Hanoi (VNA) – As Vietnam races to embrace digital transformation and artificial intelligence (AI), a homegrown innovation is poised to redefine how users handle the country’s complex language and documents.
The CMC Research Institute for Applied Technology (CMC ATI) has unveiled CATI-VLM, an AI model for document reading that has already claimed a spot among the world’s elite, ranking first in Vietnam and among the top 12 globally in the Document Visual Question Answering (DocVQA) category of the Robust Reading Competition (RRC) in June 2025.
The RRC, launched in 2011 by the Computer Vision Centre at the Autonomous University of Barcelona, is a prestigious global stage for advancements in computer vision and document recognition. It has attracted heavyweights like Tsinghua University, Hyundai Motor Group, and Tencent. Held alongside the International Conference on Document Analysis and Recognition, it tests the ability of AI to interpret intricate documents, which is a critical need in Vietnam, where the language’s diacritics and handwritten texts pose unique challenges.
Unlike traditional optical character recognition systems that merely extract text, CATI-VLM goes deeper, analysing not just words but also checkboxes, charts, signatures, formulas, and document layouts.
Trained on a sprawling 5-terabyte dataset, the model can answer questions posed on document images, much like conversational AI tools such as ChatGPT, without requiring prior training on specific formats. Its versatility makes it a powerful tool for digitising documents, automating business processes, and strengthening governance.
Remarkably, CATI-VLM achieves this with just 3 billion parameters, a fraction of the size of competitors like Deepseek (27 billion parameters), GPT-4 Vision Turbo paired with Amazon Textract OCR (Top 34), and Baidu (Top 22). Yet it topped four of the seven RRC benchmark datasets, showcasing a balance of efficiency and accuracy tailored to Vietnam’s computing infrastructure.
Nguyen Trung Chinh, Chairman and Executive President of CMC Corporation, attributed the breakthrough to over a decade of research investment and a commitment to developing Vietnamese-origin technologies. “This milestone aligns with our vision for AI-driven transformation and global expansion,” he said.
Looking ahead, CMC plans to embed CATI-VLM into its C.OpenAI ecosystem, powering tools like CLS – a legal document auditing assistant, SmartDoc – a platform for digital document transformation, and CMC KMS – a knowledge management system. The model will also drive automated reporting and next-generation document applications./.
