Surya vs SmolDocling
Pick Surya when you need layout detection, reading order and bounding boxes across 90+ languages, and you can run a vLLM or llama.cpp backend. Pick SmolDocling only for English page conversion experiments on tiny hardware. It is a preview release and its authors point users to Granite-Docling.
Suryadatalab/surya | SmolDoclinghuggingface/smoldocling | |
|---|---|---|
| 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 | 1.0 s / page |
| Best for | Layout and line-level OCR | Very cheap page conversion |
| Inputs | PDF, images | PDF, images |
| Languages | 90+ languages | Mainly 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
- It covers 90+ languages, while SmolDocling is English only.
- One model returns layout labels, reading order, OCR, HTML tables and KaTeX math with per-block bounding boxes.
- It scores 83.7% on olmOCR-bench and runs on CPU or Apple Silicon via llama.cpp, with peak memory of about 4 to 6 GB on an M4 Max.
Choose SmolDocling if
- At 256M parameters it is small enough for modest hardware, against 689M for Surya.
- It runs at about 0.35 seconds per page on an A100 with vLLM.
- It is Apache-2.0 with no revenue cap, while Surya's weights need a commercial license above $5M in funding or revenue.
Key differences
| Surya | SmolDocling | |
|---|---|---|
| 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 per the model summary (the metadata tag says CDLA-Permissive-2.0) |
| Model size | 689M parameters | 256M parameters |
| How you run it | Lightweight client against a separate vLLM or llama.cpp server; NVIDIA GPU, CPU or Apple Silicon | Self-hosted VLM; about 0.35 s per page on an A100 with vLLM |
| Languages | 90+ languages; only 21 of 91 reach a 90% pass rate | English only |
| Output formats | JSON, HTML, LaTeX, bounding boxes, plain text | DocTags, Markdown, HTML, JSON, LaTeX, bounding boxes |
| Reported benchmarks | olmOCR-bench 83.7%; 91-language multilingual pass rate 74.9% | Table TEDS 0.82 on FinTabNet; full-page OCR F1 0.80; OCRBench 338 |
| Known failure | Surya 2.5 table output lacks row and column geometry; multilingual tables score 58.1; handwriting and complex tables are better sent to the hosted Datalab API | Output can loop on some images and become very slow; endless token repetition and missing closing tags are typical defects |
Sources
- Surya GitHub README
- Datalab pricing and license FAQ
- SmolDocling model card
- Docling issue #1515
- Granite-Docling comparison table
What each one is
Surya
Open-source OCR toolkit with line detection, layout analysis, reading order and tables.
Surya detailsSmolDocling
Tiny 256M vision model from Hugging Face and IBM that outputs DocTags. Predecessor of Granite Docling.
SmolDocling detailsTry both on your own file
Run the file once with each model and compare the output.
PDF, PNG, JPG, WebP, HEIC. Up to 20 MB each, 10 files.No file?
Use one, fall back to the other
{"model": "datalab/surya","fallbacks": ["huggingface/smoldocling"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Surya vs MonkeyOCRSurya vs Granite Doclingdots.ocr vs SmolDoclingChandra vs SmolDoclingAmazon Textract vs SuryaSmolDocling vs DonutPaddleOCR-VL vs SmolDoclingSurya vs dots.ocr
Surya alternativesSmolDocling alternativesBest OCR models in 2026