The Yandex B2B Tech team has introduced a new generation of speech models for creating voice AI agents. The updates will allow companies to automate more customer inquiries and reduce the need for operator involvement.
The new models can maintain a natural dialogue, change intonation depending on the situation, respond to interruptions, and switch between topics. In addition, the agents can perform several related actions within a single conversation, accessing corporate systems in real time.
In particular, the updated Speech Realtime Max allows handling multi-step requests. For example, a fitness club’s voice assistant rescheduled classes for a client and her daughter in three minutes, applied a discount, offered a family membership, calculated the surcharge, and sent a payment link.
The Speech TTS Live model is responsible for natural-sounding speech: it independently places pauses, stresses, and selects intonations. For different scenarios, companies can choose voices and communication styles—from businesslike to softer.
Agents can be created and configured using text prompts in Yandex AI Studio, connecting them to internal systems without developing a prototype from scratch.
According to Gartner’s forecast, by 2030, more than 70% of enterprise voice interactions will be handled automatically. Speech models capable of reasoning will be able to autonomously resolve up to 80% of incoming inquiries.