Skip to content
NewsProduct launch

OpenAI brings GPT-Live full-duplex voice to ChatGPT desktop on macOS and Windows

· by Pondero Newsdesk

The short version

ChatGPT desktop gains voice control over Codex and ChatGPT Work threads starting July 23, with Appshots screen context on macOS, available to paid subscribers on Plus, Pro, Business, Enterprise, and Education plans.

OpenAI brings GPT-Live full-duplex voice to ChatGPT desktop on macOS and Windows

Developers can now voice-command running Codex threads, pull requests, and ChatGPT Work tasks from the desktop. OpenAI pushed GPT-Live into its ChatGPT desktop application on July 23, 2026, giving paid subscribers a live voice layer over the same agentic systems that previously required keyboard input only.

What changed

GPT-Live, OpenAI's full-duplex voice model that arrived on smartphones on July 8, now powers ChatGPT Voice inside the desktop app, per OpenAI's release notes. Full-duplex means the model speaks and listens simultaneously, handling interruptions without forcing rigid turn-taking. Background reasoning stays delegated to GPT-5.5 while GPT-Live manages the audio layer.

The feature works across Chat, Work, and Codex threads. A developer can open a voice session and instruct the app to start a new Codex task, check the status of a running job, or steer an active pull request, without switching to a text prompt.

On macOS, Appshots integration adds screen context to the voice session. ChatGPT Voice can read the frontmost window, local files, and codebase structures alongside active plugins, per 9to5Mac's coverage of the release notes. Windows users get full voice control of agents but without the screen-reading layer. A multi-folder project update also ships in build 26.715, per VentureBeat.

Access is limited to Plus, Pro, Business, Enterprise, and Education plan subscribers. Free-tier accounts are excluded.

Why it matters

The substantive shift is that Codex workflows no longer require direct keyboard interaction between tasks. A developer can walk through a bug verbally while an agent runs fixes in a separate thread, redirect that thread when the output is wrong, and confirm status without dropping focus on other work. In a demo posted by OpenAI, a developer issued a single spoken command to create a new thread, file a pull request, and identify a bug root cause, per TechCrunch.

Codex and ChatGPT Work together had more than 10 million weekly active users as of the announcement, per VentureBeat citing OpenAI figures. Voice control lowers the friction for kicking off and monitoring tasks, which could accelerate compute consumption for that user base. Users on iOS can also connect to a running desktop Codex session remotely through ChatGPT's Remote feature.

The release represents OpenAI's clearest push toward voice-driven developer workflows. Existing keyboard-centric tools such as Cursor and GitHub Copilot do not currently offer a voice layer for agent orchestration. Whether the experience holds up under the latency and error conditions of real coding sessions will determine how widely developers adopt it.

What to watch next

IDE integration is the open question. The voice layer today lives inside the ChatGPT desktop app, not inside VS Code, Cursor, or other editors. If OpenAI extends it to IDE plugins, it competes on the home ground of keyboard-native tools. OpenAI has not announced a timeline for that step. It is also worth watching whether Appshots screen-reading reaches Windows before Q3 closes, given that the macOS-only restriction limits the feature for teams on mixed platforms.

Sources