Surya is an OCR and document intelligence tool that performs optical character recognition, layout analysis, reading order detection, and table recognition across 90+ languages.
The tool addresses the challenge of extracting and understanding text from scanned documents and images. It uses a 650-million-parameter neural network model trained to recognize text with high accuracy while simultaneously analyzing document structure. The model identifies layout elements such as tables, images, and headers, determines the logical reading order of content, and recognizes table structure including rows and columns. It handles multilingual documents, supporting over 90 languages in a single model.
Surya suits projects that need to process diverse document types at scale, particularly those requiring both text extraction and structural understanding. The tool achieves 83.3% accuracy on a standard OCR benchmark while maintaining fast throughput of 5 pages per second on high-end hardware. It also includes smaller companion models for specialized tasks like line-level text detection and OCR error detection. For teams preferring managed infrastructure rather than self-hosting, the project's creators offer a cloud platform with free trial credits and a public playground for testing.
The project maintains active development with regular model improvements and expanded language support. The codebase receives consistent updates addressing performance optimization and feature additions. Community engagement is supported through an official Discord channel for user discussion and feedback.
