T-Bank Launches Russia’s First LLM with Hybrid Reasoning

T-Pro 2.0, a new open-source language model from T-Bank’s tech division, rivals global peers with hybrid reasoning and commercial-grade performance—free to use for businesses.

T-Technologies, part of the T-Bank holding group, has released T-Pro 2.0—the first Russian large language model (LLM) to feature hybrid reasoning capabilities. The model is open-source and free to use commercially, including for building AI agents to automate business processes. Until now, Russian companies largely relied on Chinese models for such tasks.

T-Pro 2.0 can break down complex problems into logical steps, build reasoning chains, and refine its own hypotheses—making it suitable for analytics, engineering, and even code generation. According to a T-Technologies spokesperson, the model can handle tasks that take a human up to eight minutes to solve, compared to earlier AIs that maxed out at 30-second operations. The hybrid reasoning architecture also reduces the rate of hallucinations—situations where the AI generates inaccurate or misleading information.

One of T-Pro 2.0’s standout features is its efficiency. The model’s tokenizer has been optimized for the Russian language, resulting in a 30% reduction in computing resource demands. This translates to faster response times and lower deployment costs. Total development costs, including R&D and fine-tuning, were kept under 120 million rubles.

Industry experts see T-Pro 2.0 as a milestone for Russia’s tech landscape. The open-access release of a hybrid-reasoning model boosts AI sovereignty, while local deployment mitigates data leakage risks—especially critical for industrial and financial sectors.