ChatGPT now works like an employee. OpenAI on Thursday launched ChatGPT Work, an autonomous agent that can research, build documents, manage spreadsheets, and create web applications for hours at a time without constant human supervision.
The agent runs on GPT-5.6, the company's latest model family that also dropped Thursday after a limited preview period. Bloomberg reported the release was initially delayed by the Trump administration, though OpenAI did not comment on the holdup.
ChatGPT Work is designed for multi-step tasks that would normally require a human assistant: monitoring connected apps, running scheduled jobs, and producing finished deliverables. Users can follow progress, change direction mid-task, or approve actions as the agent works.
It functions across web, mobile, and desktop and can tap into local files, a built-in browser, and third-party plugins. The release doubles as a major platform consolidation. OpenAI is merging its Codex developer tool directly into the ChatGPT desktop app for macOS and Windows.
The new unified app combines Chat, Work, and Codex in one interface. The old ChatGPT desktop app is being renamed ChatGPT Classic, and existing Codex users can update their app to the new version while keeping their projects, settings, and workflows. A new Sites beta feature lets users turn work into interactive websites or lightweight web apps without leaving ChatGPT, then publish them publicly via URL.
ChatGPT Work is available today for Pro, Enterprise, and Edu users. Plus and Business subscribers get access in the coming days.
Free and Go users are excluded from the agent. The GPT-5.6 lineup brings three tiers. Sol, the flagship model, costs $5 per million input tokens and $30 per million output tokens through the API.
Terra runs $2.50 input and $15 output. Luna, the budget option, is $1 input and $6 output.
Free and Go users default to GPT-5.6 Terra, while Plus and above can choose among all three.
OpenAI claims Sol delivers state-of-the-art results across coding, cybersecurity, and science while using fewer tokens than competing frontier models. The company also highlighted improved design judgment, saying GPT-5.6 can create tasteful, functional interfaces from high-level direction alone.
A new ultra acceleration mode, available to Pro and Enterprise users in ChatGPT Work, coordinates multiple agents across parallel workstreams to finish complex tasks faster. The models are rolling out over the next 24 hours.
OpenAI will retire GPT-5.4 on July 23, while GPT-5.5 models remain available.
