Ukraine has selected Google Gemma 3 as the foundational model for its national large language model initiative, the Ministry and Kyivstar announced.

Compute resources will be provided via Google Vertex AI to support model training and deployment within Ukrainian infrastructure constraints.

Selection and partners

The decision was taken jointly by the Ministry of Digital Transformation and the telecommunications operator Kyivstar, the agencies confirmed.

Authorities said Gemma 3 supports more than 140+ languages and includes multimodal capabilities for text, images and other modalities.

Technical details

Model training and inferencing will rely on Google Vertex AI infrastructure, chosen to provide managed services and scalable compute for the project.

Project architects indicated integration work will tailor Gemma 3 to Ukrainian linguistic and domain needs, preserving model safety and performance requirements.

Goals and data

The primary objective is to fine-tune the base model on Ukrainian datasets and create a sovereign LLM under local operational control and oversight.

Developers plan to use publicly available corpora, licensed resources and domestically collected data while adhering to privacy and legal requirements.

Governance and deployment

Authorities emphasised that the resulting system will be administered within Ukraine and optimised for national use cases across government and private sectors.

Specific timelines, funding arrangements and technical partners beyond Google and Kyivstar were not disclosed at the announcement stage.

Officials said the initiative aims to reduce dependence on external models and to support Ukrainian language technology growth domestically.

Further updates are expected as the project advances through testing, fine-tuning and phased deployment steps across sectors.