The Institute for Computer Science, AI, and Technology (INSAIT), part of Sofia University, Bulgaria, announces the release of MamayLM v1.0 - the first free, open and efficient Ukrainian Large Language Model that can handle both text and visual data.
MamayLMv1.0 outperforms up to 5x larger open models on Ukrainian tests and surpasses proprietary models such as GPT-5 mini in chat performance on Ukrainian-specific topics. Due its compact size, 4B and 12B parameters, it can run on a single GPU, enabling cost efficient adoption of AI.
The release of MamayLMv1.0 is coordinated with the Ministry of Digital Transformation of Ukraine as part of their ongoing national LLM initiative as well as with the Bulgarian Ministry of Innovation and Growth.
The model is ideal for enabling rapid Ukrainian AI autonomy and deployment across multiple sectors in Ukraine, including education, law, healthcare, and the public sector. It is also attractive in scenarios where data privacy is important, as it can be run on a local machine.
MamayLMv1.0 is INSAIT’s second Ukrainian LLM release and is based on Google’s Gemma 3 series of models. The first model, MamayLM v0.1, based on Gemma 2 and released a few months ago, was positively received with more than 10,000 downloads in few months. This new model however, comes with several key improvements and new capabilities, including larger context size enabling processing of large documents, the ability to handle visual data, and stronger language performance.
MamayLM v1.0 is completely free to download on the Hugging Face platform. All benchmark results and comparisons are also publicly available. We also publish more technical information in the release blog, available in English and Ukrainian.
The release of MamayLMv1.0 is part of INSAIT’s strategy for creating specialized open models which serve as the basis for achieving national AI autonomy in a cost-efficient, controlled, secure and predictable manner.