According to a recent post by a the Ukrainian Ukraine is working on a sovereign, national large language model. The model, which should enter beta in spring of 2026, is focused on giving the country it’s own fully-controlled LLM that will allow it to presumably reduce dependence on any other models, be they from the United States or anywhere else. This type of sovereign infrastructure brings to mind Russia’s GLONASS global positioning system built as a reaction to the technical hegemony of the world standard, GPS.
Locally-run Kyivstar is acting as the technical and financial partner and private partners, not the government, will pay for the development.
“Ukraine’s mission is to rank among the world’s top three countries in artificial intelligence development. One of the most ambitious steps toward that goal is launching a homegrown large language model trained on unique Ukrainian data,” wrote on LinkedIn.
The model will be trained on Google’s Gemma model, an open version of the company’s powerful Gemini models.
“These are advanced technologies that we are adapting to the Ukrainian language and national context,” wrote Fedorov.
The model will ingest data from Ukrainian government institutions, media outlets, and academic groups which will bring together “terabytes of unique data” with a definitively Ukrainian focus. It will also include a framework to fairly and responsibly manage the training data.
Interestingly, Fedorov said the LLM would be named by the Ukrainian people through an online poll. These kinds of sovereign LLMs should start becoming the norm as countries realize that commercial LLMs could contain biases or, at worst, be poisoned by bad actors. While none of these model are being used in mission critical situations, the time will come when a sovereign LLM will be key to a country’s technical future.
