ChatGPT Work
OpenAI
Long-running ChatGPT agent for research, connected workflows, and review-ready work artifacts.
Overview
Freshness note: AI products, plan access, and credit rules change quickly. This profile is a point-in-time snapshot last verified on July 10, 2026.
ChatGPT Work is OpenAI’s agent for longer tasks with a concrete deliverable. It launched on July 9, 2026 as one of three first-class ChatGPT experiences: Chat for questions and conversation, Work for research and finished materials, and Codex for software development.
Work is the right surface when the job involves several sources, tools, or stages and should end in something reviewable: a document, spreadsheet, presentation, report, Site, analysis, or recurring update. It can use files, plugins, approved tools, and connected systems while showing progress and pausing for questions or important approvals.
Key Features
Work can plan and execute a multi-step task instead of returning only a conversational answer. You can follow its progress, add context, change direction, and review the finished artifact. Scheduled Tasks can run work once, on a schedule, after a trigger, or while monitoring for changes.
The local and cloud surfaces differ. Work on web and mobile runs in the cloud. In the new ChatGPT desktop app, Work can also use local files, the built-in browser, and approved macOS or Windows applications. At launch, cloud Work conversations do not appear in desktop Work, while desktop threads and local files remain on that computer.
OpenAI’s current model guide gives Work access to the GPT-5.6 family where the plan exposes it. The default Power setting uses GPT-5.6 Sol with medium reasoning, while Terra and Luna cover balanced and higher-volume work.
Strengths
Work closes the gap between a useful chat and a finished business artifact. It is especially effective when the source material is scattered across files and connected tools but the output still needs a human owner and an explicit review step.
The ability to steer a long-running task without restarting it is also practical. Missing information can be resolved while the agent is working instead of being discovered only after a polished but unusable final draft.
Limitations
Access is still rolling out. On web and mobile, OpenAI launched Work for eligible paid plans except Free and Go, with Pro, Pro Lite, Enterprise, and Edu receiving access first and Plus and Business following. Enterprise and Edu begin with an admin-controlled preview period.
Work shares Codex usage, credits, and plan limits, so a complex deck or research workflow can cost materially more than a short chat. Plugin access, local tools, regional availability, approvals, and workspace controls can also change what the agent may actually do.
Practical Tips
Define the outcome, source set, constraints, review criteria, and stop conditions in the first prompt. Ask Work to show its plan and flag missing information before building the final artifact. For external messages, publishing, record updates, or destructive actions, require approval before execution.
Use ChatGPT Workspace Agents instead when the workflow should become a reusable shared agent for a team, with organization-wide sharing, schedules, Slack deployment, analytics, and admin controls.
Verdict
ChatGPT Work is a strong new default for substantial individual knowledge-work tasks that are too involved for ordinary chat but do not belong in a coding-first surface. Its value is highest when the deliverable and approval boundaries are clear.