GOT-OCR v2

A cutting-edge, end-to-end General OCR Theory model engineered to unify all optical character recognition tasks into a single vision-language network. Moving past traditional, rigid multi-stage text scrapers, this architecture processes raw pixels to instantly extract standard multilingual text, complex mathematical formulas, chemical equations, musical scores, and geometric charts. It preserves original layout coordinates natively, rendering outputs directly into clean Markdown or structured JSON formats with pinpoint accuracy. The model handles low-contrast scans, curved book pages, and complex multi-column paper formats effortlessly without specialized pre-processing. It stands as a vital open-weights foundation for automated enterprise document ingestion, legal archiving, and advanced scientific data parsing.