Vietnam’s AI model ranks global top 12 for document recognition
CATI-VLM, an AI model developed in Vietnam, ranks among the world’s best in visual document understanding.
THE HANOI TIMES — CATI-VLM, an AI model developed by the CMC Applied Technology Institute (CMC ATI), was named one of the world’s top 12 models and ranked first in Vietnam in the document visual question answering (DocVQA) category of the June 2025 robust reading competition (RRC).
Artificial intelligence is emerging as a dynamic and fast-growing sector in Vietnam’s digital landscape.
Built on a 5TB dataset, CATI-VLM is designed to tackle challenges specific to Vietnamese documents, particularly those involving diacritics and handwritten text. Unlike standard OCR, the model understands not only text, but also visual elements such as tables, checkboxes, signatures, and page layouts. Like ChatGPT, it can answer natural-language questions directly on document images without needing to be trained on specific forms.
Hosted by the Computer Vision Center at the Autonomous University of Barcelona, RRC is a global benchmark in document AI. Since 2011, it has been held alongside the International Conference on Document Analysis and Recognition (ICDAR) and attracts entries from top universities and technology firms worldwide, including Tsinghua University, Hyundai Motor Group, and Tencent.
"This result proves that Vietnamese researchers can compete globally while solving real-world problems tailored to our language and industries," said Dr. Dang Minh Tuan, Director of CMC ATI.
