Ukrainian LLM to enter beta testing in spring 2026 - Fedorov
Kyiv • UNN
The Ukrainian national large language model (LLM) will enter beta testing in the spring of 2026. The development is financed by Kyivstar, and the model is being trained on the Gemma family of models from Google.
The Ukrainian national LLM (large language model) will enter beta testing in spring 2026, First Deputy Prime Minister - Minister of Digital Transformation Mykhailo Fedorov announced on Wednesday in Telegram, detailing the stage of development, writes UNN.
The national LLM will enter beta testing in the spring. Ukraine's mission is to enter the top 3 countries in the world for artificial intelligence development. One of the largest steps is the launch of its own large language model, trained on unique Ukrainian data.
Who is developing the national LLM
"LLM development requires resources, and in times of war, all funds must be directed to defense. Therefore, Kyivstar became the technical partner of the project. The company finances the development of the national large language model and will then transfer it to the state," the Deputy Prime Minister stated.
According to him, the Ukrainian LLM will be trained on Google's Gemma family of models – our strategic partner. "These are advanced technologies that we are adapting to the Ukrainian language and context," he noted.
"And the quality of the national LLM development will be tested by a group of independent experts from various fields, who will check the model for: technical quality; ethics and safety for users; knowledge of the Ukrainian language; understanding of the national context," Fedorov said.
What stage is the development at now?
"Currently, the most important part of the work is underway - collecting data for LLM training. For high quality, information from the internet is not enough, so we are working with government agencies, media, universities, and other institutions - these are terabytes of unique data," Fedorov said.
In parallel, according to him, a "legal framework is being created to responsibly work with the data provided to us for model training."
Also, he reported, "a group of experts is already working on creating benchmarks - tests for the language model that will help evaluate and improve its quality." This, the Deputy Prime Minister explained, "will allow us to keep the effectiveness and safety of the language model in focus."
When to expect the launch of the Ukrainian LLM
In January 2026, we will already have: the first text base for LLM training; an improved tokenizer - a tool that divides words into elements so that LLM processes language faster and more productively; our own benchmarks for quality assessment. In spring 2026, we plan to launch beta testing. And the name for the Ukrainian LLM will be chosen by Ukrainians themselves - in January, we will start voting in "Diia". We will make Ukraine an AI-first state.
Addition
Last March, when announcing the start of development, Fedorov stated that the launch of the national LLM would kickstart the mass production of Ukrainian AI tools. According to him, thousands of AI products are created based on LLM - virtual assistants like ChatGPT, text generators like Notion AI, and programming tools like GitHub Copilot.