Qianfan OCR vs GOT-OCR 2.0
Pick Qianfan OCR for self-hosted, multilingual document parsing where you need structured output (Markdown, JSON, HTML, LaTeX, bounding boxes) and a published benchmark score. Pick GOT-OCR 2.0 for a small 580M model that reads formulas, sheet music and chemical structures, or for OCR on a chosen region. Mind the licence split: Qianfan OCR is Apache-2.0, while GOT-OCR 2.0 training data is CC BY-NC 4.0.
Qianfan OCRbaidu/qianfan-ocr | GOT-OCR 2.0stepfun/got-ocr-2 | |
|---|---|---|
| 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 | Formatted text, formulas and charts |
| Inputs | PDF, images | images |
| Languages | Chinese, English and more | Chinese, English |
| Max pages per request | 200 | 100 |
| 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
- It scores 93.12 on OmniDocBench v1.5, 79.8 on olmOCR-Bench and 880 on OCRBench, all reported by its authors.
- It covers 192 languages, handwriting and scene text, so it suits multilingual document parsing.
- It outputs JSON, HTML and bounding boxes as well as Markdown and LaTeX, which helps with key information extraction and tables.
Choose GOT-OCR 2.0 if
- At 580M parameters it is much smaller than the 4B Qianfan OCR.
- It reads formulas, sheet music and chemical structures, with output formats such as TikZ, SMILES and kern.
- Region OCR guided by box coordinates or color lets you read one part of a page, and it supports multi-page OCR.
Key differences
| Qianfan OCR | GOT-OCR 2.0 | |
|---|---|---|
| Size | 4B parameters (Qwen3-4B language model) | 580M parameters |
| Licence | Apache-2.0 | Code Apache-2.0; training data CC BY-NC 4.0 |
| How you run it | Self-host weights from Hugging Face, or use the Baidu Qianfan platform | Self-host only, via Hugging Face Transformers; CUDA 11.8 and PyTorch 2.0.1 per README, CPU users are pointed to a llama.cpp port |
| Output formats | Markdown, JSON, HTML, LaTeX, bounding boxes | Plain text, Markdown, LaTeX, TikZ, SMILES, kern |
| Benchmarks | OmniDocBench v1.5 93.12, olmOCR-Bench 79.8, OCRBench 880 (authors) | Not published |
| Languages | 192 languages, plus handwriting and scene text | English and Chinese documents |
| Known failure | Thinking mode can emit up to 16,384 tokens, adding latency; 1.024 pages per second on an A100 with W8A8 | Output can loop characters midway (issue #226), invents text on blank images (issue #235), and format mode is much worse than plain OCR mode (issue #317) |
Sources
What each one is
Qianfan OCR
Baidu's OCR-tuned vision model. Strong on Chinese documents and tables.
Qianfan OCR detailsGOT-OCR 2.0
Unified OCR model that reads plain text, tables, formulas, sheet music and charts.
GOT-OCR 2.0 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": ["stepfun/got-ocr-2"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Qianfan OCR alternativesGOT-OCR 2.0 alternativesBest OCR models in 2026