Surya vs OpenOCR
Pick Surya for local OCR with layout detection, reading order and bounding boxes across 90+ languages, including on a laptop with CPU or Apple Silicon. Pick OpenOCR for Chinese and English academic PDFs with formulas, where its 0.1B models are light and score 90.57 on OmniDocBench v1.5. The main split is language coverage and output structure versus a smaller model limited to two languages.
Suryadatalab/surya | OpenOCRfudan/openocr | |
|---|---|---|
| Type | Open source | Open source |
| Price | 2 credits / page | 1 credit / page |
| Per 1,000 pages | $2.00 | $1.00 |
| Typical speed | 1.5 s / page | 0.5 s / page |
| Best for | Layout and line-level OCR | Lightweight OCR pipeline on CPU |
| Inputs | PDF, images | PDF, images |
| Languages | 90+ languages | Chinese and English |
| Max pages per request | 500 | 500 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | No |
| Math | No | ✓ Yes |
| Bounding boxes | ✓ Yes | ✓ Yes |
| Confidence | No | No |
| Forms | No | No |
| Multilingual | ✓ Yes | No |
| 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 Surya if
- You need OCR in more than two languages, since Surya covers 90+ languages and 21 of 91 tested languages reach a 90% pass rate.
- You need layout labels, reading order, HTML tables, KaTeX math and per-block bounding boxes from one model.
- You want to run on a laptop, because it works on CPU and Apple Silicon through llama.cpp with peak RAM of about 4-6 GB on an M4 Max.
Choose OpenOCR if
- Your documents are Chinese or English academic PDFs with formulas, and UniRec-0.1B reads plain text and single or multi-line formulas in one model.
- You want the smallest model, since OpenDoc-0.1B reports 90.57 on OmniDocBench v1.5 with 0.1B parameters.
- You want a plain Apache-2.0 licence with no revenue cap, plus ONNX export for deployment.
Key differences
| Surya | OpenOCR | |
|---|---|---|
| Licence | Code Apache-2.0; weights under a modified AI Pubs Open Rail-M. Free for research, personal use and companies under $5M funding or revenue. | Apache-2.0 |
| Languages | 90+ languages. Multilingual pass rate is 74.9% across 91 languages. | Chinese and English only for OpenDoc and UniRec. |
| Model size | 689M parameters | 0.1B parameters (OpenDoc-0.1B and UniRec-0.1B) |
| Benchmark | 83.7% on olmOCR-bench and 85.9 on the OmniParseBench headline (Surya 2.5), reported by the authors. | 90.57 on OmniDocBench v1.5 (OpenDoc-0.1B), reported by the authors. Different benchmarks, so the scores are not directly comparable. |
| How you run it | A lightweight client against a separate vLLM or llama.cpp inference server. Needs an NVIDIA GPU via vLLM, or CPU/Apple Silicon via llama.cpp. | A self-hosted Python library. Document parsing needs a separate layout model, PP-DocLayoutV2 from PaddlePaddle. No hardware requirements published. |
| Output formats | JSON, HTML, LaTeX, bounding boxes, plain text | Plain text, LaTeX, bounding boxes |
| Known failure | Surya 2.5 table output lacks row/column geometry, and multilingual tables score only 58.1. The README sends handwriting, complex tables and math to the hosted Datalab API. Not for photos or natural scenes. | No independent replication of its benchmarks found. No throughput figures published. |
Sources
What each one is
Surya
Open-source OCR toolkit with line detection, layout analysis, reading order and tables.
Surya detailsOpenOCR
Fudan's open toolkit for text detection and recognition, formulas, tables and document parsing.
OpenOCR 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": "datalab/surya","fallbacks": ["fudan/openocr"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Surya alternativesOpenOCR alternativesBest OCR models in 2026