Surya vs Nougat
Pick Surya for layout, reading order and bounding boxes across 90+ languages, and for local use on a laptop or in a commercial product under the licence terms. Pick Nougat only for English arXiv-style papers where you need LaTeX math in Markdown and your use is non-commercial. The main split is scope and licence: Surya is a general toolkit, Nougat is a narrow scientific-paper converter with non-commercial weights.
Suryadatalab/surya | Nougatmeta/nougat | |
|---|---|---|
| Type | Open source | Open source |
| Price | 2 credits / page | 2 credits / page |
| Per 1,000 pages | $2.00 | $2.00 |
| Typical speed | 1.5 s / page | 3.0 s / page |
| Best for | Layout and line-level OCR | arXiv-style papers |
| Inputs | PDF, images | PDF, images |
| Languages | 90+ languages | Mainly English |
| Max pages per request | 500 | 200 |
| Tables | ✓ Yes | ✓ Yes |
| Handwriting | No | No |
| Math | No | ✓ Yes |
| Bounding boxes | ✓ Yes | No |
| 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
- Its weights are free for research, personal use and companies under $5M funding or revenue, so small commercial products can use it.
- It returns layout labels, reading order, OCR, HTML tables, KaTeX math and per-block bounding boxes from one model, across 90+ languages.
- It runs on CPU and Apple Silicon through llama.cpp with peak memory of about 4-6 GB on an M4 Max, or on an NVIDIA GPU through vLLM.
Choose Nougat if
- It is trained on arXiv and PubMed Central papers and outputs equations and tables as LaTeX in Mathpix-compatible Markdown.
- It is a self-hosted Python package with small and base checkpoints and a simple CLI and API, with no separate inference server to run.
- The code is MIT licensed, which suits non-commercial research pipelines on English scientific PDFs.
Key differences
| Surya | Nougat | |
|---|---|---|
| Licence | Code Apache-2.0; weights modified AI Pubs Open Rail-M, free under $5M funding or revenue | Code MIT; weights CC-BY-NC, no commercial use |
| How you run it | Lightweight client against a vLLM or llama.cpp inference server; NVIDIA GPU, CPU or Apple Silicon | Self-hosted Python package with CLI and API |
| Output formats | JSON, HTML, LaTeX, bounding boxes, plain text | Markdown (.mmd) and LaTeX |
| Languages | 90+ languages; 74.9% pass rate on the 91-language test, 21 of 91 languages reach 90% | Works best on English; Chinese, Russian and Japanese do not work |
| Benchmarks | 83.7% on olmOCR-bench (Surya 2.5, author-reported) | Not published |
| Known failure | Surya 2.5 table output lacks row and column geometry; multilingual tables score 58.1 | Repetition loops on 1.5% of test pages, more on out-of-domain documents; pages can be dropped as MISSING_PAGE |
| Best document type | Layout detection and reading order; hard cases like handwriting and complex tables are better sent to the hosted Datalab API | English scientific papers and math-heavy PDFs; not for forms, invoices or business documents |
Sources
- Surya GitHub README
- Datalab pricing and license FAQ
- Nougat GitHub README
- Nougat paper
- Issue #11 repetitions
What each one is
Surya
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.
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": ["meta/nougat"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
Surya vs DonutSurya vs MonkeyOCRdots.ocr vs NougatChandra vs NougatDocling vs NougatClaude vs SuryaSurya vs PixtralSurya vs Granite Docling
Surya alternativesNougat alternativesBest OCR models in 2026