gpt-realtime and Realtime API general availability
The Realtime API goes generally available with gpt-realtime, a speech-to-speech model, plus remote MCP, image input and SIP phone calling.
gpt-realtime is a single speech-to-speech model with better instruction following and tool calling, new voices Marin and Cedar, and 20% lower prices than the preview model. Adds remote MCP servers, image input and SIP telephony for production voice agents.
- Date
- Thursday, 28 August 2025
- Lab
- OpenAI
- Kind
- model
- Access
- closed API
- Price
- $32 input / $64 output per 1M audio tokens (2025-08-28)
Figures
| Measure | Value | Measured by |
|---|---|---|
| MultiChallenge audio (instruction following) | 30.5% previous December 2024 model scored lower | company |
| ComplexFuncBench audio (function calling) | 66.5% | company |
| API price (audio input / output per 1M tokens) | $32 / $64 20% below gpt-4o-realtime-preview | company |
The older Realtime API beta was removed 2026-05-12; the model line continued with gpt-realtime-1.5 (2026-02-23), gpt-realtime-2 (2026-05-07) and gpt-realtime-2.1 (2026-07-06) per the changelog.
Sources
This record was checked against its sources on 6 October 2026. How we check