OpenAI has moved ChatGPT Voice Mode to the new GPT-6 model lineup — Astra, Sol, and Luna — and connected the voice core to external plugins: email, calendar, and Slack. End-to-end actions are now performed by voice without text input, and in the enterprise environment ChatGPT Work, the assistant creates documents, presentations, spreadsheets, and websites by voice command. The update is rolling out globally starting September 23, 2026, but so far it is backed only by the company's official post — without technical documentation or independent measurements.
What happened
OpenAI announced the update in an official post on X. ChatGPT Voice Mode has been moved to the GPT-6 lineup of three models — Astra, Sol, and Luna. The voice core has gained direct access to external plugins: email, calendar, and Slack. End-to-end actions — checking the calendar, dictating emails, working with Slack — are performed by voice command without switching to text input. In the enterprise environment ChatGPT Work, available in the web version and mobile apps, the assistant creates documents, presentations, spreadsheets, and websites by voice command. The global rollout started on September 23, 2026, and is ongoing in the latest version of the app.
Context
Prior to this update, voice in ChatGPT worked as a "question–answer" interface: the model answered aloud, but any actions with work services required text input or manual operations. The new step changes the architecture: on top of speech works a tool-calling mechanism — the same "LLM plus plugins" combination that has already become the standard for text agents, only now it is built into a mass-market voice product. This is the first time the GPT-6 lineup has entered OpenAI's main product not as a separate API offering, but through a consumer scenario. The second part of the context is enterprise: ChatGPT Work continues the line of turning the assistant from a chat into a work tool with access to office services, and voice here becomes another way to launch work operations.
Why this matters for the industry
For the industry, this is a platform shift, not a cosmetic feature update: voice has for the first time become an agentic channel with access to work services. The "voice → plan → actions via API" scheme is becoming a public reference for agentic UX, and thin voice wrappers around email, calendar, and office messengers are being devalued — their niche has been taken by a product with OpenAI's distribution. At the same time, a window opens for vertical voice agents in narrow domains where specialization, not universality, is important. The impact on production pipelines is currently zero: this is OpenAI's closed functionality, and there is no data on the API, price, and latency of end-to-end voice actions. In the coming months, independent measurements of the accuracy and latency of voice agents and competitor responses are likely; if OpenAI opens the GPT-6 models or the voice core via API, tool calling on top of speech will become a standard integration pattern, and the main engineering task will be controlling latency, cost, and quality of multi-step chains.
Why this matters for users
To get the new capabilities, it is enough to update the ChatGPT app to the latest build — the global rollout is already underway. After the update, you can check the calendar, dictate emails, and work with Slack by voice without text input, and in ChatGPT Work on the web and in mobile apps — create documents, presentations, spreadsheets, and websites by voice command. Since the features are available immediately, it makes sense to test them manually before trusting the assistant with important work operations: see how it handles ambiguous phrasing, how quickly actions are performed, and what happens if it mishears a task.
What is still unknown / limitations
The only source on the update is OpenAI's marketing post on X: there is no system card, benchmarks of agentic capabilities, or testing methodology. The accuracy of tool calling by voice, the latency of end-to-end actions, the frequency of erroneous actions, and how irreversible operations are confirmed are not disclosed. There is no information about the API, price, and rollout rates for all users — the global announcement of September 23, 2026, does not rule out phased rollout. Until independent measurements, claims about agentic UX should be treated as a company declaration.
Sources
Author
Look at AI, editorial team