- Direction:
- GovTech
Ukraine Moves Toward a Sovereign AI Model: National LLM to Enter Beta in 2026
- Publication date and time:
- Reading time:
- 4 min
Ukraine has entered the implementation phase of one of its most ambitious digital initiatives to date: the development of a national large language model (LLM). After moving from concept to concrete technical architecture in 2025, the first beta version of Ukraine’s sovereign LLM is expected to launch in spring 2026.
The project is being developed in partnership with Kyivstar, Ukraine’s largest telecom operator, which is financing and leading the technical development. Upon completion, the model will be transferred to the state. No public budget funds are being used at this stage.
The technological foundation of the model is based on Gemma, Google’s open family of language models, which is being adapted to the Ukrainian language and national context. This includes linguistic structure, cultural references, historical sources, and domain-specific data relevant to public administration, education, and security.
Data as the Core Challenge
The most critical phase currently underway is data preparation. Building a high-quality Ukrainian-language LLM requires far more than scraping publicly available online content. Training datasets are being assembled from a wide range of sources, including government institutions, media organizations, universities, scientific institutions, and historical archives.
A significant share of relevant Ukrainian-language material still exists only in offline or paper form, highlighting a broader structural challenge: the need for large-scale digitisation of national knowledge assets. Addressing this gap is essential not only for AI development, but for long-term digital sovereignty.
To ensure quality and accountability, the development process is overseen by an independent expert structure working across four dimensions: scientific and technical, legal, cultural-historical, and linguistic. One of the committee’s current priorities is the creation of professional benchmarks — a dedicated evaluation system that will test model performance, safety, and alignment with Ukrainian-language norms.
By January 2026, the team plans to deliver:
- the first structured text corpus for model training,
- an improved Ukrainian-language tokenizer to enhance processing speed and accuracy,
- proprietary benchmarks to assess model quality and safety.
In parallel, a legal framework is being developed to govern data usage, intellectual property compliance, and transparency in model training.
From Vision to Infrastructure
The national LLM is not an isolated technology project. It aligns with Ukraine’s broader ambition to build sovereign AI infrastructure as part of its long-term innovation and security strategy.
As previously outlined by Mykhailo Fedorov in his op-ed for Digital State UA, “Why Does Ukraine Need Its Own LLM”, sovereignty in AI is not only about language or convenience. It is about control over data, the ability to deploy AI services without reliance on foreign infrastructure, and the capacity to integrate AI across public services, defence technologies, and the private sector under Ukrainian jurisdiction.
This approach is being coordinated through the WINWIN AI Center of Excellence, which operates within Ukraine’s Global Innovation Strategy WINWIN 2030. Unlike many international AI initiatives led by individual companies, Ukraine’s model is designed to embed AI horizontally across the state — from governance and education to defence and economic development.
Once tested, the national LLM will be made available to public institutions, research and educational organisations, and civic initiatives. Following a validation phase, the model is expected to be open-sourced, allowing businesses and developers to build services on top of it.
What Comes Next
The beta release planned for spring 2026 will be trained on uniquely Ukrainian datasets and evaluated against international benchmarks. The stated goal is not symbolic independence, but competitiveness — ensuring the model can perform alongside other open LLMs in global rankings.
In early 2026, Ukraine will also launch a public vote via the Diia ecosystem to select the model’s official name, reinforcing the project’s positioning as national digital infrastructure rather than a closed government system.
Taken together, the national LLM represents a shift from declarative AI ambitions to executable state capacity — where artificial intelligence becomes part of how public institutions function, scale, and adapt over time.
