The Ukrainian National Large Language Model (LLM), jointly created by the Ministry of Digital Transformation and the mobile operator Kyivstar, has entered the closed beta testing stage. Currently, developers are testing an experimental prototype to verify architectural solutions and the specifics of working with the Ukrainian language.
The Ukrainian National Large Language Model (LLM), jointly created by the Ministry of Digital Transformation and the mobile operator Kyivstar, has entered the closed beta testing stage. Currently, developers are testing an experimental prototype to verify architectural solutions and the specifics of working with the Ukrainian language.
This was reported to DOU by Kyivstar, which is responsible for coordinating the development process, forming engineering teams, and managing cloud computing infrastructure.
The engineers have now completed the first two stages of training the AI model—pre-training and fine-tuning with the help of instructors. The team is now focusing on final calibration and tuning of the algorithm’s responses.
To evaluate the prototype and find its strengths and weaknesses, the project’s specialized scientific committees — linguistic, historical, technical, and ethical and legal — were involved. Based on their findings, the algorithms will be further improved.
At the current stage, experts are testing a reduced version of the neural network, which will later be scaled up to a full-fledged large model. It is expected that the final product will be integrated into government services and commercial IT services.
The Memorandum on the creation of a domestic AI was signed by the Ministry of Digital Affairs and Kyivstar in the summer of 2025. Subsequently, a survey was conducted in the «Diya» application, according to the results of which the future model received the official name «Syaivo». The technological foundation for the development was the open model Gemma 3 from Google, which is optimized for working with the Ukrainian language and supports multimodal functions.
Earlier, Roman Kyslyi, head of the Artificial Intelligence Center at the Ministry of Digital Transformation, explained that most of the data on which modern models are trained is English-language content, so «the more we add high-quality Ukrainian data, the more publicly available datasets we create, the better the models understand the Ukrainian language and context.»
