Qianfan OCR vs Pixtral
Pick Qianfan OCR for new document parsing work, especially multilingual pages, handwriting, tables and formulas that you host yourself. Pixtral 12B only makes sense if you already run it through the Mistral API and mainly ask questions about charts and documents. Mistral marks it as deprecated and no longer maintained, and it is reported to miss on dense, small-font and non-Latin text.
Qianfan OCRbaidu/qianfan-ocr | Pixtralmistral/pixtral | |
|---|---|---|
| Type | OCR API | Vision model |
| Price | 2 credits / page | 2 credits / page |
| Per 1,000 pages | $2.00 | $2.00 |
| Typical speed | 2.0 s / page | 3.0 s / page |
| Best for | Chinese documents | Questions about document images |
| Inputs | PDF, images | images |
| Languages | Chinese, English and more | Multilingual |
| Max pages per request | 200 | 500 |
| Tables | ✓ Yes | No |
| Handwriting | ✓ Yes | No |
| Math | ✓ Yes | No |
| Bounding boxes | No | No |
| 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 is built for document parsing and 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, and returns Markdown, JSON, HTML, LaTeX and bounding boxes.
- It is a 4B model under Apache-2.0, benchmarked on a single A100 with vLLM, so self-hosting is cheaper than running a 12B model.
Choose Pixtral if
- You already have an integration on the Mistral API and want to keep it, at a launch price of $0.15 per million input and output tokens.
- You need question answering over charts and documents, where Mistral reports 90.7 ANLS on DocVQA and 81.8 on ChartQA.
- You want many images in one request, since the 128K token context allows it.
Key differences
| Qianfan OCR | Pixtral | |
|---|---|---|
| Model type | 4B end-to-end document parsing model from Baidu (Qwen3-4B language model) | 12.4B general vision language model from Mistral AI (12B decoder + 400M vision encoder) |
| Maintenance status | Current, with weights on Hugging Face and hosted access on the Baidu Qianfan platform | Marked deprecated and no longer maintained by Mistral |
| Licence | Apache-2.0, commercial use allowed | Apache-2.0 open weights |
| Pricing | Free to self-host (you pay for compute); hosted access via Baidu Qianfan platform, price not published | Mistral API launch price $0.15 per million input and output tokens; free to self-host |
| Output formats | Markdown, JSON, HTML, LaTeX, bounding boxes | Plain text, Markdown, JSON |
| Headline benchmark | OmniDocBench v1.5 93.12, olmOCR-Bench 79.8, OCRBench 880 (reported by the authors) | DocVQA 90.7 ANLS, ChartQA 81.8, MMMU 52.5 (reported by Mistral) |
| Handwriting and non-Latin text | Covers 192 languages, handwriting and scene text | Reported accuracy drop on handwriting, small fonts and non-Latin scripts; Q4 quantization reported to degrade small text OCR |
Sources
- Qianfan-OCR model card
- Qianfan-OCR paper
- Mistral announcement
- Mistral Sept 2024 pricing
- Third-party deployment guide
What each one is
Qianfan OCR
Baidu's OCR-tuned vision model. Strong on Chinese documents and tables.
Qianfan OCR detailsPixtral
Mistral's open-weight vision models, Pixtral 12B and Pixtral Large. Mistral's API no longer serves them, so they run self-hosted.
Pixtral 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": ["mistral/pixtral"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Qianfan OCR alternativesPixtral alternativesBest OCR models in 2026