release

Introducing gpt-realtime and Realtime API updates

These updates streamline the creation of real-time, multimodal voice agents that can access external data and handle phone calls, opening new possibilities for automated customer service and voice-driven applications.

OpenAI Blog·August 28, 2025·1 min readrelease

releaseIntroducing gpt-realtime and Realtime API updates

openai.com

What happened

OpenAI has announced an upgraded speech-to-speech model and several new features for its Realtime API, according to the OpenAI Blog. The updated model delivers more natural and responsive voice interactions, while the API now supports MCP (Model Context Protocol) servers, enabling real-time data source integration. Additionally, developers can input images alongside audio, allowing multimodal queries within voice conversations. The API also now supports SIP (Session Initiation Protocol) phone calling, making it possible to build voice agents that can make and receive phone calls. For developers building AI workflows, these updates reduce friction in creating real-time voice applications with access to external tools and visual context. The practical angle is that integrating voice, vision, and data sources into a single API call simplifies the development of conversational AI for customer support, virtual assistants, and phone-based services.

Key takeaways

OpenAI released an improved speech-to-speech model for more natural voice interactions.
Realtime API now supports MCP server integration for real-time data access.
Developers can provide image inputs alongside audio for multimodal voice queries.
SIP phone calling support enables voice agents to make and receive phone calls.
All features are available via the Realtime API, reducing the need for separate audio pipelines.