GPT-Realtime-2
last releaseMay 2026
goblin vibe check:
use this when voice needs to feel like a loop, not a three-step loading bar
openai's realtime speech model for low-latency voice agents, live conversation, and multimodal interaction. it is a practical model-level tool for building npc dialogue tests, support agents, tutors, and interactive voice interfaces without stitching separate stt, llm, and tts pieces together.
single realtime model path for speech-to-speech voice agentsdesigned for low-latency interactive conversationsupports tool use and multimodal agent workflows through the realtime apiuseful for live npc tests, coaching, support, and voice interface prototyping
key features
single realtime model path for speech-to-speech voice agentsdesigned for low-latency interactive conversationsupports tool use and multimodal agent workflows through the realtime apiuseful for live npc tests, coaching, support, and voice interface prototyping
spec & usage
realtime model update surfaced in May 2026 release tracking
available through OpenAI's Realtime API documentation and platform stack
requires hosted API usage rather than local deployment
limitations
public release details are spread across platform docs and release notes rather than one deep launch post
not the right choice when you need local/offline voice inference
scope:
audiolanguagemodelvoicechatbotapicloudpaidreal-time
launchMay 2026
last releaseMay 2026