Russian Researchers Introduce a New Generation of Compact AI
The AIRI Institute, with support from Sber, unveiled a new family of compact AI models at the St. Petersburg International Economic Forum 2026. The technology is designed to process requests faster while requiring significantly fewer computing resources than large language models.
The new models are based on an approach known as Optimal Cognitive Core. Unlike traditional systems that store vast amounts of knowledge within the model itself, AIRI's development acts as a cognitive core. It analyzes information, builds logical connections, and accesses external databases, search engines, and other sources whenever additional information is needed.
According to AIRI CEO Ivan Oseledets, the developers focused not on expanding the model's memory but on improving its ability to work with data.
Tests showed that the systems process requests 1.5 to 2 times faster than large language models while using roughly 1.5 times fewer computing resources. Their compact size allows them to run not only on server infrastructure but also on laptops and smartphones. The models are already available as open-source releases and can be used in financial services, customer support, corporate knowledge bases, as well as legal and medical applications.
