GPT-Realtime-2.1
OpenAI · Audio model
GPT-Realtime-2.1 is OpenAI's current reasoning speech-to-speech model for the Realtime API, an update to GPT-Realtime-2 (May 2026). It listens and replies in audio, calls tools, and lets developers tune reasoning effort. A cheaper gpt-realtime-2.1-mini is also available ($10 audio input…
Current
ReleasedJul 6, 2026
Context window128000
Input → outputtext, audio, image → text, audio
API priceText: $4 input / $0.4 cached input / $24 output per 1M tokens; Audio: $32 input / $0.4 cached input / $64 output per 1M tokens; Image: $5 input / $0.5 cached input per 1M tokens
Consumer access—
Best for
Developers building speech-to-speech voice agents that need reasoning and tool calls in the Realtime API.
Overview
GPT-Realtime-2.1 is OpenAI's current reasoning speech-to-speech model for the Realtime API, an update to GPT-Realtime-2 (May 2026). It listens and replies in audio, calls tools, and lets developers tune reasoning effort. A cheaper gpt-realtime-2.1-mini is also available ($10 audio input / $20 audio output per 1M tokens).
Strengths and limits
- Reasoning speech-to-speech model with configurable reasoning effort and tool use
- 128K context window, 32K max output tokens
- Improved alphanumeric recognition, noise/silence handling and interruption behaviour over GPT-Realtime-2
- Audio tokens are expensive ($64 per 1M output) compared with the mini variant ($20)
- No official benchmark scores published for the 2.1 update