Surya OCR

Surya is a 650-million-parameter OCR model from Datalab (founded by Vik Paruchuri) for multilingual document understanding: line-level text detection, OCR, layout analysis with reading order, and table recognition across more than 90 languages. It scores 83.3 percent on the olmOCR benchmark, the top result under 3B parameters, while running at about 5 pages per second on an RTX 5090 and working on CPU and Apple Silicon too.

The code is Apache 2.0 but the model weights use a modified open-rail license: free for research, personal use, and startups under 5M in funding or revenue, with larger commercial use requiring a Datalab license. Surya OCR 2 continues the line with stronger handwriting, math, and form support. For document-ingestion pipelines feeding RAG systems, it is one of the most practical self-hostable options available.