He detailed why Ukraine needs a national LLM, how it will be trained, and its projected cost.
"Creating a national LLM is a strategic security step for the country," he highlighted.
“Countries are building national models trained on local data. This is necessary to develop sovereign AI and launch products without the risk that user data fed into AI systems will end up in the hands of foreign private companies.”
Bornyakov argues that national LLMs outperform English-language models when it comes to local relevance, thanks to unique training data. These models better grasp national context, dialects, and sociopolitical nuances.
“Ukraine needs its own AI — one that doesn’t think using content publicly available on the internet, including Russian propaganda,” he said.
“It should provide answers based on real data stored in libraries, state archives, and public institutions.”
How the model will be trained
Training a Ukrainian-language model is both logical and essential, Bornyakov said. The LLM is being built using an existing open-source foundation model that already includes English-language data, so there's no need to duplicate it.
Instead, it will undergo additional pre-training in Ukrainian, which will help it adapt to Ukraine’s culture, context, and dialects — a critical requirement for delivering accurate and relevant responses.
Training data will include open Ukrainian-language datasets, namely Maliuk, NER-UK, UA-GEC, among others. But the model’s core will be formed from unique, high-quality Ukrainian-language materials — sourced from universities, government institutions, researchers, educators, and libraries — not available on the open web. While this is not classified information, it’s data that exists offline in institutional repositories. The goal is to gather everything that’s been digitized and use it to fine-tune the model.
How much will it cost?
Cost is a key concern for the government — and using a Ukrainian-trained model will be cheaper than relying on foreign LLMs, where Ukrainian-language tokenization often leads to higher costs. A national LLM will lower the price of AI integration into public services.
Kyivstar will not receive any government funding for the LLM. The company is fully financing the development and serving as the lead implementer. After the testing phase is complete, the government will receive full rights to the model.
The ministry aims to make the model open-source and accessible to all, meaning it could be downloaded and deployed on independent infrastructure.
So how much will it cost Kyivstar? It’s hard to say. Bornyakov noted the figure is “dynamic”, depending on how long the project takes and what infrastructure deals are made. Preliminary estimates suggest $1–3 million.
What to expect?
The first version, planned for release by December 2025, will focus on text-based capabilities. It will answer questions, translate, and manage basic conversations. The computing power will be sufficient for simple interactions.
Future updates will expand the model’s abilities, eventually developing a multimodal version that can understand and generate audio and images. But for now, the top priority is perfecting text processing, which will serve as the foundation for other features.
Ukraine now lacks AI-specific legislation, so the LLM project will follow existing laws on data protection, intellectual property, and human rights — including anti-bias and anti-discrimination safeguards in development and training.
In addition to Ukrainian law, Bornyakov said the team will adhere to international AI standards. Ukraine recently signed the Council of Europe’s Convention on AI, Human Rights, Democracy, and the Rule of Law, and its principles will be incorporated into the LLM’s design.
What’s next?
For now, government officials plan to encourage businesses to self-regulate their use of AI, with legislation similar to the EU AI Act likely to follow.
The national language model initiative was officially launched in June, when Kyivstar and Digital Transformation Ministry agreed to cooperate.
The ministry will coordinate the process and shape the ethical and technical vision for the LLM, establishing key structures for its success — including a coordinating committee and both technical and ethical boards.