OpenAI Turns ChatGPT Into a Super App With Work and Codex

OpenAI spent Thursday reshaping what its desktop software is for. Alongside the public arrival of GPT-5.6, the company launched ChatGPT Work, an agent built to hand back finished documents rather than answers, and folded its Codex coding app into a single ChatGPT desktop application for macOS and Windows.

The pitch for ChatGPT Work is that it takes an outcome instead of a prompt. Give it a goal and it gathers context from your connected apps, breaks the job into steps and works through them on its own, staying with a complex project for hours before returning spreadsheets, slides, documents or a working web app. It runs on GPT-5.6 and carries the same underlying technology as Codex.

Three tools, one window

The reorganisation is the visible half of the change. The standalone Codex app is gone, absorbed into a new desktop client that presents Chat, Work and Codex side by side. The old desktop app has been renamed ChatGPT Classic for anyone who wants the previous layout, and developers can still set Codex as their default view. All three are available on every plan, including the free tier, with the web and mobile versions reaching Pro, Enterprise and Edu first and Plus and Business following within days.

For developers, the merged Codex adds inline diff editing, pull-request review inside a side panel, faster Computer Use powered by GPT-5.6 and support for multi-repository projects. ChatGPT Work leans on more than 1,400 plugins to reach outside tools, a plan mode that asks for approval before it acts, and scheduled tasks that can run on a timer or fire on an event.

Billed by the job, not the month

The economics differ from a normal chat subscription. ChatGPT Work uses usage-based metering, closer to how Codex has always been billed, on the logic that long agentic runs consume far more than a single question. OpenAI has not published per-task rates, which leaves teams to measure real consumption before they wire the agent into anything recurring. Anyone who watched metered coding bills land harder than expected will recognise why that caution matters.

The launch puts OpenAI directly alongside the assistants its rivals shipped this month, including Anthropic's Cowork agent, which reached web and mobile days earlier, and Claude's new usage dashboard. The competitive shape is clear enough. The frontier labs no longer want to sell a chatbot. They want to sell the thing that does the work, and ChatGPT Work is OpenAI's plainest statement yet of that ambition.