Qianfan OCR vs PaddleOCR (PP-OCR)
Pick Qianfan OCR when you need a full document turned into Markdown, JSON, HTML or LaTeX, with tables, formulas and handwriting, and you have an A100-class GPU. Pick PaddleOCR (PP-OCR) when you need fast text lines and boxes from Chinese or English print, receipts, signs or screenshots, on CPU, edge devices or in the browser. The main split is a 4B document parser against a 34.5M text detector and recognizer.
Qianfan OCRbaidu/qianfan-ocr | PaddleOCR (PP-OCR)paddle/paddleocr | |
|---|---|---|
| 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 | 0.7 s / page |
| Best for | Chinese documents | Fast printed text, Chinese and English |
| Inputs | PDF, images | PDF, images |
| Languages | Chinese, English and more | 80+ languages |
| Max pages per request | 200 | 500 |
| Tables | ✓ Yes | No |
| Handwriting | ✓ Yes | No |
| Math | ✓ Yes | No |
| Bounding boxes | No | ✓ Yes |
| Confidence | No | ✓ Yes |
| Forms | No | No |
| Multilingual | ✓ Yes | ✓ Yes |
| 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, the top end-to-end score in Baidu's table, and 79.8 on olmOCR-Bench.
- It outputs Markdown, JSON, HTML, LaTeX and bounding boxes, so tables and formulas come out in one pass.
- It covers 192 languages, handwriting and scene text, and suits key information extraction.
Choose PaddleOCR (PP-OCR) if
- It runs on CPU, NVIDIA GPU and Kunlunxin XPU, and PP-OCRv6 reports 0.13s on an A100 and a 5.2x CPU speedup with OpenVINO.
- PP-OCRv6 medium is 34.5M parameters, and one model covers 50 languages including Chinese, English, Japanese and 46 Latin-script languages.
- It can run in the browser through the official PaddleOCR.js SDK, and also from Python or C++.
Key differences
| Qianfan OCR | PaddleOCR (PP-OCR) | |
|---|---|---|
| Licence | Apache-2.0, commercial use allowed | Apache-2.0 |
| How you run it | Open weights on Hugging Face, or the hosted Baidu Qianfan platform | Local library: paddleocr Python package, C++ or PaddleOCR.js in the browser |
| Model size | 4B (Qwen3-4B language model) | PP-OCRv6 medium: 34.5M |
| Hardware | Benchmarked on one NVIDIA A100 with vLLM; not for CPU-only deployment | CPU, NVIDIA GPU, Kunlunxin XPU and other accelerators |
| Output formats | Markdown, JSON, HTML, LaTeX, bounding boxes | Plain text, bounding boxes, JSON; tables and formulas need the separate PP-Structure or VL pipelines |
| Benchmarks | OmniDocBench v1.5 93.12, olmOCR-Bench 79.8, OCRBench 880 (author-reported) | Not published |
| Known issues | 1.024 pages per second on an A100 even with W8A8; thinking mode can emit up to 16,384 tokens, adding latency | PP-OCRv6 weights contain denormals that make CPU inference 4 to 11x slower (RapidOCR issue #743); NotImplementedError on PaddleOCR 3.7.0 with PaddlePaddle 3.3.1 (issue #18162) |
Sources
- Qianfan-OCR model card
- Qianfan-OCR paper
- PaddleOCR GitHub README
- PP-OCRv6 on Hugging Face
- RapidOCR issue #743 denormal weights
- Issue #18162
What each one is
Qianfan OCR
Baidu's OCR-tuned vision model. Strong on Chinese documents and tables.
Qianfan OCR detailsPaddleOCR (PP-OCR)
Baidu's classic open-source text detection and recognition pipeline. Light enough to run on CPU.
PaddleOCR (PP-OCR) 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": ["paddle/paddleocr"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Qianfan OCR alternativesPaddleOCR (PP-OCR) alternativesBest OCR models in 2026