What happened

The Ministry of Digital Transformation announced that beta testing of the first national large language model (LLM) trained on unique Ukrainian data will begin in spring 2026. The project's technical partner is the operator Kyivstar, which is funding the development and will transfer the model into state ownership.

"Ukraine will launch beta testing of the first national LLM in spring 2026..."

— Ministry of Digital Transformation (statement)

Technical base and timeline

The model will be adapted from Google's open LLM — Gemma — for the Ukrainian language and local context. An independent group of experts is involved in testing to assess technical quality, ethics, and conformity with Ukrainian realities. Completion of the initial text corpus, an improved tokenizer and an evaluation system are planned for January; public testing is expected to begin in spring. Citizens will choose the model's name through a vote in the "Diia" app.

Why it matters: security, language, public services

This is not just a technological project — it is an element of information and digital security. A model trained on local data will better understand the Ukrainian context, language specifics and legal realities, and therefore can improve the quality of state services, automated responses and analytics. In addition, having a national LLM reduces dependence on foreign services in critical areas.

Data sources and oversight

The team is collecting terabytes of data not only from open sources but also from media, universities and more than 90 government institutions — collection began in December 2025. At the same time, a legal framework for secure data handling is being developed, and experts are forming national benchmarks to evaluate the model.

"An independent group of experts is involved in testing to check technical quality, ethics, language proficiency and understanding of Ukrainian realities."

— Official statement of the Ministry of Digital Transformation

Risks and safeguards

Risks include privacy issues, biases in the data and the possibility of misuse. Measures already being applied include independent review, national benchmarks, legal frameworks and public testing, which will allow the public and specialists to identify problematic cases. It is important that funding and technical support are sustainable and that oversight is transparent.

What could change for the citizen

Gradually, the model could improve the quality of chat support in state services, automate the preparation of documents and localized services, and become a tool for education and media analytics. At the same time, effectiveness will depend on the quality of the data and transparency of the process.

Conclusion

The project combines technology, state policy and public oversight. If declarations turn into stable procedures and transparent supervision, Ukraine will gain not only a tool for services but also an important asset for linguistic and information security. Whether this becomes a long‑term resource depends on how the state and society arrange governance and funding after the beta test.