OpenAI has launched ChatGPT Work, a new agent mode for longer workplace tasks that can research, analyze, work across connected apps and files, and create finished deliverables such as documents, spreadsheets, slide decks, reports, and Sites.

The launch matters because OpenAI is no longer presenting ChatGPT as just a conversational assistant for drafting and Q&A. ChatGPT Work is positioned as a project-running agent: users give it an outcome, attach or connect the relevant context, follow progress, answer questions, redirect the task, and approve important actions as it works.

It is also tied directly to OpenAI's new GPT-5.6 model family. OpenAI says ChatGPT Work uses GPT-5.6 to reason through multi-step work, follow templates and reference files, and produce polished materials with less prompting.

What ChatGPT Work does

ChatGPT Work is built for tasks that normally spill across multiple tools. OpenAI's examples include month-end budget variance analysis, sales-meeting preparation, marketing campaign briefs, executive dashboards, and product launch reviews.

The common pattern is not "write a paragraph." It is closer to:

  • gather information from connected tools and files;
  • plan the work;
  • create or update a deliverable;
  • ask for clarification when needed;
  • keep going in the background while the user reviews progress;
  • hand back a finished document, deck, sheet, report, or Site.

OpenAI says Work can use plugins for tools like Slack, Microsoft Teams, Google Drive, SharePoint, email, calendars, CRMs, project trackers, and internal systems. Users can also call a specific app by typing @ and the app name in a prompt.

On desktop, the new ChatGPT app adds another layer: Work can use local files and desktop apps with permission. It also includes a built-in browser for web-based work, research, and supported online files.

Sites, scheduled tasks, and desktop apps

Three adjacent releases make ChatGPT Work more than a renamed agent mode.

First, ChatGPT Sites is entering public beta. Sites lets users create, preview, publish, and share interactive websites or lightweight apps from ChatGPT Work on the web, or from Work and Codex in the new desktop app. OpenAI lists dashboards, project trackers, launch calendars, prototypes, internal portals, and reports as likely use cases.

Second, Work connects into Scheduled Tasks. That means a workflow can run once, repeat on a schedule, trigger from an event, or monitor for changes. OpenAI gives examples such as reviewing new Slack updates each week, checking dashboards each morning, monitoring customer feedback, or updating a presentation when new email feedback arrives.

Third, OpenAI is merging Codex into a new ChatGPT desktop app. The new app combines Chat, Work, and Codex on macOS and Windows. Chat is for quick questions and brainstorming, Work is for research and finished deliverables, and Codex remains the software-development mode for local folders, repositories, terminals, and developer tools.

That product split is useful. It keeps Codex focused on code while giving non-developers a similar long-running agent surface for business work.

Availability and models

Availability is slightly different by surface.

OpenAI says ChatGPT Work is rolling out on web and mobile to paid plans except Free and Go. Pro, Pro Lite, Enterprise, and Edu users receive access first, with Plus and Business following over the next few days. Enterprise and Edu workspaces get a two-week preview period on web and mobile where Work is off by default before admins decide whether it should turn on automatically.

The desktop story is broader. OpenAI says the updated ChatGPT desktop app is available globally for Mac and Windows, with Chat, Work, and Codex available to users on every plan, including Free. The help center adds a caveat: if Work is not visible yet, access may still be rolling out.

Model access also depends on plan. In ChatGPT Work and Codex, OpenAI says Free and Go users get GPT-5.6 Terra, while Plus, Pro, Business, and Enterprise users can choose GPT-5.6 Sol, Terra, and Luna and set effort levels. max is available to users with GPT-5.6 access in Work and Codex. ultra, OpenAI's highest-capability multi-agent setting for complex work, is available in ChatGPT Work for Pro and Enterprise users.

Why this is a bigger move than agent mode

The important shift is that OpenAI is packaging several pieces into a single workplace loop:

  • GPT-5.6 for higher-effort reasoning and computer use;
  • plugins for connected company context;
  • Work for long-running deliverables;
  • Sites for interactive outputs;
  • Scheduled Tasks for recurring or monitored workflows;
  • desktop computer use for local files and apps;
  • admin controls for plugins, connected tools, browser/network access, and sensitive actions.

That makes ChatGPT Work look less like a one-off feature and more like OpenAI's main push into everyday office automation.

It also changes the competitive frame. The comparison is no longer only Claude, Gemini, or Microsoft Copilot in chat. ChatGPT Work is competing with the messy bundle of spreadsheets, slide decks, CRM exports, Slack threads, dashboards, docs, and manual weekly processes that teams already maintain.

The caveats

The launch is still a rollout, not instant universal access. Users may see different availability depending on plan, region, workspace settings, and app version.

There are also obvious governance questions. The more useful Work becomes, the more sensitive the data and actions become. OpenAI says admins can manage access, connected tools, browser and network use, and sensitive actions, and that auto-review can check important connected-tool and API actions before they happen. Enterprise buyers will still need to test those controls against their own data policies.

Finally, the real value will depend on workflow quality. A good Work setup could save hours by turning source context into a finished deck or dashboard. A poor setup could simply produce another artifact that humans must verify and clean up. The product is promising because it aims at finished work, but finished work still needs accountable review.

Our take

ChatGPT Work is one of OpenAI's clearest moves toward agentic workplace software. It gives non-developers a Codex-like idea: define the outcome, provide context, let the agent work in steps, and review the output.

For teams, the immediate test is not whether Work can write a nice memo. It is whether it can reliably handle recurring, source-heavy workflows: weekly reporting, account-plan updates, launch checklists, budget variance analysis, customer research, and internal dashboards.

If it performs well, ChatGPT Work could become the default place teams start longer AI-assisted office tasks. If it does not, it will still be an important signal: OpenAI is turning ChatGPT from a response engine into a work surface.

Sources: OpenAI launch post, ChatGPT Work product page, GPT-5.6 announcement, ChatGPT release notes, ChatGPT Work and Codex help page, ChatGPT Sites help page, new ChatGPT desktop app help page