OpenAI has announced the launch of ChatGPT Work, an autonomous agentic system designed to execute complex projects across corporate databases, local files, and web browsers. According to the official OpenAI announcement, the release is powered by a new frontier model called GPT 5.6. This launch marks a major shift from conversational artificial intelligence toward active, independent task execution, allowing the system to break down massive corporate projects into smaller steps and execute them over several hours without constant human intervention.
The operational foundation of ChatGPT Work is the newly released GPT 5.6 model, which provides advanced reasoning capabilities for multi step tasks. This model is designed to interpret reference files and maintain strict adherence to corporate templates. Alongside this model, OpenAI is merging its developer focused Codex application directly into the main ChatGPT desktop interface. Codex has reached 5,000,000 weekly active users, with 1,000,000 of those professionals operating entirely outside of traditional software development. The older desktop client has been renamed ChatGPT Classic, while the unified app now serves both general business and technical coding needs.
ChatGPT Work is built to manage entire workflows from a single prompt. For example, the agent can autonomously synthesize raw customer research, draft a comprehensive marketing campaign brief, design corresponding creative assets, and then translate and format those assets for 10 distinct regional markets. Internal metrics indicate that nearly 100% of corporate teams inside OpenAI, ranging from finance to sales, are currently utilizing these agentic tools to reduce operational overhead.
To support asynchronous productivity, the platform introduces Scheduled Tasks. This feature allows the software to monitor external communication channels like Slack and Microsoft Teams in the background, compile new messages into updated documents or presentation slides, and distribute those updates to corporate teams. This automation operates even when the user is away from their computer, shifting the role of the employee from active creator to final editor.
On desktop environments, the application utilizes a feature called Computer Use. This capability allows the agent to interact with local files and desktop software by simulating clicks, typing commands, and moving files within the operating system. A built in browser also allows the agent to research market trends, compare online databases, and edit cloud documents in 1 centralized space. To ease the transition for existing users, OpenAI is sunsetting its standalone Atlas browser and migrating those web capabilities into the primary ChatGPT desktop experience.
OpenAI is also launching a public beta for a feature called Sites, which allows users to convert raw data and concepts into interactive web applications. These pages can function as live project trackers, corporate portals, and interactive dashboards that update automatically as the underlying source data changes. Users can test and deploy these sites directly within the desktop application interface.
To address enterprise security concerns, the system includes automated review protections. This protocol uses advanced modeling to inspect sensitive API actions and prevent the unauthorized sharing of proprietary data. Corporate administrators can centrally manage plugin access, regulate browser permissions, and set precise financial spend limits within the Admin Console. This setup allows companies to distribute agentic capabilities across large teams while maintaining strict compliance and budgetary oversight.
The roll out of ChatGPT Work begins immediately for users on Pro, Enterprise, and Edu plans, with Plus and Business plans scheduled to receive access shortly. The updated desktop application is available globally for Windows and Mac operating systems, offering basic chat functions, the Work agent, and Codex tools to all users, including those on the Free tier. Because these long running agentic operations require substantial computing power, usage is tracked based on the complexity of the task, utilizing the same billing structure previously established for Codex.
