On July 6, 2026, the AI Research Center - VinUni University officially introduced V-Bench. This is a non-profit toolkit developed to serve the community and promote responsible AI development in Vietnam.
According to VinUni, V-Bench is built on a dataset of more than 40,000 questions and tasks, with the participation of a scientific council of 18 Vietnamese AI experts at home and abroad. The dataset not only evaluates academic knowledge but also reflects elements of culture, language, digital sovereignty, information security and local characteristics of Vietnam.
The toolkit is divided into two major categories with five groups of evaluation criteria including: understanding of implicit culture, ability to handle regional diversity, digital security and sovereignty, practical application capacity in fields such as education and health, along with the ability to plan and use tools of AI models.
According to VinUni, V-Bench not only tests the ability to answer questions but also assesses the level of cultural understanding, regional language and AI's ability to act in the Vietnamese context. The assessment results are built according to a unified methodology, with the appraisal of a panel of experts to ensure objectivity and reference value.
In addition, V-Bench is developed according to common standards in major language model research in the world, helping to be compatible with the international benchmark ecosystem, creating conditions for global AI developers to participate in evaluation.
Prof. Dr. Duong Nguyen Vu - Vice Rector, Scientific Director of the AI Research Center of VinUni University said that V-Bench aims to build an AI reference system for Vietnam, helping agencies, businesses and organizations have more basis to choose AI models suitable for each usage needs.
Currently, the toolkit is provided free of charge at vbench.ai and has announced benchmark evaluation results for 15 major language models.
In the near future, VinUni will continue to expand V-Bench with evaluation criteria related to image and sound such as identification of Nom script, regional signs, charts, cultural heritage, three-region voice recognition, understanding context through short videos and processing Vietnamese documents with a length of over 100,000 tokens such as legal documents, contracts and textbooks.
According to VinUni, V-Bench is one of the school's key AI research projects, aimed at contributing to the research community and promoting responsible AI development in Vietnam.
