Jina OCR vs Pixtral
Pick Jina OCR for non-commercial PDF to Markdown, research and evaluation, where its OCR benchmarks and small active parameter count help. Pick Pixtral only if you already run on the Mistral API or need Apache-2.0 weights, and note that Mistral marks it deprecated, so it is a poor choice for new OCR work. The main split is that Jina OCR is a purpose-built page-to-Markdown model with a non-commercial licence, while Pixtral is a general vision model with a permissive licence and weaker verbatim OCR.
Jina OCRjina/jina-ocr | Pixtralmistral/pixtral | |
|---|---|---|
| Type | Open source | Vision model |
| Price | 1 credit / page | 2 credits / page |
| Per 1,000 pages | $1.00 | $2.00 |
| Typical speed | 1.5 s / page | 3.0 s / page |
| Best for | Clean Markdown at low cost | Questions about document images |
| Inputs | PDF, images | images |
| Languages | About 100 languages | Multilingual |
| Max pages per request | 200 | 500 |
| Tables | ✓ Yes | No |
| Handwriting | No | 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 Jina OCR if
- You want page-to-Markdown output, with HTML and LaTeX also available, for non-commercial PDF conversion.
- You want a model with OCR benchmark results: 91.14 on OmniDocBench v1.6 and 83.4 on olmOCR-Bench, both reported by the authors.
- You need throughput on modest hardware, since only about 570M of its 3B parameters are active and it runs 2.57 pages per second on an A100.
Choose Pixtral if
- You need Apache-2.0 weights that can be used commercially without a separate licence deal.
- You already have an integration on the Mistral API, where launch pricing was $0.15 per million input and output tokens.
- Your task is question answering over charts and documents, where it scores 90.7 ANLS on DocVQA and 81.8 on ChartQA (Mistral's numbers), and the 128K context fits many images per request.
Key differences
| Jina OCR | Pixtral | |
|---|---|---|
| Licence | CC-BY-NC-4.0. Commercial use needs a license from Jina/Elastic. | Apache-2.0 |
| Model type | Page-to-Markdown model built on the DeepSeek-OCR architecture | General vision language model with a 12B Mistral Nemo decoder and a 400M vision encoder |
| Size | 3B MoE, about 570M active | 12.4B |
| How you run it | Self-host or Jina's hosted API. GPU recommended, vLLM 0.21+ for FastMTP, CPU fallback in Transformers. | Open weights or the Mistral API as pixtral-12b-2409 |
| Price | Free to self-host for non-commercial use, you pay for compute | $0.15 per million input and output tokens at API launch. Free to self-host, you pay for compute. |
| Output formats | Markdown, HTML, LaTeX | Plain text, Markdown, JSON |
| Benchmarks | OmniDocBench v1.6 91.14, olmOCR-Bench 83.4 | DocVQA 90.7 ANLS, ChartQA 81.8, MMMU 52.5 |
| Known failures | Hosted API returns HTTP 503 on cold start, so retry after 30 to 60 seconds. Reported scores use a stricter prompt than the recommended default. | Deprecated and no longer maintained. Reported accuracy drop on handwriting, small fonts and non-Latin scripts. Q4 quantization degrades small text OCR. |
Sources
- jina-ocr-v1 model card
- Jina model page
- Mistral announcement
- Mistral Sept 2024 pricing
- Third-party deployment guide
What each one is
Jina OCR
Compact page-to-Markdown model from Jina AI, built on the DeepSeek-OCR encoder.
Jina 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": "jina/jina-ocr","fallbacks": ["mistral/pixtral"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Jina OCR alternativesPixtral alternativesBest OCR models in 2026