XNews

ChatGPT Desktop Adds Voice Control via GPT-Live

ChatGPT Desktop Adds Voice Control via GPT-Live

OpenAI added voice control to the ChatGPT desktop app (formerly Codex App) on macOS and Windows, powered by the GPT-Live voice model launched on July 8. GPT-Live features a full-duplex architecture, enabling simultaneous listening and speaking without turn-taking pauses. Users can speak to control the computer, launch Codex for coding, schedule tasks via ChatGPT Work, and manage multiple background agents in one voice session. Complex reasoning tasks are offloaded to a GPT-5.5 backend so the conversation remains fluid. macOS users get Appshots, allowing the voice session to see the frontmost window for contextual responses; Windows lacks this feature. iPhone users can remotely pair iOS ChatGPT voice to control the desktop Codex. Android support is planned. The feature is available globally to Plus, Pro, Business, Edu, and Enterprise subscribers, but not free users. Early demonstrations show users watching YouTube together with Codex, performing vibe coding entirely by voice via AirPods, and projecting the desktop to a TV for ambient interaction.

Key insights

GPT-Live's Full-Duplex Architecture Enables Natural, Uninterrupted Voice Interaction

Unlike previous voice models that required turn-taking, GPT-Live can listen and speak simultaneously, making conversations feel more natural. Offloading heavy reasoning to GPT-5.5 prevents the voice interaction from stalling. This design is critical for real-time computer control, where users need immediate feedback while issuing commands.

Voice Control Transforms the Desktop into an Ambient Computing Platform

With voice control and screen context, users can interact with their computer from anywhere in the room, using AirPods or a paired iPhone. This enables hands-free coding, media consumption, and task management, blurring the line between traditional desktop use and ambient AI assistants. The ability to project the screen and control it by voice suggests a shift toward more flexible, less keyboard-dependent workflows.

Source posts