Jina OCR vs Surya
Pick Jina OCR for non-commercial page-to-Markdown work, research and evaluation, where its 91.14 on OmniDocBench v1.6 and a hosted API option help. Pick Surya when you need layout, reading order and per-block bounding boxes from a small model you run yourself, including on a laptop CPU. The main split is licence and output: Jina OCR blocks commercial use without a deal, while Surya is free for companies under $5M funding or revenue.
Jina OCRjina/jina-ocr | Suryadatalab/surya | |
|---|---|---|
| Type | Open source | Open source |
| Price | 1 credit / page | 2 credits / page |
| Per 1,000 pages | $1.00 | $2.00 |
| Typical speed | 1.5 s / page | 1.5 s / page |
| Best for | Clean Markdown at low cost | Layout and line-level OCR |
| Inputs | PDF, images | PDF, images |
| Languages | About 100 languages | 90+ languages |
| Max pages per request | 200 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | No |
| Math | ✓ Yes | No |
| 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 Jina OCR if
- You want a page-to-Markdown model that also outputs HTML and LaTeX, and it scores 91.14 on OmniDocBench v1.6 (authors' figure).
- You want a hosted API as well as self-hosting, and you can retry after 30 to 60 seconds when a cold start returns HTTP 503.
- Your work is non-commercial or research, and you want only about 570M active parameters at 2.57 pages per second on an A100.
Choose Surya if
- You need layout labels, reading order, HTML tables, KaTeX math and per-block bounding boxes from one model, across 90+ languages.
- You want to run locally on CPU or Apple Silicon via llama.cpp, with peak memory of about 4.3 to 5.9 GB on an M4 Max.
- You are a company under $5M in funding or revenue and want to use it commercially for free, and you can run a vLLM or llama.cpp backend yourself.
Key differences
| Jina OCR | Surya | |
|---|---|---|
| Licence | CC-BY-NC-4.0; commercial use needs a licence from Jina/Elastic | Code Apache-2.0; weights under modified AI Pubs Open Rail-M, free for research, personal use and companies under $5M funding or revenue |
| How you run it | Self-host or call Jina's hosted API | Self-host only, as a lightweight client against a vLLM or llama.cpp server |
| Model size | 3B MoE, about 570M active | 689M parameters |
| Output formats | Markdown, HTML, LaTeX | JSON, HTML, LaTeX, bounding boxes, plain text |
| olmOCR-Bench | 83.4 overall (authors), using a stricter prompt than the recommended default | 83.7% for Surya 2.5 (authors) |
| Hardware | GPU recommended (A100, L4); falls back to CPU in Transformers | NVIDIA GPU via vLLM, or CPU and Apple Silicon via llama.cpp; only 0.198 pages/s on an M4 Max at 96 DPI with Q8_0 |
| Known failure | Hosted API returns HTTP 503 on cold start | Surya 2.5 tables lack row/column geometry, multilingual tables score 58.1, and the README sends handwriting and complex tables to the hosted Datalab API |
Sources
What each one is
Jina OCR
Compact page-to-Markdown model from Jina AI, built on the DeepSeek-OCR encoder.
Jina OCR detailsSurya
Open-source OCR toolkit with line detection, layout analysis, reading order and tables.
Surya 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": ["datalab/surya"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Jina OCR alternativesSurya alternativesBest OCR models in 2026