The Institute for Computer Science, Artificial Intelligence and Technology (INSAIT) at Sofia University “St. Kliment Ohridski” has unveiled MamayLM v1.0 – the first open-source and practical Ukrainian model capable of processing both text and images.
According to the institute, MamayLM v1.0 delivers performance that surpasses models up to five times larger when working with Ukrainian language tasks, while also achieving strong results in English. It even outperforms closed conversational systems such as GPT-5 mini. Designed to be efficient and affordable, the model can run smoothly on a single graphics processing unit (GPU), lowering the cost barrier for wider adoption.
The project builds on the earlier MamayLM v0.1, released just a few months ago. That first version was downloaded more than 10,000 times and received widespread positive feedback. The new generation introduces two variants – with 4 billion and 12 billion parameters – and features major upgrades, including the ability to handle visual data, extended context windows for better document processing, and improved accuracy in natural language tasks.
MamayLM v1.0 is developed on top of Google’s open Gemma 3 model series, with INSAIT incorporating its own innovations based on its growing expertise in building national AI systems. The institute highlights the release as an important step toward ensuring Ukrainian technological sovereignty in artificial intelligence. Because the model can be deployed locally in government institutions, businesses, and organizations without heavy investment, it provides an option to safeguard sensitive data by keeping it in-house.
The initiative has been coordinated with the Ministry of Digital Transformation of Ukraine, as part of Kyiv’s strategy to establish a sovereign AI model, as well as with Bulgaria’s Ministry of Innovation and Growth. The model is freely available on the Hugging Face platform, while additional documentation and updates are being published on INSAIT’s official blog.
With MamayLM v1.0, INSAIT continues its mission to help countries secure autonomy in artificial intelligence through transparent, secure, and accessible solutions, combining open research with practical applications for governments and businesses alike.
Source: press release