gpt-realtime at a glance
The gpt-realtime speech-to-speech model improved instruction following, tool calling and natural speech, while the Realtime API gained SIP phone calling, remote MCP servers and image input at 20% lower prices.
What changed with gpt-realtime
On August 28, 2025, OpenAI made its Realtime API generally available and released gpt-realtime, its most advanced speech-to-speech model at the time. OpenAI said it follows complex instructions better, calls tools more precisely, captures non-verbal cues such as laughter, can switch languages mid-sentence, and sounds more natural. Two new voices, Marin and Cedar, were added.
The API gained remote MCP server support, image input, and phone calling through SIP, plus asynchronous function calling so long-running tools no longer stall a conversation. OpenAI cut gpt-realtime prices by 20% versus gpt-4o-realtime-preview, to $32 per million audio input tokens and $64 per million audio output tokens.
Why gpt-realtime matters for businesses
For businesses, native SIP calling and better tool use made it much simpler to put an AI receptionist or support agent on a real phone line. Call recording disclosures, consent rules and a clear handoff to a human remain essential.
Put it to work: AI Voice Agents
Sources for this story
- Introducing gpt-realtime and Realtime API updates for production voice agents (opens in a new tab)Official
OpenAI · openai.com
Last verified: . IA Compa summarizes public announcements for business readers. We are not affiliated with the companies mentioned. Always confirm details, pricing, and availability with the official source.
- #voice-agents
- #speech-to-speech
- #openai
- #realtime-api
- #mcp
