ChatGPT Voice Now Controls Email, Calendar and Slack Using GPT-6 Astra, Sol and Luna Models

OpenAI collapsed the boundaries between ChatGPT’s text, voice and agentic task modes into a single interface Wednesday, enabling users to manage email, calendar events and Slack channels entirely through spoken conversation without switching back to typing.

The update ships across iOS, Android, web browsers and desktop applications. It represents the first time ChatGPT’s plugin ecosystem — which connects the AI to third-party productivity tools — has been fully exposed to the voice layer. Previously, initiating a voice session meant losing access to integrations like Gmail, Outlook, or workplace messaging platforms. Users had to end the audio call and revert to text prompts to trigger any action outside the model’s native knowledge base.

Voice interactions now run on three distinct GPT-6 variants: Astra, Sol and Luna. Astra has powered select voice sessions for several weeks, but Sol and Luna were integrated into the voice pipeline only in the last 24 hours. OpenAI has not published detailed architectural differences between the three models, though offering multiple backend options typically allows the system to route queries based on latency requirements, complexity or computational cost. This mirrors how enterprise API providers segment model tiers for different workload intensities.

The unification also brings ChatGPT Work — the company’s agentic system designed for multi-step document creation, presentation assembly and spreadsheet generation — directly into the voice interface. A user can now dictate instructions to build a financial model or draft a slide deck while walking away from a keyboard. While OpenAI’s Codex product has offered similar agentic capabilities on desktop environments for software engineering tasks, embedding this functionality in the mobile voice layer significantly broadens its accessibility beyond traditional workstation setups.

This release accelerates a deliberate pivot toward ambient computing interfaces. The trajectory began earlier this year with the deployment of GPT-Live-1, an end-to-end audio model that replaced the legacy three-part architecture. Older voice systems required transcribing speech to text, generating a written response, and then synthesizing that text back into audio — a process that introduced noticeable latency and robotic cadence. GPT-Live-1 processes raw audio input and output simultaneously, allowing the model to pause, listen for interruptions and maintain conversational context over extended periods. Atty Eleti, product lead for ChatGPT Voice, noted during a July briefing that the architecture supports continuous dialogues lasting up to 40 minutes.

The push toward seamless, persistent voice assistants carries historical friction for the company. In May 2024, OpenAI faced intense backlash after debuting a voice mode featuring a sound profile widely compared to Scarlett Johansson’s performance in Spike Jonze’s 2013 film “Her.” Johansson publicly stated she had declined OpenAI’s request to use her voice and was surprised by the similarity of the resulting system. The company subsequently pulled the voice option. That controversy underscored the cultural weight attached to human-sounding AI companions and forced OpenAI to navigate the line between natural interaction and perceived impersonation.

By focusing the current update on utility rather than personality — prioritizing Slack summaries and calendar management over conversational mimicry — OpenAI is attempting to ground its voice ambitions in measurable workplace productivity rather than science fiction parallels.

The unified voice, text and agent interface is rolling out globally to all ChatGPT accounts as of Oct. 1, 2026, with no tier restrictions announced for basic voice access.