OpenAI launched GPT-Live yesterday, a full-duplex speech model for ChatGPT Voice. The new solution can respond while listening and is easy to interrupt. It also allows for rapid context switching. Processing time is filled with a 'mhmm' or other interjection.
The previous version contained serial conversions of speech-to-text, LLM, and text-to-speech, which, according to analyses, increased the latency to nearly two seconds. GPT-Live delegates more complex questions to a frontier model, at the launch of GPT-5.5, for searches and reasoning, while the conversation continues uninterrupted.
TechCrunch reports that OpenAI views voice as the future primary interface for more complex, long-running agential work. The current rollout of the voice solution is aimed at consumers, but it is also becoming the standard for paying Go, Plus, and Pro users. GPT-Live-1 mini can be used without an account on iOS, Android, and the web.
Ziptone asked GPT Live a few questions about latency:
(Ziptone/editors)
Also interesting
-
Every healthcare institution has customer contact. Now there is also a new conference specifically for customer contact in…
-
The transition from chatbot to agent remains an Achilles' heel of customer contact. Ronald Rubens wants this…
-
With growing attention to strategic autonomy, the on-premise solution for the contact center is also…