According to MTS, companies in Russia will for the first time have access to the Qwen3-235B (235 billion parameters), Kimi K2 (1 trillion parameters) and GLM-4.6 (357 billion parameters) models. It is planned that other models of comparable class will be added to the platform in the near future.
These models can be used in various scenarios – from analyzing documentation to generating code and preparing reports. Ultra-large models are also suitable for tasks where accuracy of analysis and quality of decisions are critical: legal expertise, financial calculations, technical support, working with documents and software code development.
The level of factual and contextual errors (errors in understanding the situation, conditions, or relationships between data) in models of this class is 40–60% lower than that of smaller LLMs. This result is achieved through more accurate data processing and self-checking mechanisms, which allow the system to check its conclusions several times, reducing the risk of errors and "hallucinations."
As a reminder, MWS GPT is a B2B platform that allows you to work with different large language models simultaneously in a single window and, if necessary, switch from one to another. More than 50 LLMs are available on the platform. MWS GPT includes its own chat interface with the ability to save and share prompts, as well as tools for designing and creating applications without involving developers. The service is available both in the cloud (with fast scaling on the MWS GPU infrastructure) and in on-premise format (on the customer's infrastructure).