| The name of the base system (platform): | Artificial intelligence (AI, Artificial intelligence, AI) |
| Developers: | Yandex |
| Last Release Date: | 2026/03/03 |
| Branches: | Information Technology |
| Technology: | Application Development Tools |
The Yandex AI Studio platform allows you to create AI applications and agents to solve business problems, combines all the technologies, models and tools necessary for this (for November 2025).
2026: Ability to create reasoning agents
Yandex B2B Tech on March 3, 2026 introduced a large update to Yandex AI Studio, an AI application development platform. Now with its help, business can create reasoning agents that can consistently execute complex orders. Security tools have appeared on the platform: for example, you can configure the agent to transfer data to the cloud model over the corporate network and without logging requests. The use of AI agents in a number of scenarios will become 4 or more times cheaper due to the new charging model.
Yandex AI Studio gave the business the ability to build AI agents using the DeepSeek-V3.2 model. Among her advantages are improved reasoning skills: she is able to hold a long context and draw consistent conclusions. In addition, AI agents based on it can simultaneously think and work with external applications - this significantly increases the speed of responses.
Reasoning agents are well suited for complex multi-composition tasks. For example, they can compare vendor bids or assist in the analysis of technical incidents. Another area of their application is working with code: they can be instructed to run autotests, create a repository, deploy applications from it, and much more. This model has already entered the ensemble of neural networks used by SourceCraft CodeAssistant, an assistant built into the SourceCraft development platform created by Yandex.
Using DeepSeek-V3.2 on a cloud platform is often easier, cheaper, and safer than deploying your infrastructure yourself. Yandex AI Studio is adapted to the specifics of business requirements, including large ones: data processing using generative models takes place in Russia, while the platform meets all domestic and international security standards.
For large businesses, such as banks and industrial companies, Yandex AI Studio has offered tools that allow you to work with artificial intelligence in the cloud in compliance with specific information security requirements.
In particular, you can now connect to Yandex AI Studio over a dedicated secure connection. This allows you to use AI agents inside the corporate network, without accessing the Internet. Requests are not logged in any way, which means that all information remains within the company.
Yandex AI Studio became fully available in on-premieses format. Customers were able to deploy the platform on their infrastructure or work with neural networks using a hybrid model: process confidential data within the company, and solve part of resource-intensive AI problems in the cloud through a secure connection.
Yandex AI Studio will add two kinds of tokens. The first is cached tokens: for cases where part of the information is reused. The second is tool tokens: for cases where data is transferred to the model as a result of a call to external services and applications. Now it will be easier to track what makes up the price, and customers will be able to optimize costs on large requests - for example, using file or web search. For a number of neural networks, tool and caching tokens will cost 4 times cheaper than incoming and outgoing tokens.
| We not only added capabilities to Yandex AI Studio, but also finalized the entire platform infrastructure. Now it works better with a long context, supports multi-step and takes into account repeated calls to the model. All this ensures the fast and uninterrupted operation of AI agents even under high loads. The updated infrastructure and new tariffs will allow businesses to significantly reduce the final cost of using AI agents. According to our estimates, in a number of scenarios, costs can be reduced by 4 times or more, "said Artur Samigullin, head of Yandex AI Studio. |
