charts on full pages.
  • At 0.9B parameters it runs on much smaller GPUs than 7B document VLMs like olmOCR.

Choose Hunyuan OCR if

  • One ~1B model handles document parsing, text spotting, information extraction and image translation.
  • It exposes 12 task types through the vLLM client, including low-resource and ancient-script OCR.
  • It runs on a CPU, consumer GPU or laptop through a llama.cpp GGUF conversion, and it outputs bounding boxes as well as Markdown, HTML, LaTeX, JSON and plain text.

Key differences

PaddleOCR-VLHunyuan OCR
LicenceApache-2.0Tencent Hunyuan Community License. It excludes the EU, UK and South Korea, and needs a separate licence above 100M monthly active users.
How you run itSelf-host via PaddleOCR, vLLM or transformers. NVIDIA GPU recommended, CPU possible but slow. The transformers path does element-level recognition only.Self-host via vLLM, Transformers or llama.cpp. The unified vLLM environment needs CUDA 13 and vLLM 0.25.1+. The llama.cpp GGUF path runs on CPU, consumer GPU or laptop.
PriceFree, self-hosted (you pay for compute)Free, self-hosted (you pay for compute)
Output formatsMarkdown, JSONMarkdown, HTML, LaTeX, JSON, plain text, bounding boxes
BenchmarksOmniDocBench v1.6 overall 96.33 and v1.5 overall 94.5 (authors). olmOCR-Bench overall 80.0 (independent).ParseBench-2 overall 67.11 and ParseBench mean 63.86 (independent). Not published on OmniDocBench.
ScopeDocument parsing: text, tables, formulas and charts. 109 languages.Parsing, text spotting, information extraction and image translation. Images up to 4K and a 128K context window in v1.5.
Known failureVersion 1.6 regresses Russian and Kazakh Cyrillic on born-digital PDFs, hangs on blurry images, and ships use_cache false, which makes decoding about 11x slower. Users also report missing table content.Archived 1.0 note said Transformers and vLLM outputs differed in accuracy. Text score is 62.88 on ParseBench-2 in public CI replay.

Sources

What each one is

PaddleOCR-VL

Open-source document model from Baidu. Excellent on Chinese, Japanese and Korean.

PaddleOCR-VL 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": "paddle/paddleocr-vl","fallbacks": ["tencent/hunyuan-ocr"],"file_url": "https://example.com/document.pdf"}

Questions

More comparisons

PaddleOCR-VL alternativesHunyuan OCR alternativesBest OCR models in 2026