MTS AI unveiled Cotype Pro 2, a new AI model for business with a focus on autonomy and accuracy

MTS AI has released an updated version of its major language model for the enterprise segment – Cotype Pro 2. The new generation is noticeably faster and more accurate, with a 40% increase in processing speed and a 50% improvement in quality for long texts up to 128,000 tokens (approximately 900,000 characters). This model will be the technological foundation for future intelligent agents focused on solving business problems with a high degree of autonomy.
The new generation of intelligent agents will become the technological foundation for future intelligent agents.
Solutions for business with a focus on security
Cotype Pro 2 has already been customized for more than a hundred application scenarios in the enterprise environment. And it can be deployed entirely on the customer’s internal servers-without connecting to external cloud services. This is especially important for companies that place a high priority on protecting sensitive information.
According to the MERA rating compiled by the Alliance for Artificial Intelligence, Cotype Pro 2 is among the top 5 best Russian-speaking AI models. In 2025-2026, it will be used to produce specialized AI-agents for the public sector, financial organizations, retail and other areas.

Emphasis on logical thinking and structured answers
MTS AI Executive Director Dmitry Markov emphasized that the key area of development remains AI’s ability to reason: from analyzing a situation and formulating conclusions to making decisions under conditions of uncertainty. This is especially relevant, for example, in the field of financial analytics or scientific research, where AI can assess risks and find the most informed decisions.
“In modern business, it’s important not only for AI to be able to follow instructions and process large data sets, but also to produce clear, structured answers suitable for use in other digital systems,” Markov said.
Improvements in all key metrics
- idea generation – improving accuracy from 44% to 57%;
- information extraction – from 81% to 86%;
- general communication – from 55% to 64%;
- creative writing – up to 65%;
- text summarization – up to 85%;
- classification accuracy remained at a high level – 87%;
The new version also has significantly improved logical reasoning ability, which makes it especially valuable for analytical and scientific tasks.
Practical testing and new features
Before its public release, Cotype Pro 2 underwent three months of testing in real-world business scenarios. In one bank, the AI automatically distributed customer service calls, and in an MTS Live project, it created descriptions for events. In addition, the model was successfully integrated with Retrieval-Augmented Generation (RAG) technologies that allow extracting relevant information from internal knowledge bases and generating answers based on it. Such tools are especially needed in intelligent chatbots and digital assistants.
Role interaction and proprietary quality metrics
The new model supports “role simulation” mode, allowing dialog on behalf of a specialist – this is relevant for both customer support and employee training. In addition, MTS AI has developed a unique approach to assessing the quality of long text processing, where Cotype Pro 2 demonstrated a twofold improvement over the previous version.