ChatGPT Work
ChatGPT Work is an OpenAI AI agent within ChatGPT for longer, multi-step projects and finished deliverables. OpenAI announced it on July 9, 2026. Work can gather context from files, connected systems, and websites, then create outputs such as documents, spreadsheets, presentations, reports, analyses, and interactive Sites. OpenAI says a complex task can continue for hours while the user follows its progress, answers questions, changes direction, and approves important actions.[1]
Work is powered by GPT-5.6 and incorporates technology developed for OpenAI Codex.[1][2] It is nevertheless a distinct ChatGPT experience: OpenAI describes Chat as the place for quick conversational help, Work as an agent for multi-step work, and Codex as the dedicated environment for software development and other technical work.[3]
Key facts
| Field | Detail |
|---|---|
| Developer | OpenAI[1] |
| Announced | July 9, 2026[1] |
| Product type | Agent in ChatGPT for longer projects and finished deliverables[1][3] |
| Model | GPT-5.6[1][2] |
| Main surfaces | ChatGPT web, mobile, and the macOS and Windows desktop app[2][3] |
| Current general availability | All plans on desktop; Plus, Pro, Business, Enterprise, and Edu on web and mobile, subject to workspace and feature controls[2][3] |
| Connected systems | Plugins and their underlying apps, when installed, enabled, and authorized[1][6] |
| Browser options | A built-in desktop browser and a separate cloud browser for supported delegated tasks[4][5] |
| Usage | Follows the Codex usage structure; consumption varies by task[1][3] |
Product scope
ChatGPT Work is a product mode, not a separate subscription plan or a model name. The user specifies an outcome and can supply files, constraints, source material, and review criteria. Work can divide the request into steps and produce a finished artifact rather than only returning a conversational answer. Examples in OpenAI's documentation include research, data analysis, documents, spreadsheets, presentations, reports, and Sites.[1][3]
The Work experience brings long-running tasks, artifact creation, plugins, scheduled workflows, and browser or desktop actions into one mode. OpenAI's documentation presents it alongside Chat and Codex rather than as a new base model.[1][3]
Work and Chat share ChatGPT projects and recent-chat navigation. Codex remains a separate view with separate history. In the desktop app, eligible users can also open a local folder. Local files and outputs stay on that computer unless the user explicitly moves or shares them; Work on web and mobile cannot directly access files on the user's computer.[3]
Availability and execution environments
OpenAI's current product page lists Work on every plan in the ChatGPT desktop app for macOS and Windows, including Free. It lists web and mobile access for Plus, Pro, Business, Enterprise, and Edu.[2] Actual access can still depend on rollout, app version, region, workspace settings, and role permissions.[3][4]
Work on web and mobile runs in the cloud. Cloud Work conversations can sync among supported web, mobile, and desktop surfaces. The desktop app can use cloud execution or, when the plan and workspace allow it, local files and desktop applications. A local conversation remains on that computer rather than appearing in cloud chat history.[3]
For managed organizations, Work Cloud and Work Local are separate permissions. Codex Local is controlled separately again, and browser use and network access have their own controls. As a result, access to one surface or execution mode does not imply access to all of them.[3]
Deliverables and ongoing workflows
Work can draw on connected context to create documents, presentations, spreadsheets, analyses, and other deliverables. Sites, introduced in public beta with Work, lets a user turn work into a shareable interactive website or web application such as a dashboard, tracker, prototype, portal, or report. OpenAI says Work can update a Site as underlying information changes.[1][2]
The user remains part of the execution loop. During a task, the user can inspect progress, answer a clarifying question, change direction, and approve an important action. Descriptions of tasks running for hours establish the intended operating horizon, not a guarantee that every long task will finish without intervention.[1]
Plugins and connected systems
ChatGPT Work uses ChatGPT plugins to connect workflows and context. On July 9, 2026, OpenAI moved its former App Directory into the Plugins Directory. In the current terminology, a plugin packages a workflow and may include reusable skills, apps, and app templates. Apps are the integration layer that connects ChatGPT to external data or actions.[6]
OpenAI's product page lists more than 1,400 plugins and names systems such as Slack, Microsoft Teams, Google Drive, SharePoint, email, calendars, customer-relationship-management tools, and project trackers.[1][2] A user can invoke an installed plugin explicitly with an @ mention, while ChatGPT may also select a relevant plugin from the request.[1][6]
Plugin visibility, installation, and usable app access are different things. Availability can depend on the plan, region, product surface, workspace, role, and included capability. If a plugin requires an app, that app must be enabled and the user may need to complete OAuth and sync setup. Installing the plugin does not grant broader rights in the connected service: source-system permissions still determine which files, records, repositories, workspaces, or channels the user may access.[6]
Built-in and cloud browsers
ChatGPT Work has two browser environments with different operating models.[4][5]
| Browser | Where it runs | State and use |
|---|---|---|
| Built-in browser | Inside the ChatGPT desktop app | The user and ChatGPT can inspect the same page. It has its own browser state and supports multiple tabs, downloads, sign-in, navigation, and page annotation.[5] |
| Cloud browser | On a separate remote computer | Work can carry out supported web steps in the background and continue after the user's computer or phone is closed. It pauses for required input, sign-in, or confirmation.[4] |
The built-in browser does not use the user's existing Chrome profile. OpenAI directs users to the Codex Chrome extension when a task needs an existing Chrome profile, signed-in session, open tabs, or Chrome extensions. Credentials should be entered in the browser interface, not in the chat.[5]
Cloud browser is available on paid ChatGPT plans in supported regions, excluding Free and Go, with availability also subject to rollout and workspace permissions. ChatGPT decides when to use it; a connected app or plugin may be used instead, or the two mechanisms may be combined.[4]
Cloud sign-in and persistent sessions
When a supported website requires authentication, cloud browser pauses for the user to sign in through a secure form and complete any two-factor step. OpenAI says credentials entered there go directly to the remote browser, are not visible to the model, and are not stored by ChatGPT. After authentication, the signed-in browser session can remain available for later tasks until it expires or the user clears that site's browser data.[4]
This persistence applies to the remote browser session, not to the user's ordinary browser. Cloud browser has its own cookies, browser data, and signed-in sessions. It does not inherit the device browser's open tabs, browsing history, saved passwords, cookies, extensions, or existing sign-ins.[4]
Website permission is also distinct from action approval. A user can configure default and site-specific access, but allowing a website does not pre-approve a consequential step. Cloud browser is designed to request confirmation before an action that could create a financial, legal, account, or other hard-to-reverse commitment. The user can also take over the remote browser directly.[4]
Scheduled and event-triggered tasks
Scheduled Tasks can run once, recur on a schedule, monitor for changes, or respond to a supported event. They can use connected systems and, when available, a browser. Event-triggered tasks run in Work and currently respond to supported Gmail, Slack, or GitHub activity.[1][8]
OpenAI lists event-triggered tasks for eligible Plus, Pro, Business, Enterprise, Edu, and Healthcare accounts. They are not available to Free or Go accounts or in FedRAMP workspaces. An action that requires approval can pause the task. A task created inside a ChatGPT project also cannot access files uploaded to or stored in that project, according to the current Scheduled Tasks documentation.[8]
Sharing a task does not transfer the creator's identity, chat history, past results, memories, custom instructions, attached files, app data, credentials, or permissions. A recipient schedules a separate copy using that recipient's own Work access and connected systems.[3][8]
Permissions, security, and privacy
For managed workspaces, administrators can control access to Work, plugins, apps, browser use, network access, and actions. Plugin installation and access to an underlying app are separate controls. Apps may also distinguish read access from the ability to take actions and may require confirmation before particular actions run.[3][6][7]
Enterprise and ChatGPT Edu workspaces disable plugins and their underlying apps by default; administrators can make them available by role. ChatGPT Enterprise and Edu also support role-based controls for Work Cloud and Work Local. The behavior of a managed workspace should therefore be checked rather than inferred from consumer-plan availability.[3][7]
OpenAI says data handled by connected apps is encrypted in transit and at rest and that OAuth tokens use audited key-management practices. For ChatGPT Business, Enterprise, and Edu, OpenAI says information accessed from apps is not used to train its models. Connected third-party services still apply their own data policies and residency rules to data sent to them.[7]
For cloud browser, website information used in a task follows the user's ChatGPT data-control settings. Users can clear data for all sites or a single site, which also ends the corresponding saved session. OpenAI says it tests for prompt injection, phishing, and unintended actions, while warning that those safeguards do not eliminate every risk.[4]
Limitations
Work cannot necessarily complete every website, sign-in, action, or transaction. A website can block automated traffic, a step can be unsupported or restricted, or the user may need to take over and finish the task. OpenAI advises users to review sources, website addresses, sign-in previews, screenshots, confirmation requests, sensitive steps, and important results.[4]
Capabilities and access vary across plans, regions, app versions, rollout stages, workspace roles, and connected systems. Usage also varies with task complexity and follows the Codex usage structure rather than a fixed amount of work per request.[1][3][6]
The available evidence is first-party product documentation. It establishes what OpenAI has announced and how its controls are documented; it does not independently demonstrate that Work is more accurate, productive, or secure than another system.
References
- ^OpenAI. "ChatGPT is now a partner for your most ambitious work." July 9, 2026. openai.com/...chatgpt-for-your-most-ambitious-work
- ^OpenAI. "ChatGPT Work for every team." Accessed Aug. 27, 2026. openai.com/chatgpt-work
- ^OpenAI Help Center. "ChatGPT Work and Codex." Accessed Aug. 27, 2026. help.openai.com/...20001275
- ^OpenAI Help Center. "Using cloud browser in ChatGPT." Accessed Aug. 27, 2026. help.openai.com/...-using-cloud-browser-in-chatgpt
- ^OpenAI Help Center. "Using the built-in browser in the ChatGPT desktop app." Accessed Aug. 27, 2026. help.openai.com/...wser-in-the-chatgpt-desktop-app
- ^OpenAI Help Center. "Plugins in ChatGPT and Codex." Accessed Aug. 27, 2026. help.openai.com/...56-plugins-in-chatgpt-and-codex
- ^OpenAI Help Center. "Admin controls, security, and compliance for plugins and apps." Accessed Aug. 27, 2026. help.openai.com/...11509118
- ^OpenAI Help Center. "Scheduled tasks in ChatGPT." Accessed Aug. 27, 2026. help.openai.com/...10291617
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation. Every suggestion is reviewed for sourcing before it goes live.
v1 · 1,836 words · full history
Fact-checks are independent of edits: a reviewer re-verifies the article against its sources and stamps the date. How we verify
Research and drafting on this wiki are AI-assisted, under named human editorial standards. How AI is used here
Reviewer note: Independently fact-checked against the cited sources on Aug. 27, 2026; claims were limited to what those sources support.
Cite this page: AI Wiki. "ChatGPT Work." aiwiki.ai, updated 27 Aug 2026, fact-checked 27 Aug 2026. CC BY 4.0. https://aiwiki.ai/wiki/chatgpt_work