MTS AI: Cotype (Large Language Model, LLM)

The main articles are:

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

2025

Cotype Pro 2.5 with Advanced Agent Skills

MWS AI October 30, 2025 released LLM Cotype Pro 2.5 with advanced agent skills.

Cotype Pro 2.5 - received improved capabilities for integration with corporate knowledge bases and working with computer code.

Cotype Pro 2.5 is designed to develop AI assistants in the corporate segment. It can be deployed on the customer's infrastructure as a separate solution, as well as as as part of the MWS AI platform for creating AI assistants.

The model supports integration with external systems and APIs, including enterprise scheduling tools, mail services, CRM platforms, project management systems, and web traffic analytics.

A distinctive feature of the model is the improved ability of multi-step operation in agent mode. With this, Cotype Pro 2.5 is able to automate complex multi-step processes that require sequential access to current data from various sources and offline interaction with corporate systems. The efficiency of the model in such scenarios is ten times superior to the Cotype Pro 2, and 2.5 times Qwen3-32B. In particular, the model showed high accuracy during testing in scenarios in the telecom sphere, where the agent and the user coordinate actions to solve technical problems with communication services, and in air travel, where interaction with booking systems is required when changing flight schedules, flight parameters or baggage.

Using RAG (Retrieval-augmented generation) technology, the model can access databases, text archives and other external sources for up-to-date information and use it to generate more accurate and relevant answers. This significantly improves the quality of data analysis and processing, making it easier to use the model to solve a variety of business tasks and information requests. In terms of accuracy when working with Russian-language materials (interpretation and use of content in Russian when generating answers), Cotype Pro 2.5 exceeds Qwen3-32B by 22%.

This version has significantly improved the ability to work with computer code. Cotype Pro 2.5 shows growth across all major metrics: code documentation - + 37.4%, test generation - + 20.3%, code generation - + 16.4% compared to Cotype Pro 2. The average estimate rose by 24.4%, reflecting overall improvements in output stability and quality and making the model more applicable to engineering and integration challenges.

Cotype Pro 2.5 is an important step in the development of our AI platform, which is not only the basis for the implementation of applied intelligent solutions, but also a technological enabler for the start of the AI transformation of corporations. The model has improved capabilities in terms of generating ideas, extracting, classifying and generalizing information, is able to work with computer code, call functions and schedule the process of performing tasks. This helps us and our clients create AI assistants whose abilities go far beyond a simple user dialogue, "said Denis Filippov, CEO of MWS AI.

To deploy the Cotype Pro 2.5 model, you need one graphics processor with at least 40 GB of video memory (80 GB is recommended, for example, Nvidia A100 80GB or H100 80GB), 100 GB of disk space, a 16-core processor, 48 GB of RAM.

Cotype VL Model View

MWS AI introduced its multimodal model - Cotype VL, capable of simultaneously analyzing and interpreting images and text. The company announced this on October 9, 2025. The model is supplied both as a separate product and as part of the MWS AI AI assistants for a wide range of scenarios: from searching for documents containing visual information to client support for users from screenshots and preparing reports based on graphic data.

Cotype VL contains 32 billion parameters and recognizes images with printed, handwritten and mixed text, takes into account the visual context when translating from one language to another. The model is also able to create a short and detailed description of images and answer complex and logical questions about their content, requiring reasoning, comparisons and conclusions.

Cotype VL supports Russian, English, Chinese and other languages, which makes it convenient for companies with international document management. Cotype VL can be deployed in the closed loop of the company and, if necessary, retrained together with MWS AI on customer data.

Integration into the "Thesis" EDMS

MWS AI (part of MTS Web Services) Haulmont and join forces for development artificial intelligence EDMS THESIS in other solutions. Within the framework of the partnership in the EDMS THESIS, large language models of MWS AI, placed both cloudy and locally, can be used. This was MWS announced on September 29, 2025. More here.

Cotype Pro 2 as part of PAC dl AI Rubytech

MWS AI began supplying large language models of Cotype Pro 2 - as part of the domestic software and hardware complex for AI from the Rubytech Group. This was announced on June 4, 2025 by the head of Rubytech Igor Vedyokhin. Read more here.

Cotype Pro 2

MTS AI has released the second generation of its large Cotype language model designed for corporate clients - Cotype Pro 2. The company announced this on April 1, 2025. This version is 40% faster than the previous one and 50% more accurate in processing long texts - up to 128 thousand tokens (about 900 thousand characters). The model will become the basis for a line of AI agents capable of solving business problems at a high level of autonomy.

Cotype Pro 2 is adapted to more than 100 enterprise scenarios and can be deployed on the internal infrastructure of companies - without connecting to external servers, which increases data protection. In 2025-2026, based on it, MTS AI plans to release a series of AI agents for the public sector, banks, retail and other areas.

A key requirement is the development of reasoning, which will allow models to analyze, draw conclusions and make informed decisions in conditions of uncertainty, for example, to assess risks in financial analytics or find optimal approaches in scientific research. It is also important for AI agents not only to follow the instructions well and work with a large amount of information, but also to issue answers in a structured form, understandable to other systems, "said Dmitry Markov, Executive Director of MTS AI.

This model demonstrates an increase in efficiency in all main tasks: generation of ideas (from 44% to 57%), extraction of information (from 81% to 86%), communication on common topics (from 55% to 64%). Summarization (up to 85%) and creative writing (up to 65%) have also been improved. At the same time, the classification accuracy remained at a high level - 87%. The Cotype Pro 2 also does a better job of reasoning, which is useful for financial analytics and scientific research.

Before the release of MTS, AI tested the Cotype Pro 2 model in real conditions for three months: in the bank, it automatically sorted requests for support, and for MTS Live created descriptions of events. Also, as part of the testing, the possibility of integrating with the RAG (Retrieval-Augmented Generation) tools for extracting information from knowledge bases with text generation was confirmed. Such solutions are in demand, for example, in chat bots, where you need to regularly update data and respond to the user in a few seconds.

In addition to basic tasks, Cotype Pro 2 supports "role-playing" - conducting a dialogue on behalf of a specialist, which is useful for supporting customers and training employees. The company also developed its own method for evaluating working with long texts, where the model showed a twofold increase in quality compared to the previous version.

Implementation in MTS Live

The service for buying tickets MTS Live and the developer of large language models MTS AI (MTS Artificial Intelligence Center) completed a pilot project in which Large Language Model (LLM) Cotype was tested to prepare SEO-optimized descriptions for performances for the ticket window. This was MTS announced on March 26, 2025. More. here

2024

Launching Cotype Nano

The company MTS AI has released an open large language model (LLM) Cotype Nano to solve business problems related to the creation and analysis of texts on. Russian The company announced this on November 26, 2024. The model can be run locally on personal devices - mobile phones desktop computers laptops and medium performance, which makes it available to a large range of users.

Model weights - the parameters that it uses to make decisions are open to researchers and developers. This allows them to study how the model works, customize it for their own needs and use it in their own projects without the need to build everything from scratch.

The model performed well in its class at the Ru Arena Hard benchmark. Testing assesses the accuracy, quality, and relevance of responses to user questions compared to other models.

Cotype Nano is capable of processing context up to 32,000 tokens (about 45 pages of text) at a time, which allows you to work with large amounts of data. The model is trained to create content, accurate and fast translations between Russian and English, process and analyze text data to improve client service, and can be used to develop chatbots and virtual assistants. In addition, it has advanced data classification capabilities - this is necessary for scenarios such as automatic search and analysis of information in corporate knowledge bases. 

Cotype Nano is optimized to run on CPU and GPU, with additional optimization for, processors Intel which allows you to run it on laptops and even smartphones makes the model available to a wide range of developers and companies that do not have access to powerful computing resources.

MTS AI adheres to the principles of openness and transparency in the development of generative artificial intelligence. We launch an open large language model with a commercial license and develop new services for automatic training of neural networks and code creation that will speed up the development process. The development of open LLM models in Russia will allow companies and novice developers and researchers to create solutions based on neural networks without investing significant resources in development and equipment, "said Sergey Ponomarenko, director of LLM products at MTS AI.

Deployment of MTS AI Cotype Plus in the AFL Process Sandbox

The AFL Technology Sandbox has deployed a GPU-oriented infrastructure for high-performance matrix and vector operations in order to pilot solutions using technologies. artificial intelligence The Association announced this on September 12, 2024. In particular, a large MTS AI Cotype Plus language model has already been deployed on this infrastructure. More. here

Optimization for Tatar language texts

MTS AI developed an updated version of the large Cotype Lite language model for working with texts in Tatar the language. The company announced this on September 10, 2024.

LLM is able to process documents up to 8 thousand tokens (approximately 5 A4 sheets), extract and summarize data in a few seconds.

Cotype Light can be used in archives, libraries, state in and private organizations - wherever there is a need for processing information and analyzing documents in Tatar. For example, with the help of a large language model, it is possible to speed up the processing of applications to government agencies.

The Cotype will extract key information such as the subject matter of the request, the location and the personal data of the applicant, and transfer them to the appropriate database. Like other models of the Cotype family, this version can be installed in the contour of the organization, which allows you to eliminate information leaks.

{{quote 'Creating a large language model in Tatar, the developers of MTS AI pursued several goals. First, we wanted to support the diversity of languages ​ ​ existing in Russia, help them develop and be in demand in the digital era. Secondly, this project has shown that we are able to adapt our models to any scientific and business tasks, including such non-trivial ones as information processing in the languages ​ ​ of the peoples of Russia, "said Dmitry Markov, Executive Director of MTS AI. }}

So that the Cotype Light model can understand an unfamiliar language, the developers collected the datacet and translated it from Russian into Tatar. After that, all the data and answers of the model were checked by Turkologists and native speakers. Cotype Light training takes place at MTS Web Services facilities .

According to the developers, Cotype Light is among the best LLM in its class: it contains 8 billion parameters. If necessary, MTS AI can create LLM in Tatar with a large number of parameters - up to 70 billion parameters, as well as a large context window of up to 32 thousand tokens - so that the model can perform tasks such as translation and generation of long texts. Also, MTS AI is ready to adapt models of the Cotype family for other regional languages ​ ​ of Russia.

Ability to process a long user context

MTS AI has released an updated version of its large Cotype PRO business language model. This model can handle a long user context - up to 20 pages, which allows you to produce personalized and accurate responses when computing power is low. The company announced this on August 28, 2024.

Cotype is a large language model created by MTS AI specifically for working with corporate data. It is trained on a large amount of business correspondence, job descriptions, documentation and other texts, which provides high LLM expertise in this area and allows the use of AI in business processes. Cotype is in the top 3 Russian-language models, according to the MERA benchmark.

An increased amount of context storage without data loss is the Cotype Pro function, which is used in more than 10 pilot projects of the corporate sector and government agencies to create an end-to-end search system for internal documentation and databases, generate technical instructions and corporate letters, SEO-optimization of materials on the site and analyze and summarize the results of meetings - said Sergey Ponomarenko, senior manager of LLM products of MTS AI.

Cotype PRO was created using a unique two-stage method of further training and its own benchmarks developed by MTS AI specialists. This approach has achieved high quality performance for a large language model that runs on just one NVIDIA A100 graphics card, unlike competitive solutions that need 4 graphics cards.

A large context window gives the Cotype PRO language model the ability to process a significant amount of information at a time. That is, the user can download a 20-page contract or other document. Thanks to which, the model has a better understanding of what it is about and gives more accurate answers, which is extremely important when it comes to analyzing corporate documents and regulations.