97 milliseconds! Baidu PP-OCRv6 brings OCR into the millisecond era
Baidu's Wenxin officially released the new OCR model PP-OCRv6, offering Tiny, Small, and Medium versions that support over 50 languages and are deployable across various scenarios from browsers to servers. The Tiny model, weighing just 1.5MB, can perform OCR in as little as 97 milliseconds directly within a browser, enhancing privacy and reducing deployment barriers. PP-OCRv6 has set new benchmarks in OCR performance, outperforming major multimodal models in specialized OCR tasks and solidifying PaddleOCR's position as a leading open-source OCR project. AI
IMPACT Sets new SOTA for browser-based OCR, potentially accelerating AI agent capabilities in edge and privacy-sensitive applications.