OpenAI expands Realtime API for voice agents

The big shift here is that voice is no longer being framed as a novelty layer on top of text models; OpenAI is positioning it as a first-class agent runtime.

• –Production readiness matters more than raw demo quality here: interruptions, tool calling, and stateful conversation flow are the real bottlenecks for voice apps.
• –The translation model broadens the use case from English-first assistants to global, real-time communication products.
• –This is a strong signal that OpenAI sees voice agents as a deployment category, not just an interface experiment.
• –The main question is whether developers will adopt the Realtime stack directly or keep using orchestration layers around it for reliability and control.