ChatGPT Desktop App Gets a Voice That Takes Action

OpenAI has rolled out a major update to its ChatGPT desktop application, adding a voice mode that lets users talk directly to the AI to control agents and perform tasks on their computer. The feature, announced Thursday, taps into the company’s new ChatGPT-Live voice model family, which debuted earlier this month.

Unlike the smartphone version, which offered smoother conversations and better interruption handling but couldn’t act on the device, this desktop upgrade is built for action. Users can dictate complex, multi-step commands—like creating a new thread, making a pull request, or finding a bug’s root cause—and ChatGPT will respond when it needs input. On macOS, the app can also access on-screen content through Appshots, including alt-text, and it works with both ChatGPT Work and Codex to look up websites and apps.

In a demo, OpenAI showed a developer using a single voice command to trigger a chain of coding tasks. The company also noted that users can access Codex via voice from the iOS app through remote control. Meanwhile, rival Anthropic has updated its own voice mode for Claude, which now taps its Opus, Sonnet, and Haiku models to complete tasks inside apps like Gmail, Calendar, Slack, Notion, and Canva.