OpenAI’s Latest Voice Model Won’t Interrupt You When You Pause to Think

This week, OpenAI introduced GPT-Live, a line of voice models it claims can hold natural, real-time conversations. The models can listen to you and speak at the same time—perhaps the most natural conversational element of all—allowing you and any other human conversationalists to interrupt, change topics, or ask follow‑up questions.

But GPT-Live won’t interrupt. Often, those who used OpenAI’s previous voice models had to fill their thinking pauses with “um” or “hmm” so the model wouldn’t think they’d finished their query. Though OpenAI attempted to fix this with an update to ChatGPT’s Advanced Voice Mode last year, GPT-Live can reportedly “make interaction decisions many times per second: whether to speak, continue listening, pause, interrupt, or invoke a tool.” OpenAI says this results in smoother pacing, fewer awkward cut‑offs, and more human‑like responses during extended conversations.

When a user’s question is fairly complex, GPT-Live passes it to a more advanced model (GPT-5.5) and then shares the answer back in the conversation.

OpenAI’s system card says GPT‑Live‑1 clearly outperforms the older voice mode on reasoning and search benchmarks, as well as in human evaluations focused on comfort, interruptions, and flow.

GPT‑Live is called GPT‑Live‑1 for paid plans and GPT‑Live‑1 mini for free users. OpenAI is rolling it out across iOS, Android, and the ChatGPT website as the new standard for voice chat, with the nine ChatGPT voices remastered for GPT‑Live.

Finally, GPT-Live incorporates certain safety guardrails, like referring users to crisis resources when self-harm is mentioned and allowing parents to control whether teens can use voice features.

Disclosure: Ziff Davis, ExtremeTech’s parent company, filed a lawsuit against OpenAI in April 2025, alleging it infringed Ziff Davis’s copyrights in training and operating its AI systems.