Ukraine Develops Independent AI System Using Google Open Framework
Ukraine is developing its own independent artificial intelligence system using Google's open framework Gemma.
The Gaze reports on it, referring to Reuters.
The Ministry of Digital Transformation of Ukraine, together with mobile operator Kyivstar, announced the creation of a large language model that will first be trained on Google's computing infrastructure and later fully deployed on local servers.
The Gemma 3 model was chosen as the basis for creating a Ukrainian large language model (LLM) to train this model on unique Ukrainian data, thus minimizing linguistic and ethical risks and ensuring high-quality performance with the Ukrainian language.
“When making our choice, we focused on how well the model already processes Ukrainian-language texts and how controllable it is during training. This will help minimize linguistic and ethical risks in our LLM,” said Danylo Tsivok, Chief AI Officer at the Ministry of Digital Transformation and CEO of WINWIN AI Center of Excellence.
As part of the model adaptation, Ukrainian tokenizer is planned to be improved, trained on specially collected texts, and benchmarks for fine-tuning are to be created. Gemma supports over 140 languages, has multimodal capabilities, up to 128,000 tokens, and a flexible architecture that allows the model to be effectively adapted to various tasks.
This decision was made in order to fully control the system and minimize dependence on foreign proprietary solutions, such as ChatGPT from OpenAI or alternative models from China.
The project aims to support both government and military operations, as well as civilian services and the private sector. The system will help analyze military data, coordinate actions on the battlefield, control satellite and aerial reconnaissance, as well as combat disinformation and prioritize demining.
The developers are also preparing security measures against potential cyberattacks, including malicious instructions for AI.
The key advantages of choosing Gemma are its optimal balance of performance and resources, multilingual support, multimodality, and an extended tokenizer for long context. The model has already been successfully used to create leading Ukrainian LLMs, such as Lapa LLM and MamayLM, as well as the Bulgarian LLM INSAIT BgGPT.
Earlier, Ukraine has partnered with NVIDIA to develop a state-controlled AI infrastructure aimed at creating a “sovereign state of artificial intelligence.” The initiative is aimed to provide secure data processing, advanced computing capabilities, and a trained workforce to develop and operate AI technologies, with the first project being Diia AI LLM, a large language model tailored to Ukrainian legislation, public administration, and government workflows.
As The Gaze previously reported, over the past three years, Google had provided Ukraine with more than $45 million in financial assistance and more than $10 million in technical solutions and educational resources.
Read more on The Gaze: Ukraine’s Government Action Program 2025–2026: Building a Resilient State Through Defense, Reforms, and European Integration