·
·
Wednesday, 12 August 2026dailybeirut.com
AI
OpenAI has launched GPT-Live, a new voice mode for ChatGPT that uses full-duplex architecture to enable simultaneous listening and speaking, reducing interruptions and supporting natural back-and-forth dialogue.
ChatGPT’s voice interactions have become significantly more fluid and human-like with the introduction of GPT-Live — a new real-time voice mode built on a full-duplex architecture. Unlike earlier voice systems, GPT-Live can listen and speak at the same time, minimizing premature interruptions and bringing spoken exchanges closer to conversations between people.
The original Voice Mode, released by OpenAI in 2023, relied on three separate components: speech-to-text conversion, language model response generation, and text-to-speech synthesis. Later, Advanced Voice Mode introduced a multimodal model to improve speed and naturalness. GPT-Live replaces both as the default voice option, using what OpenAI describes as a full-duplex structure — allowing continuous, overlapping audio input and output without requiring users to pause before the model begins responding.
GPT-Live introduces vocal feedback cues such as “Mmm” and “Yeah” while users are still speaking — mirroring natural conversational behavior. When a query demands deeper reasoning or broader information retrieval, GPT-Live can internally route processing to more capable models like GPT-5.5, all while maintaining uninterrupted voice interaction. This background model switching occurs transparently to the user.
GPT-Live is available in the ChatGPT mobile app for iPhone and Android, as well as via web browser. To begin, tap the soundwave-shaped microphone icon located at the far right of the message input box. On first use, ChatGPT will request microphone access permission. Once enabled, a floating animated circle appears on screen, signaling that voice input is active. Users may speak naturally, and the voice mode remains engaged even when switching to other apps or locking the device screen.
Users can adjust the voice model’s intelligence level via the settings icon at the top of the screen. Three options are offered:
This intelligence-level customization is not available in the GPT-Live-1 mini version, meaning it is restricted to paid subscription plans.
Subscribers to ChatGPT Pro, Plus, and Go receive access to GPT-Live-1. Free-tier users are limited to GPT-Live-1 mini. GPT-Live has fully superseded the previous Advanced Voice Mode as the default voice interface across all supported platforms.
A defining feature of GPT-Live is its reduced mechanical feel. Instead of waiting for the model to finish speaking before resuming, users may interrupt or continue talking seamlessly. The model is also less likely to misinterpret brief pauses in speech as sentence endings — resulting in smoother, longer-form spoken interactions.
GPT-Live’s responsiveness makes it suitable for live translation during spoken conversations. During voice chat, users can scroll down to view the written transcript of both their own speech and ChatGPT’s spoken responses. If the model determines a visual element would aid understanding, it may generate an interactive widget alongside the audio reply.
GPT-Live initiates every session at the Instant intelligence level by default. While this prioritizes quick answers for routine questions, the system automatically delegates more demanding tasks — such as multi-step reasoning or extensive search — to more powerful underlying models. For users regularly engaging with intricate subjects, switching to Medium or High levels extends processing time by one or two seconds, yet preserves conversational fluency.
GPT-Live does not currently support screen sharing or video sharing. Users dissatisfied with the new voice mode may revert to the prior Advanced Voice model through ChatGPT’s settings menu under Voice > Model Selection. From the same menu, users may also change ChatGPT’s voice, select language preferences, and set Voice Mode as the default interface upon launching the application.
