T-Pro (Large Language Model, LLM)

The main articles are:

• Speech Recognition (Technology, Market)
• Speech technology: On the path from recognition to understanding

2025: T-Pro 2.0

On July 18, 2025, T-Technologies introduced T-Pro 2.0, its own large language model with a hybrid reasoning mode. Hybrid models allow you to choose between quick simple responses and deep analysis, which helps optimize computing costs when implementing AI.

𝓲

Фото: T-Technologies

According to the company, the total development costs, including the costs of computing power for R&D and final further training, as well as the salaries of employees, do not exceed 120 million rubles.

Through reasoning, the model learned to solve complex analytical problems, structurally reason and plan its actions, put forward hypotheses and refute them on its own. If earlier models could be given an autonomous task, which takes an average of 30 seconds for a person, then models with a reasoning mode can be assigned tasks for which a person takes an average of 8 minutes. Solutions in such models are more transparent and reproducible, making them the technological base for next-generation agents who are able to interact with external interfaces and perform tasks without human control. For example, in supporting AI agents help not only to generate answers, but also to resolve issues without involving an operator, in T-Bank the share of such requests has already reached 45%.

T-Bank plans to integrate the model into the work of its AI assistants, which are available to all clients of the bank, as well as apply to improve the efficiency of operational processes through copy solutions and autonomous agents.

The transition to the internal version of T-Pro 2.0, trained on its own internal data, optimized the quality of work of T-Bank's autonomous support agents: they began to solve user problems without operator participation 10% more often without additional costs for development or operational processes.

T-Pro 2.0 with 32 billion parameters is available on Hugging Face in the public domain under the Apache 2.0 license, which allows companies to freely download, use, adapt and retrain their data and integrate it into their solutions. Together with the model, benchmarks were published, and the T-Wix instructional dataset was first posted, which contains about 500 thousand examples.

Key features of the T-Pro 2.0:

• Hybrid reasoning mode: the model allows you to choose an answer strategy: give it immediately with simple questions or use reasoning to solve complex problems, which allows you to use the same model for different business processes and save resources.
• Implementation savings: Thanks to a denser tokenizations Russian language and faster token generation by larger language models, using an additional smaller model, T-Pro 2.0 provides answers twice as fast as the Qwen3 and DeepSeek R1 Distil model of similar sizes, which reduces the cost of processing information and developing solutions based on the model. On average, the model allows you to generate an answer with reasoning in Russian on typical equipment 2 times cheaper.

The model is part of Gen-T - a family of their own large language models of the T-Technologies group. In 2024, the company released the first large language models with 32 billion parameters and 7 billion parameters: T-Pro and T-Lite.

2024: Opens access to T-Pro and updated T-Lite models

The T-Technologies Group has opened access to two large language models (LLM) - a T-Pro with 32 billion parameters and an updated T-Lite with 7 billion parameters. Now any Russian company will be able to use their capabilities for free. Representatives of the group told TAdviser on December 11, 2024. According to T-Technologies, as shown by industrial benchmarks (MERA, ruMMLU, Ru Arena Hard, MT Bench and AlpacaEval), T-Pro and T-Lite "surpass all Russian and foreign models in terms of general level of knowledge, ability to conduct dialogue and perform practical tasks." Read more here.

Both models (T-Pro and T-Lite) are included in Gen-T - a family of their own specialized language models of the T-Technologies group.

Compared to the predecessor T-Lite, the T-Pro model has increased the number of parameters - from 7 to 32 billion. This makes the model more powerful and productive. More parameters allow the model to take into account more context and language features, better memorize information, and draw more accurate and complex conclusions.

𝓲

Фото: "T-Bank"

The model works in two modes: it can be further trained for specific business tasks (Fine-tuning), as well as used in prompting mode - to set tasks for the model in dialog mode.

𝓲

Фото: "T-Bank"

According to T-Technologies, among the open models of its weight category, T-Pro ranks first in solving problems in Russian according to the industrial benchmarks MERA, ruMMLU, Ru Arena Hard, MT Bench and AlpacaEval. Among proprietary (closed) models, the T-Pro ranks second in a number of benchmarks, behind only GPT-4o.