MTS MWS GPT

The main articles are:

• Speech Recognition (Technology, Market)
• Speech technology: On the path from recognition to understanding
• SaaS - History. Philosophy. Development drivers

2025

Over 40 different models available

MTS Web Services (MWS) June 4, 2025 announced a significant expansion of the functionality of cloudy the b2b platform for working with. artificial intelligence Now more than 40 different models are available in MWS GPT, including leading Open source LLM in different sizes.

Increasing the number of available LLMs gives customers access to a single environment to compare and test models, accelerating the adoption of AI into business processes. Now companies can immediately assess which model copes better with specific cases: document analysis, report generation, customer support, or automation of routine operations.

Speech-to-text models allow you to create voice assistants, automate audio decryption, and receive conversational analytics.

One of the key changes is support for multi-agent architectures in the low-code constructor, which opens up opportunities for automating multitasking scenarios. Users can customize complex scripts where a single orchestrator agent manages several highly specialized assistants, each of whom performs their function using connected instruments.

In addition, ChatUI introduced RAG (from the English Retrieval-Augmented Generation, a method in the field of artificial intelligence that combines generative models with search mechanisms from external data sources) and Vision (this is the model's ability to understand and process images, not just text). Users can now upload documents and images, and AI will analyze their contents and squeeze long texts, answer questions on instructions or describe graphic materials.

We see business increasingly adopting generative AI, but many companies face the challenge of choosing the right model and the complexity of integration. With the MWS GPT update, we give customers a ready-made tool with a large selection of LLMs, multi-agent capabilities and simple low-code development. Now you don't need to spend months on tests and customization - you can immediately deploy AI assistants, voice interfaces and complex automated workflows, saving time and resources, "commented MWS CEO Pavel Voronin.

For June 2025, FineTune is available on the platform to a limited number of customers - additional training of a pre-trained language model on a specific set in data order to adapt it to specific user tasks or improve the quality of work in a specific area, in the future it will be available to everyone. In the near future, functionality for working with images will also appear.

The MWS GPT platform is available both in the cloud (with fast scaling on the MWS GPU infrastructure) and on-premium. MWS provides full platform deployment support.

Running MWS GPT

MTS, a digital ecosystem, on February 5, 2025 announced the launch of MWS GPT, a cloud B2B platform that unites the company's large language models. With the help of the service, small, medium and large businesses will be able to simultaneously use different LLMs, as well as work with any Open source models, learn them more, use function calling functionality, and then implement and use ready-made solutions on the ecosystem infrastructure.

Inside the platform there is its own chat interface, which allows you to save prompts and exchange them, as well as a No-Code designer for designing and creating applications without involving developers. At the first stage, you can connect the cloud version of MWS GPT, in the future it will be possible to deploy the service on the customer's infrastructure. At the next stage, it is planned to increase the number of available models, including with the help of popular products of other companies on the market.

To achieve good results, it is important to constantly experiment. It was for this purpose that we launched the LLM aggregator inside MTS. As of February 2025, we record over 1.2 million calls to the platform per day. Having made sure of the efficiency and reliability of the service, we decided to enter the foreign market. Working with different models, our customers will be able to choose the most suitable, without issuing numerous subscriptions and optimizing the cost of technical support. In my opinion, in order to get the best result in solving problems using LLM, you do not need to focus on one. MWS GPT is based on an agent architecture that allows any model to be an AI Agent. In addition to the usual text response, MWS GPT decides to use this or that skill depending on the task. For example, in the case of mathematical calculations, the model refers directly to the interpreter, instead of trying to continue the text based on the training corpus, and if the training data is not enough for a complete answer, it makes a request to the search engine, - commented Pavel Voronin, First Vice President for MTS Technologies.

Payment for using the service will be charged depending on the number of tokens processed, which reduces the entry threshold for small businesses.