OpenAI launches GPT-Live voice model

The launch of GPT-Live represents a notable step in OpenAI's effort to make voice interaction with AI feel like a real conversation. The previous ChatGPT Advanced Voice Mode improved on the original cascaded system, which chained separate models for speech-to-text, language processing, and text-to-speech, resulting in slow and stilted responses. However, Advanced Voice Mode still operated through discrete turns, leading to rigid back-and-forth and interruptions caused by silence detection errors.

GPT-Live addresses these limitations with two key architectural changes: continuous interaction through a full-duplex architecture and decoupling from deeper work by delegating tasks like search to models such as GPT-5.5. This allows the voice model to keep the conversation flowing while complex tasks are handled in the background. In head-to-head evaluations, GPT-Live-1 and GPT-Live-1 mini were strongly preferred over Advanced Voice Mode in matched 5–10 minute conversations.

The success of this launch may hinge on how well the new model handles real-world, non-English conversations, as OpenAI notes some languages may have a non-native accent or gaps in fluency. Additionally, the coming API availability could be a key factor for developers and enterprises looking to build on the new voice capabilities.