OpenAI on Thursday updated its ChatGPT desktop app with voice capabilities that let users control autonomous agents through spoken commands. The feature, rolling out globally on macOS and Windows, connects ChatGPT Voice to both ChatGPT Work and Codex, turning the desktop app into a voice-driven agent orchestration layer.
How It Works
ChatGPT Voice on desktop is powered by OpenAI’s GPT-Live model family, which launched on July 8. GPT-Live models are full-duplex, meaning they can speak and listen simultaneously. Users can interrupt mid-sentence, and agents can ask clarifying questions in real time before executing tasks.
The desktop implementation goes beyond what ChatGPT Voice offered on mobile. On smartphones, voice mode handled conversation but could not take action. The desktop version accepts multi-step commands, according to TechCrunch. In a demo video, OpenAI showed a developer using a single voice command to create a new thread, open a pull request, and locate the root cause of a bug.
ChatGPT Voice also integrates with computer use capabilities, allowing it to look up websites and applications. On macOS, the Appshots feature lets the app read on-screen content, including alt-text.
Availability
The update is available to Plus, Pro, Business, Edu, and Enterprise plan subscribers, OpenAI confirmed on X. Free-tier users do not have access at launch.
The Friction Problem Voice Solves
Agent adoption has been constrained by interface complexity. Directing ChatGPT Work or Codex has required typing detailed prompts, switching between contexts, and monitoring outputs through text-based dashboards. Voice removes that overhead. A sales team can redirect a research agent mid-call. A developer can triage a failing deployment without leaving a terminal session. An ops lead running a war room can dispatch agents to parallel tasks by speaking.
Anthropic is making a similar bet. The company updated Claude’s voice mode on July 23 with support for Opus, Sonnet, and Haiku models, enabling voice-driven task completion in Gmail, Calendar, Slack, Notion, and Canva. The two largest agent platforms converging on voice as the primary control interface within 48 hours of each other signals that text-only agent interaction may be entering its final phase.