At the IT Arena 2025 conference, Ukrainian scientists presented Lapa LLM — the first large language model of artificial intelligence, created specifically for working with the Ukrainian language and context.
The presentation of the model was conducted by a graduate student of the Ukrainian Catholic University and a data specialist of the Nortal company Yuriy Paniv. He noted that the Lapa LLM is designed to address critical issues that are overlooked by global open LLM models, especially in national security, culture and linguistic accuracy. The model is developed taking into account the needs of national security and privacy to work with sensitive data in the defense sector and large companies without the involvement of cloud providers.
The Lapa LLM is culturally coherent, learning with a focus on Ukrainian values and context, using automatic data filtering to prevent Russian propaganda, disinformation and “hallucinations” about knowledge about Ukraine. In addition, the project is aimed at increasing productivity for the Ukrainian language, which remains a problem for many open LLMs.
Lapa LLM is based on Google’s Gemma model with 12 billion parameters, which provides an optimal balance between performance and resource requirements. Among the technical innovations of the developers is the Ukrainian tokenizer, which allows you to reduce the number of tokens by 50% when processing Ukrainian text. The release of the model, along with training scripts and datasets, is scheduled for early October, and the team plans to publish Lapa LLM under the MIT Open License.
