Qianfan OCR vs Granite Docling
Pick Qianfan OCR for self-hosted multilingual parsing of scans, handwriting and forms, where it scores 93.12 on OmniDocBench v1.5 and covers 192 languages. Pick Granite Docling for English PDFs with tables, code and equations inside a Docling pipeline, where its 258M size runs on a CPU or small GPU. The main split is size and language coverage: 4B and 192 languages versus 258M and English first.
Qianfan OCRbaidu/qianfan-ocr | Granite Doclingibm/granite-docling | |
|---|---|---|
| Type | OCR API | Open source |
| Price | 2 credits / page | 1 credit / page |
| Per 1,000 pages | $2.00 | $1.00 |
| Typical speed | 2.0 s / page | 1.5 s / page |
| Best for | Chinese documents | Small-model document conversion |
| Inputs | PDF, images | PDF, images |
| Languages | Chinese, English and more | English, experimental CJK and Arabic |
| Max pages per request | 200 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | ✓ Yes | No |
| Math | ✓ Yes | ✓ Yes |
| Bounding boxes | No | ✓ 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 Qianfan OCR if
- You need 192 languages, handwriting and scene text; Granite Docling marks Japanese, Arabic and Chinese as experimental.
- You want key information extraction plus Markdown, JSON, HTML, LaTeX and bounding boxes from one model.
- You want a hosted option: it is also offered on the Baidu Qianfan platform, and the weights are Apache-2.0.
Choose Granite Docling if
- You convert English PDFs with tables, code and equations: table structure TEDS is 0.97 on FinTabNet, code F1 is 0.988 and equation F1 is 0.968.
- You run on CPU or a small GPU, since the model has only 258M parameters.
- You already use the Docling library, because the model outputs DocTags built for it.
Key differences
| Qianfan OCR | Granite Docling | |
|---|---|---|
| Size | 4B (Qwen3-4B language model) | 258M |
| Licence | Apache-2.0 | Apache-2.0 |
| How you run it | Self-host (benchmarked on one A100 with vLLM) or the Baidu Qianfan platform | Self-host only, typically inside Docling |
| Output formats | Markdown, JSON, HTML, LaTeX, bounding boxes | DocTags, Markdown, HTML, LaTeX, bounding boxes |
| Languages | 192 languages, plus handwriting and scene text | English primary; Japanese, Arabic and Chinese experimental |
| OCRBench | 880 | 500 |
| Known failure | Thinking mode can emit up to 16,384 tokens, adding latency; 1.024 pages per second on an A100 with W8A8 | Repetitive loops on page headings and invoices; exclamation-mark output on GPUs without bfloat16 such as T4 unless run in float32 |
Sources
- Qianfan-OCR model card
- Qianfan-OCR paper
- Granite-Docling model card
- IBM announcement
- HF discussion #24 repetition
What each one is
Qianfan OCR
Baidu's OCR-tuned vision model. Strong on Chinese documents and tables.
Qianfan OCR detailsGranite Docling
Compact IBM vision model that converts pages to DocTags for the Docling pipeline.
Granite Docling 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": "baidu/qianfan-ocr","fallbacks": ["ibm/granite-docling"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Qianfan OCR alternativesGranite Docling alternativesBest OCR models in 2026