Qianfan OCR vs dots.ocr
Pick Qianfan OCR for self-hosted multilingual parsing, key information extraction, and documents heavy on tables and formulas. It reports 93.12 on OmniDocBench v1.5, covers 192 languages and handwriting, and has an Apache-2.0 licence. Pick dots.ocr for multilingual PDFs to Markdown with layout, and for chart-to-SVG work through the dots.mocr-svg variant. It is smaller at 3B and its v1.5 scores 83.9 on olmOCR-bench.
Qianfan OCRbaidu/qianfan-ocr | dots.ocrrednote/dots-ocr | |
|---|---|---|
| Type | OCR API | Open source |
| Price | 2 credits / page | 2 credits / page |
| Per 1,000 pages | $2.00 | $2.00 |
| Typical speed | 2.0 s / page | 1.5 s / page |
| Best for | Chinese documents | Multilingual document parsing |
| Inputs | PDF, images | PDF, images |
| Languages | Chinese, English and more | 100+ languages |
| 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 | ✓ 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.
- It covers 192 languages, handwriting and scene text.
- It is also offered as a hosted service on the Baidu Qianfan platform, so you can skip running a GPU yourself.
Choose dots.ocr if
- It is smaller at 3B parameters (1.7B LLM backbone), against 4B for Qianfan OCR.
- The dots.mocr-svg variant converts charts and diagrams to SVG, which Qianfan OCR does not list.
- Its v1.5 (dots.mocr) reaches 83.9 on olmOCR-bench, up from 79.1 for the original, and 91.85 on DocVQA per the authors.
Key differences
| Qianfan OCR | dots.ocr | |
|---|---|---|
| Licence | Apache-2.0 | MIT, plus a separate dots.ocr license agreement file in the repository |
| How you run it | Self-host the open weights, or use the hosted Baidu Qianfan platform | Self-host only, via vLLM or Hugging Face Transformers |
| Model size | 4B (Qwen3-4B language model) | 3B (1.7B LLM backbone) |
| Output formats | Markdown, JSON, HTML, LaTeX, bounding boxes | JSON, Markdown, HTML, LaTeX, bounding boxes, SVG |
| olmOCR-bench | 79.8 | 79.1 for dots.ocr, 83.9 for dots.mocr (v1.5) |
| Hardware | Benchmarked on one NVIDIA A100 with vLLM; no minimum VRAM stated. Not for CPU-only deployment. | Single CUDA GPU with vLLM is the documented path; no VRAM figure stated. Not suited to CPU-only production. |
| Known failure | Thinking mode can emit up to 16,384 tokens, adding latency. 1.024 pages per second on an A100 even with W8A8. | Runs of ellipses or underscores can make output repeat endlessly. A table above 1000 grid cells raises an error and drops the whole page (Docling issue). |
Sources
- Qianfan-OCR model card
- Qianfan-OCR paper
- dots.ocr GitHub README
- dots.ocr blog / limitations
- Docling issue #5146
What each one is
Qianfan OCR
Baidu's OCR-tuned vision model. Strong on Chinese documents and tables.
Qianfan OCR detailsdots.ocr
Single vision model from rednote that does layout detection and content recognition together.
dots.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": ["rednote/dots-ocr"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Qianfan OCR alternativesdots.ocr alternativesBest OCR models in 2026