OpenAI Adds ChatGPT Voice to Desktop Work and Codex, Codex Weekly Active Users Top 10 Million

On July 23, OpenAI officially introduced ChatGPT voice functionality to the Work and Codex scenarios in the desktop app. Starting that day, it is being rolled out globally in batches on macOS and Windows, covering Plus, Pro, Business, Edu, and Enterprise tiers. Android support will follow soon, while iOS can access it remotely via Remote pairing. The feature is based on the new-generation voice model GPT-Live, launched on July 8, which supports simultaneous listening and speaking. Users can interrupt at any time, and the model continues the conversation from the latest sentence.

In Work and Codex, voice can be used to start tasks, query task status, view agent status, and coordinate multiple agents within the same conversation. Tasks run in the background, and users can insert instructions at any time to adjust direction. When a task gets stuck or is completed, the system proactively alerts the user via voice or on-screen prompts. Voice can also call upon existing project context, connecting to documents, calendars, contacts, and communication records to continue unfinished work.

The macOS version additionally supports Appshots and screen context, allowing it to directly read the frontmost window, combined with local files and codebase understanding. Hotkeys are customizable. The combined weekly active users of Codex and ChatGPT Work has exceeded 10 million. In an official demonstration, multiple engineers stood in the same room and gave different instructions to the same desktop session, enabling multiple users to command a single AI simultaneously.

Voice input speed is about 240 words per minute, while typing speed is about 70 to 80 words per minute. GPT-Live offloads heavy tasks to the background GPT-5.5 and GPT-5.6 models, with the front end handling only smooth conversation. OpenAI recently integrated Chat, Work, and Codex into a single entry point on the desktop, supporting one-click switching and long-running background operation. Codex has been expanded into a general-purpose platform that can run multiple agents in parallel, execute long tasks, and read entire projects.