How ChatGPT Voice Works With Connected Apps and Work

ChatGPT Voice is becoming more than a way to ask questions aloud. OpenAI’s latest rollout connects live voice interaction to authorized apps and to ChatGPT Work, allowing users to initiate and steer more involved tasks across web and mobile. The practical distinction is simple: Voice is the interface for speaking with ChatGPT; Work is the environment for carrying out longer, tool-using tasks. 721

What users can do with Voice

In a standard Chat conversation, Voice supports natural, real-time discussion and can use the plugins and connected apps available to an account. That can include connected email, calendar, and Slack tools, subject to the permissions already granted to those integrations. 723

In ChatGPT Work, Voice can be used to ask ChatGPT to create documents, presentations, and spreadsheets, use connected apps, or work in a browser. OpenAI describes Voice in Work as available on web and mobile; its Work documentation also lists support in the desktop app for Work and Codex. 2123

For paid subscribers, reported mobile examples include creating a document, drafting an email, and summarizing Slack messages from the Work tab. 5

Voice does not replace text—it hands work off to it

The new experience is not limited to an audio-only session. Connected-app responses can be followed in the written chat, and an unfinished Work task can continue in text after a Voice call ends. 23

That continuity matters in practice. A user can begin a request by voice on a phone, then review, edit, or continue the resulting task in text rather than having to repeat the request. OpenAI’s documentation confirms the voice-to-text continuation; the exact experience still depends on the user’s device, plan, and enabled tools. 2123

The model roles: GPT-Live, Astra, Sol, and Luna

Several model names are involved, but they serve different roles.

• GPT-Live is the real-time voice layer. GPT-Live-1 is designed for full-duplex interaction, meaning it can listen and speak in a more natural conversational flow. OpenAI says it can delegate deeper reasoning and tool use to the models and tools paired with it. 17
• GPT-6 Astra can be used by ChatGPT Voice when a request requires harder searching or reasoning. Model availability and limits vary by plan and workspace settings. 13
• GPT-6 Sol and GPT-6 Luna are models for ChatGPT Work and Codex. They are not available in regular ChatGPT conversations. 21

This is why a voice request may not always use the same underlying model. Voice provides the interaction layer, while the selected Chat or Work context determines which capabilities and models can handle the task.

Chat versus Work: the choice that shapes the workflow

The rollout introduces a meaningful product boundary.

Voice in Chat

Use Chat for real-time conversation, brainstorming, questions, and connected-app assistance within the tools available to the account. It is powered by GPT-Live and is available on web, iOS, and Android, subject to plan and workspace settings. 7

Voice in Work

Use Work when the request requires sustained execution: producing files, using connected apps, or taking browser-based steps. Voice in Work is available on web and mobile, and Work tasks can continue in text after the voice interaction ends. 2123

For users, the implication is that speaking a request is only the first step. To turn that request into a multi-step workflow, they may need to choose Work, connect the relevant service, and ensure the integration has the necessary access.

Access, plans, and permissions still matter

The availability of Voice and Work is rolling out across plans and platforms rather than appearing identically for every account. OpenAI’s release notes cite Plus, Pro, Business, Enterprise, Edu, and Healthcare users for web rollout in Chat and Work, while Work itself is available on web and mobile for paid plans except Free and Go. 110

In managed workspaces, administrators retain important controls. Voice in Work requires both Voice and Work access, and enabling Voice does not override a blocked plugin or grant new permissions. Existing app access, data permissions, action restrictions, and workspace controls continue to apply. 423

Usage limits also vary. OpenAI says GPT-Live-1 is available for up to three hours daily on Plus and up to 15 hours daily on one Pro tier, while plan- and workspace-specific limits may differ. 13

A practical way to use the new workflow

For a task involving work data or multi-step execution:

1. Connect the app first. Email, calendar, Slack, and other tools remain governed by their existing authorization and permission settings. 23
2. Choose Work for execution-heavy requests. Use it for creating files, browser work, and longer tasks rather than treating every voice request as a standard Chat question. 21
3. Start by voice, then inspect the written result. This is especially useful for reviewing drafts, checking outputs, and continuing unfinished work after ending a call. 23
4. Expect limits and approval boundaries. A connected tool’s existing permissions and action restrictions remain in force; voice access does not bypass them. 423

The larger shift: voice as an entry point to agentic work

OpenAI’s update changes the role of Voice from a standalone conversational feature to a front door for connected, task-oriented work. GPT-Live handles the spoken interaction; Chat can bring in more capable reasoning when needed; and Work supplies the environment for documents, connected apps, and browser-based execution. 11721

That makes the experience more useful for hands-free task initiation, but it also makes setup and context selection more important. The feature’s real value depends not just on speaking naturally, but on having the right plan, the right mode, and the right permissions in place.