Baidu Releases PP-OCRv6 with 50-Language Support, 10M-Level Parameters Match Billion-Scale VLMs

Baidu's PaddlePaddle team recently released PP-OCRv6, a new OCR system offering three versions: Tiny (1.5M parameters), Small (7.7M), and Medium (34.5M). The Medium model delivers 4.6% improvement in detection accuracy and 5.1% in recognition accuracy compared to PP-OCRv5, while integrating Chinese, English, Japanese, and 46 Latin-script languages into a single unified model.

The system employs structural reparameterization techniques to reduce computational overhead while boosting accuracy. Under OpenVINO optimization, the Medium version achieves up to 5.2x faster CPU inference speed. According to official benchmarks, PP-OCRv6 matches or exceeds performance of some billion-parameter vision-language models despite using only millions of parameters. The code has been integrated into the open-source PaddleOCR project.