We are proud to announce that Mykola Haltiuk, a PhD student and employee of the Faculty of Computer Science at AGH University of Science and Technology, was part of the team of researchers who developed Lapa LLM v0.1.2, the most effective open language model for the Ukrainian language.
The model was created by an interdisciplinary team of scientists from the UCU Faculty of Applied Sciences (Ukrainian Catholic University), AGH University of Krakow, National Technical University of Ukraine “Kyiv Polytechnic Institute,” and Lviv Polytechnic National University. The aim of the project was to create an open, highly efficient language model optimized for processing the Ukrainian language.
Lapa LLM is based on the Gemma-3-12B architecture and stands out for its efficiency and accuracy compared to previous models. Among other things, the team developed a new, optimized Ukrainian language tokenizer, authored by Mykola Haltiuk, which allows the model to process text with 1.5 times fewer tokens while maintaining high-quality results.
The model achieved the best results on 18 Ukrainian language benchmarks and also stands out as the best English-Ukrainian translator (33 BLEU on FLORES) and one of the best models for summarizing and answering questions (Q&A) in its size class.
The project is of great importance not only scientifically but also socially – it is a step towards strengthening Ukraine's technological independence by creating language tools that support the development of AI in the native language.
The team emphasizes that Lapa LLM is a fully open-source model, available for scientific and commercial applications.