Introducing gpt-realtime and Realtime API updates for production voice agents

OpenAI has made the Realtime API generally available for building production voice agents.

It also launched gpt-realtime, a new speech-to-speech model designed for more natural, low-latency conversations.

The update adds remote MCP server access, image inputs, and SIP-based phone calling, plus new Cedar and Marin voices.

OpenAI says the model improves instruction following, tool use, multilingual handling, and audio reasoning over its prior voice model.