Ukraine will collect data from 90 government institutions to create a national AI model.

Ukraine creates a national language model based on the open framework Google Gemma

Ukraine is developing a large language model (LLM) using the open framework Google Gemma, which will allow optimizing the processing of Ukrainian texts and enhance the technological independence of the country.

As part of the project, data has been collected from more than 90 governmental institutions, including judicial registers, educational publishers, and regional archives. The training of the model will begin on Google’s computational infrastructure, after which it will be transferred to Ukrainian servers. The Ukrainian developers’ team plans to improve the tokenizer for better processing of Ukrainian texts, conduct training on specially collected materials, and create tests to tailor the model for specific usage scenarios.

In particular, the language model has significant military importance: artificial intelligence will be integrated into battlefield management and enemy monitoring systems. Chinese models, such as DeepSeek and Qwen, have been dismissed due to security concerns. Experts note that current AI systems do not process local dialects well, which is an additional reason for creating a national model. Four advisory committees have been created to ensure quality control, responsible for technical, legal, cultural, and linguistic aspects.

This project is part of a broader strategy of Ukraine to increase the technological independence and data security of the country. In November 2025, another initiative was launched together with NVIDIA for the development of state AI infrastructure. A continuation was the AI Factory platform, which is based on NVIDIA solutions.