Surya

Surya is an open-source OCR and document AI toolkit covering 90+ languages with layout, reading order, and table recognition.

Overview

Surya is a family of fast, accurate document AI models from Datalab. Its toolkit includes OCR across 90+ languages, layout analysis, reading order detection, table recognition, and formula detection — all running locally. Benchmarks place it among the best open OCR systems, especially on multilingual and noisy documents. It is used standalone via Python or through Datalab’s hosted API, and pairs with Marker for full document conversion.

Surya’s models were trained with multilingual and noisy documents in mind, which makes it a go-to for archives, forms, and international paperwork. In benchmarks shared by Datalab, it competes with and often beats major cloud OCR services on layout-aware accuracy.

Key Features

• OCR supporting 90+ languages
• Layout analysis and reading order detection
• Table structure recognition and cell extraction
• Formula and math detection for technical documents
• Runs locally with open-source weights
• Hosted API available through Datalab

The model family keeps improving through public releases, and the benchmarks published by Datalab make comparisons easy. For teams handling forms, archives, and multilingual paperwork, Surya offers cloud-grade OCR with local execution.

Pricing

Surya’s models and code are free and open source. Datalab offers paid hosted processing for teams that prefer not to run models themselves.

Who Is It For

Developers building document pipelines for multilingual, technical, or scanned content who want top-tier OCR without vendor lock-in.

FAQ

Is Surya free to use?

Yes, the open-source toolkit is free; hosted processing is a paid convenience option.

Does Surya offer an API?

Yes, Datalab provides a hosted API alongside the local library.

How does Surya compare to alternatives?

Compared with Tesseract and cloud OCR APIs, Surya offers stronger multilingual accuracy and full layout analysis in one open toolkit.

Can I use Surya commercially?

Yes, the open-source license permits commercial use; verify the specific license terms for your deployment.

What hardware does it need?

Surya runs on CPU or GPU, with GPU recommended for batch workloads.

What output does Surya produce?

Text with bounding boxes, layout structure, reading order, and table data in structured formats.

Is there a hosted option?

Yes, Datalab offers hosted API processing for teams that prefer not to run models locally.

Is Surya better than Tesseract?

On multilingual and layout-heavy documents, Surya generally delivers higher accuracy.

What makes Surya different from cloud OCR?

Surya runs locally with open weights, combining multilingual OCR, layout analysis, and table recognition without sending documents anywhere.