OCRRouter

GPT vs Hunyuan OCR

Pick GPT when you call a hosted API and need handwriting, custom JSON extraction or document question answering. Pick Hunyuan OCR when you want to self-host a roughly 1B model, need bounding boxes, or work with scene text, translation and historical scripts, and you are outside the EU, UK and South Korea. The main split is hosted per-token API versus a free self-hosted model with a restricted licence.

GPTopenai/gptHunyuan OCRtencent/hunyuan-ocr
TypeVision modelOpen source
Price11 credits / page2 credits / page
Per 1,000 pages$11.00$2.00
Typical speed2.4 s / page1.5 s / page
Best forMessy scans, custom instructionsChinese documents and photos
InputsPDF, imagesPDF, images
Languages50+ languages100+ languages
Max pages per request100500
Tables✓ Yes✓ Yes
Handwriting✓ YesNo
Math✓ Yes✓ Yes
Bounding boxesNo✓ 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 GPT if

  • It is reported at about 5% errors on messy cursive, so it suits handwritten notes.
  • It extracts into custom JSON and can answer questions about a document, with files up to 50 MB per request.
  • There is no GPU or CUDA stack to run, and no regional licence limit on use.

Choose Hunyuan OCR if

  • It is free to self-host, so you pay only for compute, and it runs on laptops through llama.cpp GGUF.
  • It returns bounding boxes along with Markdown, HTML, LaTeX, JSON and plain text, which GPT's standard response does not.
  • One model of about 1B parameters covers parsing, text spotting, extraction and image translation, including low-resource and ancient scripts.

Key differences

GPTHunyuan OCR
How you run itHosted API from OpenAISelf-hosted via vLLM, Transformers or llama.cpp
LicenceProprietaryTencent Hunyuan Community License; excludes EU, UK and South Korea; separate licence above 100M monthly active users
PriceToken-based per model; page images add input tokens, so cost per page is not fixedFree software; you pay for compute
Output formatsMarkdown, JSON, plain text; no bounding boxes or confidence scores in the standard responseMarkdown, HTML, LaTeX, JSON, plain text, bounding boxes
Hardware and setupNone to manage; files up to 50 MB per requestNVIDIA GPU with CUDA 13 recommended for the unified vLLM environment; llama.cpp GGUF path for CPU, consumer GPU or laptop
BenchmarkGPT-5.2 scored 85.50 on OmniDocBench v1.5 and 86.59 on v1.6, versus 90.33 for Gemini 3 Pro in the same paperParseBench-2 overall 67.11, text score 62.88 in public CI replay (different benchmark, not directly comparable)
Known weaknessCost per page varies with page images; no word coordinatesUnified environment needs CUDA 13 and vLLM 0.25.1+; an archived 1.0 note said Transformers and vLLM outputs differed in accuracy

Sources

What each one is

GPT

General vision model. Good at messy scans and following extraction instructions.

GPT 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": "openai/gpt","fallbacks": ["tencent/hunyuan-ocr"],"file_url": "https://example.com/document.pdf"}

Questions

More comparisons

GPT alternativesHunyuan OCR alternativesBest OCR models in 2026