Russian developers presented two new large language models

T-Technology Group (which includes T-Bank) has opened access to two large language models (LLMs): T-Pro and the updated T-Lite, created by domestic developers. This was announced by representatives of the company.

As emphasized in T-Technologies, these models surpass all Russian and foreign analogs (on industrial benchmarks).

"When we started to develop products based on large language models - for example, Copilots for employees and Universe AI-assistants - we were once again convinced that the existing solutions on the market do not meet our requirements. So we started developing Gen-T, a family of specialized language models. Once we were convinced of the effectiveness of our solution, we decided to share our models with the whole industry and change the approach to using LLM. Our experience can be adopted by other companies, and the use of LLM will become much wider," said Victor Tarnavsky, Director of Artificial Intelligence at T-Bank.

The models are part of Gen-T, a family of T-Technology Group's own specialized language models. The models of the family are designed to solve specific highly specialized tasks, unlike universal solutions such as ChatGPT.

Continual Pretraining technology is used to create the models. This is a process in which a model that has already been trained on large amounts of information continues to be trained on materials specific to a certain task or domain and adapts it to Russian. The T-Lite and T-Pro models are based on the Qwen-2.5 family of models, but perform better on Russian language tasks than the original models.

As stated in T-Technologies, the presented open models will give domestic companies an opportunity to bring their technological development to a qualitatively new level and will give a new impetus to the country's economy.