Hunyuan OCR vs Florence-2
Pick Hunyuan OCR for photos, scene text, image translation, low-resource scripts and structured output such as Markdown, HTML, LaTeX or JSON, if you are outside the EU, UK and South Korea. Pick Florence-2 for small, MIT-licensed OCR of English scene text and short labels with line boxes, especially when the same model must also do captioning or detection. Avoid Florence-2 for full pages, CJK text and handwriting.
Hunyuan OCRtencent/hunyuan-ocr | Florence-2microsoft/florence-2 | |
|---|---|---|
| Type | Open source | Open source |
| Price | 2 credits / page | 1 credit / page |
| Per 1,000 pages | $2.00 | $1.00 |
| Typical speed | 1.5 s / page | 0.8 s / page |
| Best for | Chinese documents and photos | Lightweight OCR with boxes |
| Inputs | PDF, images | images |
| Languages | 100+ languages | English |
| Max pages per request | 500 | 500 |
| Tables | ✓ Yes | No |
| Handwriting | No | No |
| Math | ✓ Yes | No |
| Bounding boxes | ✓ Yes | ✓ Yes |
| Confidence | No | No |
| Forms | No | No |
| Multilingual | ✓ Yes | No |
| On OCRRouter | Available | Available |
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
- You need structured output: Markdown, HTML, LaTeX, JSON, plain text or bounding boxes.
- You work with Chinese or other non-Latin text, low-resource scripts or ancient scripts, or need image translation.
- You want one ~1B model for parsing, text spotting, extraction and translation, with images up to 4K and a 128K context window in v1.5.
Choose Florence-2 if
- You need a permissive MIT licence with no regional exclusions or user-count limits.
- You want the smallest footprint: 0.23B base or 0.77B large, run through Transformers.
- You need line-level quadrilateral boxes for scene text and short labels, or the same model must also caption, detect and ground objects.
Key differences
| Hunyuan OCR | Florence-2 | |
|---|---|---|
| Licence | Tencent Hunyuan Community License. Excludes the EU, UK and South Korea. Separate licence needed above 100M monthly active users. | MIT |
| Model size | ~1B parameters | 0.23B (base), 0.77B (large) |
| How you run it | Self-hosted via vLLM, Transformers or llama.cpp. CUDA 13 recommended for the unified vLLM environment. llama.cpp GGUF path for CPU, consumer GPU or laptop. | Self-hosted via Transformers |
| Output formats | Markdown, HTML, LaTeX, JSON, plain text, bounding boxes | Plain text and bounding boxes. No Markdown, tables or LaTeX. |
| Scope | Document parsing, text spotting, information extraction and image translation in one model | OCR and OCR with regions, plus captioning, detection and grounding |
| Chinese and CJK text | Not published | Chinese output mostly wrong. The vocabulary lacks CJK characters. |
| Known failure | ParseBench-2 overall is 67.11. Transformers and vLLM outputs differed in accuracy in the archived 1.0 note. | Spelling errors and missing spaces on cursive handwriting and French text in a Hugging Face blog test. |
Sources
- HunyuanOCR model card
- HunyuanOCR GitHub
- Hunyuan Community License
- Florence-2-large model card
- HF blog OCR test
- HF discussion Chinese very bad
What each one is
Hunyuan OCR
Small end-to-end OCR model from Tencent covering parsing, extraction and photo translation.
Hunyuan OCR detailsFlorence-2
Small Microsoft vision model (0.23B and 0.77B) with OCR and OCR-with-regions tasks that return text with boxes.
Florence-2 detailsTry both on your own file
Run the file once with each model and compare the output.
Use one, fall back to the other
{"model": "tencent/hunyuan-ocr","fallbacks": ["microsoft/florence-2"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Hunyuan OCR alternativesFlorence-2 alternativesBest OCR models in 2026