GPT-Live Codex voice control arrives in ChatGPT desktop app

OpenAI has added GPT-Live Codex voice control to the ChatGPT desktop app on macOS and Windows, bringing its full-duplex audio model into Codex and ChatGPT Work. The move matters for engineering teams because OpenAI is positioning voice as a control layer for agentic coding tasks, including pull request review, debugging and concurrent build work, rather than a separate chat feature.

OpenAI announced that GPT-Live now powers the ChatGPT desktop application and works with Codex and ChatGPT Work, which are separate experiences inside the desktop app. An OpenAI spokesperson told VentureBeat this is the first time voice activation has been built natively into Codex on desktop. Pricing details were not newly disclosed for this release, but access is limited to paid Plus, Pro, Business, Enterprise and Education subscribers.

GPT-Live was introduced on July 8, 2026 as a continuous audio model that can listen and speak at the same time. OpenAI has said the model handles the live conversation while passing more complex reasoning to background models, including GPT-5.5. In practice, that means the voice layer can acknowledge a user, keep listening and invoke tools while other agents work through coding tasks.

What can developers do with GPT-Live in Codex?

OpenAI says developers can use spoken instructions to start and direct several Codex or ChatGPT Work tasks at once. Examples cited include asking the system to investigate an authentication bug, review a pending API migration pull request and create missing unit tests from one spoken prompt.

The desktop app can coordinate work across contexts such as Slack conversations, GitHub repositories and local codebases, according to OpenAI. The company also says developers can turn design mockups into working code by splitting work across frontend, backend and testing tasks.

On macOS, the release uses Appshots and screen context so ChatGPT Voice can inspect the frontmost window along with local files, codebase structure and active plugins. OpenAI says the system can maintain conversational state while background agents handle code changes asynchronously.

Desktop, mobile and project support

The release also ties into multi-folder project support in build 26.715 and remote execution through iOS. OpenAI says engineers can check progress, respond to agent prompts and redirect active jobs without managing each process manually in the desktop interface.

OpenAI published a promotional video showing Codex developer experience engineer Jason Liu and Codex technical staffer Guinness Chen speaking to the same ChatGPT desktop app session in the same room. In the demo, both issue instructions to the same model, which is meant to show multi-person, voice-led interaction with one coding agent session.

Codex has more than 5 million weekly active users, according to OpenAI. The company has expanded the Codex brand beyond coding models into a broader desktop productivity system this year, including app control, image generation and webpage preview features.

What OpenAI is keeping closed

The voice-enabled desktop release remains a proprietary commercial product. OpenAI is not making the model weights, voice processing systems or agent state architecture available for modification or self-hosting.

Voice-triggered jobs also use the same allocations as existing Codex and ChatGPT Work activity, according to the release details. That makes the new interface a different way to spend the same paid-plan capacity, rather than a separate usage pool.

Reaction among AI developers quickly centered on the move toward hands-free agent orchestration. AI Insider journalist ChrisGPT wrote on X that the release of voice and remote guidance for Codex was “one step closer to personal AGI.” OpenAI has not said how it will measure reliability for multi-threaded voice-directed coding work, which is the question that will matter most for engineering teams considering it beyond demos.

This story draws on original reporting from VentureBeat.