ChatGPT Work: OpenAI merges ChatGPT and Codex into one agent that never clocks out

OpenAI‘s new agentic product doesn’t pause when you walk away. It takes a project, works through your files, browser, and connected apps for hours, and returns finished material — no supervision required.

ChatGPT Work is OpenAI’s answer to the question it has been circling for the past year: what does an AI that actually does the work look like? Powered by the newly released GPT-5.6 model family, the product runs multi-step tasks autonomously in the background for hours — accessing files, browsing the web, and executing actions across connected services. It does not wait for your next message.

The launch also closes a chapter. ChatGPT, the conversational assistant used by hundreds of millions of people, and Codex, OpenAI’s dedicated coding agent released earlier this year, are now a single product. The standalone Codex application is being retired; its capabilities fold directly into ChatGPT Work alongside an integrated browser and a plugin directory covering Slack, Google Drive, Salesforce, and GitHub. The old ChatGPT desktop app becomes ChatGPT Classic.

Three model tiers run underneath. Sol — the flagship — is built for complex reasoning and extended agentic runs. Terra handles daily work at roughly half the cost of GPT-5.5. Luna is the fastest and cheapest option. Free users now get access to Terra, a genuine step up from what the free tier previously offered. Sol requires a Pro or Enterprise subscription.

In practice this means handing ChatGPT Work a project brief and walking away: draft ten social posts from a document, compile a month’s sales figures into a slide deck, clear a backlog of support tickets. The agent handles each step and returns finished output without anyone watching. OpenAI says the product works across web, desktop, and connected apps — not just text in a browser tab.

The trust question is where things get harder. Granting an AI agent access to Google Drive, Slack, and email means accepting that it will act in those systems — and that its errors will land in shared documents and sent messages rather than in a chat log no one else reads. OpenAI says real-time monitoring and safeguards are in place; they have not yet been tested at production scale. The autonomous background operation also means the agent continues running after you close the laptop, useful until it acts on instructions that have since changed.

The launch positions OpenAI directly against Microsoft Copilot, which has offered comparable office automation inside Microsoft 365 for more than a year, and against Anthropic‘s Claude Cowork, which expanded from desktop to web and mobile four days before ChatGPT Work’s debut. Anthropic’s own data — drawn from 1.2 million anonymized sessions — found that fewer than 9 percent of AI agent work involves coding; business operations account for more than a third. OpenAI’s decision to fold Codex into a broader platform rather than maintain it as a developer-only tool follows the same reading.

ChatGPT Work is available now for Pro, Enterprise, and Edu subscribers, with expansion to Plus and Business plans expected within days. GPT-5.6 is also available through the OpenAI API for developers. OpenAI’s standalone Atlas browser, which previously ran agent tasks in a separate window, will be retired on August 9.

The standalone Codex app is already gone. The question OpenAI has decided to answer in public is whether one platform can handle the trust problem that two separated tools never had to face alone.

Tags: OpenAI, chatgpt, GPT-5.6, AI agents, ChatGPT Work, Codex