ChatGPT Voice Expands to Desktop With GPT-Live Powered Conversations
The desktop expansion turns ChatGPT Voice from a conversational tool into a hands-free interface for agentic tasks, letting users direct Codex and Work by speech while the assistant maintains context across background operations.
Key Facts
- ChatGPT Voice is now available on Mac and Windows desktop apps with build 26.715 or later.
- The feature is powered by GPT-Live, a full-duplex voice model announced on July 8 that processes listening and speaking simultaneously.
- Desktop ChatGPT Voice supports voice commands for Codex and Work, allowing users to start tasks and check progress while continuing conversation.
- The desktop app is available to Plus, Pro, Business, Edu, and Enterprise plans, but not to Free or Go plans.
- OpenAI also released GPT Transcribe at $0.0045 per minute and GPT Live Transcribe at $0.017 per minute for transcription tasks.
Reporting from 2 sources: GIGAZINE, GameBusiness.jp.
OpenAI has rolled out ChatGPT Voice, its voice conversation feature, to desktop applications on Mac and Windows. The feature is built on GPT-Live, a new generation voice model announced on July 8 that processes listening and speaking simultaneously in a full-duplex manner. This allows the assistant to interject with backchannel responses like nods and murmurs, and to handle interruptions naturally during conversation. Beyond standard chat, users can now issue voice commands to Codex and Work, such as starting tasks or checking progress, while continuing the conversation in the background. On Mac, the Appshots feature lets the assistant reference the focused window, and it integrates with Computer Use, local files, and ChatGPT plugins. The desktop app is available globally with build 26.715 or later, and is offered to Plus, Pro, Business, Edu, and Enterprise plans, but not to Free or Go plans. Remote access from the iOS app also enables voice control of Codex. The rollout follows a similar update from Anthropic, which refreshed Claude's voice mode for Opus and Sonnet, though Claude retains a turn-based architecture rather than GPT-Live's parallel processing.
- GPT-Live is a single model that processes input and output in parallel, replacing the older pipeline where separate models handled speech recognition, text response, and speech synthesis.
- When a task needs search, reasoning, or agentic work, the conversation is handed to a text model such as GPT-5.5 while the voice conversation continues in the background.
- OpenAI also demonstrated live translation and the ability to stay silent for long stretches while preserving context.
- Users can speak commands like "start a Codex task and run tests," and the work proceeds in the background while the conversation goes on.
- The Mac Appshots feature lets the assistant reference the focused window, and it works with Computer Use, local files, and ChatGPT plugins.
- Remote access from the iOS app enables voice control of Codex.
- Anthropic refreshed Claude's voice mode for Opus and Sonnet around the same time, but Claude keeps a turn-based architecture that processes listening, thinking, and speaking in sequence, not in parallel like GPT-Live.
Synthesized by Yomimono from the 2 cited sources below, including Japanese-language reporting where cited, then editorially reviewed before publishing.