Model overview
gpt-realtime is OpenAI's multimodal model for real-time voice interactions and its first generally available Realtime model. It directly processes and generates audio, aiming to help developers build low-latency voice agents without chaining together speech recognition and speech synthesis models. On August 28, 2025, OpenAI released the model and the generally available Realtime API; the corresponding fixed snapshot is gpt-realtime-2025-08-28. (openai.com)
Capabilities and uses
- Supports text and audio input and output, as well as image input; real-time interaction is available through WebRTC, WebSocket, or SIP connections. (developers.openai.com)
- Supports function calling and can be used to build applications with features such as Realtime API support for remote MCP servers and SIP telephony access. Officially listed uses include customer support, personal assistants, and education. (openai.com)
- OpenAI describes improvements in complex instruction following, tool calling, natural speech expression, and language switching. (openai.com)
Limitations and status
- The context window is 32,000 tokens, and the maximum output is 4,096 tokens; the knowledge cutoff date is October 1, 2023. (developers.openai.com)
- Does not support video, image output, structured outputs, or fine-tuning. (developers.openai.com)
- As of November 18, 2026, OpenAI had marked
gpt-realtimeand the fixed snapshot as deprecated and announced plans to shut them down on January 20, 2027; the officially recommended replacement model isgpt-realtime-2.1. (developers.openai.com)
API pricing
Prices listed in OpenAI's model documentation are in US dollars (USD) and are all per 1 million tokens; cached input prices apply only to cached inputs. Date checked: November 18, 2026. (developers.openai.com)
| Modality | Input | Cached input | Output |
|---|---|---|---|
| Text | $4.00 | $0.40 | $16.00 |
| Audio | $32.00 | $0.40 | $64.00 |
| Image | $5.00 | $0.50 | Not supported |