PaddlePaddle PaddleOCR-VL - 飞桨AI Studio Introduction PaddleOCR-VL is a SOTA and resource-efficient model tailored for document parsing Its core component is PaddleOCR-VL-0 9B, a compact yet powerful vision-language model (VLM) that integrates a NaViT-style dynamic resolution visual encoder with the ERNIE-4 5-0 3B language model to enable accurate element recognition This innovative model efficiently supports 109 languages and