OpenAI Announces ChatGPT Voice Now Powered by GPT-6's Astra, Sol, Luna, with Plugin Integration
OpenAI has announced that ChatGPT Voice is now powered by GPT-6's Astra, Sol, and Luna, and for the first time can invoke plugins such as email, calendar, and Slack through voice, while also entering the web and mobile ChatGPT Work.
The official demonstration includes checking the weather, online shopping, viewing schedules, sending emails, identifying and canceling duplicate charges in financial applications, and verbally creating websites with checkout pages. In the Work scenario, users can generate documents, presentations, websites, and spreadsheets via voice, or handle longer tasks in the browser. LinkedIn's synchronization states "hundreds of plugins." The update is being rolled out globally with the latest application version.
GPT-6 Astra was released as the flagship model weeks ago, with the company claiming it has reached new heights in computer usage, browsing, software engineering, and professional work; Sol and Luna have just been upgraded to GPT-6 as mid-range models. ChatGPT Work was launched mid-year, positioned as an agent that can collect materials across applications, produce outputs, and continue to advance according to plans, with a plugin directory covering Slack, Teams, cloud storage, email, calendars, CRM, and internal tools. Previously, voice was limited to conversational levels, but this time the same set of plugins and agents is connected to the speaking interface.
This transforms the "chatting voice" into a "doing voice." Sam Altman has referenced the movie "Her" as a daily assistant since the first version of voice in 2024. The risks have shifted from saying the wrong thing to sending emails incorrectly, mischanging schedules, and mismanaging accounts. The model card previously categorized Astra under the cybersecurity "critical" category in the preparation framework, meaning it can more autonomously discover and exploit vulnerabilities when obtaining tools and permissions; voice combined with plugins equates to connecting these capabilities to personal and corporate accounts.
In market mechanics, buyers are individuals and enterprises looking to type less and integrate assistants into office suites; the pressure on sellers remains on those who still treat voice as an additional feature, with plugins only existing in text windows. Funding is distributed across Plus, Pro, business and enterprise seats, as well as API, Azure, and Bedrock. Beneficiaries include plugin vendors already in the Work directory, while system-level voice assistants face pressure if they cannot read and write to the same batch of work accounts.
On the supplementary level, enterprises can restrict Astra's access to websites and desktop applications, manage uploads, downloads, and browsing history. OpenAI has not disclosed the confirmation thresholds for invoking plugins via voice, interception of false wake-ups, and differences in billing by model.
Source: Public Information
ABAB AI Insight
OpenAI is integrating its strongest models and office agents into microphones, aiming to capture the verbal layer of operating systems. Astra first proves capable of using computers and browsers, while Voice demonstrates that the same set of plugins can be driven without needing to see a screen. Work is responsible for turning conversations into documents and websites, while voice reduces initiation friction. Together, these three elements mean that the chat window is no longer the product boundary; account permissions are.
Capital seeks daily active users to shift from "asking a question" to "delegating a day's work." Seat fees only align with enterprise budgets when assistants can access email and calendars. The plugin directory positions Slack and email vendors as distribution surfaces, while concentrating data and operational risks into the same login state. Azure and Bedrock channels sell the same voice agent to clients who have already procured cloud services, reducing the need for additional vendors.
A similar structure exists in smartphone assistants connecting to system intents and desktop automation scripts linking to enterprise single sign-on. The difference is that the model decides when to invoke which plugin. The industry is at a stage where text agents can just cross-application delivery, while voice is still treated as a demonstration. Whoever first makes confirmation, cancellation, and auditing usable as commands will turn "Her" from an Easter egg into a default entry point.
The essence is a technological replacement combined with a shift in entry points. The mechanism is: keyboards provide precise commands, while voice offers high-bandwidth ambiguous commands; plugins translate ambiguous commands into write operations on real accounts. What is replaced is the point-and-click path, and what is shifted is the default location of "where to start today's work."
ABAB News · Cognitive Laws
Voice only becomes actionable when connected to accounts.
The boundaries of the assistant are not in the dialogue box, but in the permissions it can access.
The day voice can change schedules is when operating systems begin to change entry points.