PaddleOCR, the open-source optical character recognition framework, has released version 3.5 with significant architectural improvements centered on leveraging transformer-based models for enhanced performance. The update enables developers to run OCR and document parsing tasks using a transformers backend, bringing state-of-the-art natural language processing capabilities to traditional document recognition workflows. This integration allows for more sophisticated text extraction and understanding from images and scanned documents. The release represents a convergence of computer vision and NLP technologies, enabling more accurate and context-aware document processing. By adopting transformer architectures alongside traditional OCR approaches, PaddleOCR 3.5 addresses limitations in character recognition accuracy and document layout understanding. The update is particularly relevant for enterprises handling large volumes of documents that require automated parsing and data extraction, from invoices and contracts to forms and receipts.