Surya vs Granite Docling
Pick Surya if you need layout detection, reading order and block-level bounding boxes across 90+ languages, and you can live with a licence limit above $5M in revenue or funding. Pick Granite Docling if you convert English PDFs with tables, code and equations inside Docling and want plain Apache-2.0 on a 258M model. The main split is scope: Surya is a multilingual layout and OCR toolkit, Granite Docling is a small English-first document converter.
Suryadatalab/surya | Granite Doclingibm/granite-docling | |
|---|---|---|
| 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 | 1.5 s / page |
| Best for | Layout and line-level OCR | Small-model document conversion |
| Inputs | PDF, images | PDF, images |
| Languages | 90+ languages | English, experimental CJK and Arabic |
| 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
- It covers 90+ languages with layout labels, reading order, OCR, HTML tables and KaTeX math, plus per-block bounding boxes, in one model.
- It runs on CPU and Apple Silicon through llama.cpp, with peak memory of about 4.3-5.9 GB on an M4 Max.
- It reaches 83.7% on olmOCR-bench and 4.52 pages/s at 192 DPI on one H100 with vLLM.
Choose Granite Docling if
- It is Apache-2.0 with no revenue or funding threshold, and only 258M parameters.
- It scores 0.97 table structure TEDS on FinTabNet, 0.968 equation F1 and 0.988 code recognition F1.
- It outputs DocTags for the Docling library, plus Markdown, HTML, LaTeX and bounding boxes, so it fits a Docling pipeline directly.
Key differences
| Surya | Granite Docling | |
|---|---|---|
| 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 |
| Size | 689M parameters | 258M parameters |
| How you run it | Local client against a separate vLLM or llama.cpp inference server. NVIDIA GPU via vLLM, or CPU and Apple Silicon via llama.cpp. | Self-hosted, typically inside Docling. |
| Output formats | JSON, HTML, LaTeX, bounding boxes, plain text | DocTags, Markdown, HTML, LaTeX, bounding boxes |
| Languages | 90+ languages; 74.9% pass rate on a 91-language test, with only 21 languages reaching 90%. | English primary; Japanese, Arabic and Chinese are marked experimental. |
| Tables | Surya 2.5 table output lacks row and column geometry, and simple table mode errors without the older Surya 2 checkpoint. Multilingual tables score 58.1. | 0.97 TEDS structure on FinTabNet at 150 dpi |
| Known failure | Hard documents such as handwriting, complex tables and math are better sent to the hosted Datalab API. Not for photos or natural scenes. | Gets stuck in repetitive loops on page headings and invoices. On GPUs without bfloat16, such as T4, it may output only exclamation marks unless run in float32. |
Sources
- Surya GitHub README
- Datalab pricing and license FAQ
- Granite-Docling model card
- HF discussion #24 repetition
What each one is
Surya
Open-source OCR toolkit with line detection, layout analysis, reading order and tables.
Surya detailsGranite Docling
Compact IBM vision model that converts pages to DocTags for the Docling pipeline.
Granite Docling 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": ["ibm/granite-docling"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Surya alternativesGranite Docling alternativesBest OCR models in 2026