GPT-Realtime
Realtime speech-to-speech model for low-latency voice agents over WebRTC, WebSocket, or SIP
- Price per 1M tokens
- $4 / $16 input / output · OpenAI
- Context window
- 32K tokens in · 4.1K out
The GPT-Realtime-2.1 line
4 releases in the GPT-Realtime-2.1 line.
GPT-RealtimeAug 28, 2025179dGPT-Realtime-1.5Feb 23, 202673dGPT-Realtime-2May 7, 202660dGPT-Realtime-2.1Jul 6, 2026
A newer release is out: GPT-Realtime-2.1.
BENCHMARKS
Not in the Epoch Capabilities Index. Epoch benchmarks what it can run, so this is a gap in coverage, not a poor result.
OpenAI's past year
23 release days in the past year, about one every 16 days. The next is expected around Nov 5, 2026. Forecasts ↗
NovDecJanFebMarAprMayJunJulAugSepOctNov
Specifications
RELEASE DATEAug 28, 2025
WEIGHTSClosed weights
CONTEXT WINDOW32K
MAX OUTPUT4.1K
REASONINGNot listed
TOOL CALLINGNot listed
INPUTStext, audio/pcm, audio/pcmu, audio/pcma, image/jpeg, image/png, image/gif, image/webp
OUTPUTStext, audio/pcm, audio/pcmu, audio/pcma
Specifications and prices from Models.dev; availability and limits may differ by provider. Benchmarks from Epoch AI (CC BY 4.0), LiveBench and the other sources named above.