OpenAI turns ChatGPT voice into a control layer for work tasks

Key Points

1. OpenAI extended ChatGPT voice chats to create documents, presentations and handle connected apps and email.
2. Voice mode now supports plugins and connected apps on web, iPhone and Android devices.
3. The update pushes voice from a question-and-answer tool toward a control interface for AI agents.

The latest:

ChatGPT voice conversations can now run multi-step work tasks, from building documents and presentations to handling email and connected apps, under an OpenAI update. According to the company’s ChatGPT release notes, voice mode now supports plugins and connected apps on the web, iPhone and Android, while voice chats reach the Work environment on both web and mobile.

Details:

• The core change: The headline feature is the pairing of voice chats with ChatGPT Work, OpenAI’s environment for long, multi-step tasks that produce finished files and outputs rather than answers inside the app’s general interface. Users can ask by voice for a document, a presentation or a spreadsheet.
• Task continuity: According to OpenAI, a user can trigger a task by voice, act on a connected app, or have ChatGPT use a cloud browser to complete work online. If the voice conversation ends while the task is still running, the app can keep working on it in text mode.
• Connected services: Plugin and connected-app support inside voice chats closes one of the gaps that limited how far users could rely on speech alone. OpenAI said that during a voice conversation, users can ask ChatGPT to use an available plugin, including email, calendar and the team platform Slack.
• The safeguards: Access permissions and the restrictions applied to each app remain unchanged, OpenAI said. Sensitive actions requiring user approval still surface on screen for review and consent, and a spoken approval is not enough on its own to execute them.
• The model stack: The voice update lands alongside OpenAI’s expansion of the GPT-6 family. The company introduced GPT-6 Astra earlier in September, then on 22 September added GPT-6 Sol and GPT-6 Luna to the Work and Codex usage modes. ChatGPT Voice can draw on all three.
• A technical distinction: OpenAI documentation notes a gap between the voice model and the model doing the reasoning. The Live experience itself runs on GPT-Live-1 or GPT-Live-1 mini depending on subscription plan, while Voice can call on other models for tasks needing search or more complex reasoning.
• The hybrid interface: OpenAI also reworked how voice results are displayed, letting users hear a reply while following the written text inside the same conversation. Switching between typing and speaking no longer requires starting a new chat, narrowing the gap between phone and desktop use.
• Tiered access: Not every user gets the same capabilities. OpenAI said Free and Go subscribers can use voice chats with the plugins their plans support, while using voice inside Work requires access to both Voice and Work — available only on the Plus or Pro tiers.
• What Work offers: Work itself carries the broader set of agentic capabilities: generating documents, presentations and spreadsheets, operating connected apps, and using the browser to run multi-step assignments through to a final result rather than a single reply.

Between the lines:

The design reads as a bet that voice becomes a primary interface for AI agents rather than an alternative to typing, especially on phones, where speaking a long instruction is faster than writing one. The on-screen approval requirement for sensitive actions suggests OpenAI is not yet willing to let a spoken command alone move through a user’s email or calendar.

What’s next

Watch whether voice-plus-Work access widens beyond the Plus and Pro tiers, and whether further GPT-6 models join Astra, Sol and Luna in the Work and Codex modes.