OpenAI launched ChatGPT Work on July 9, 2026, a new AI agent that transforms its popular chatbot into an autonomous worker capable of creating polished documents, spreadsheets, presentations, and even interactive websites—all from a single prompt. The release, powered by the new GPT-5.6 model family, marks a significant escalation in the race to bring AI coding capabilities to non-technical users, putting it in direct competition with Anthropic’s Claude Cowork and Microsoft’s Copilot Cowork.
What ChatGPT Work Actually Does
ChatGPT Work is not just another chatbot upgrade. It combines the conversational abilities of ChatGPT with the code-generation muscle of Codex, enabling it to handle multi-step projects that stretch across hours. Instead of asking you to break a task into piecemeal prompts, the agent accepts a high-level goal—like “create a quarterly sales presentation from last month’s data and our recent marketing slide deck”—and then proposes a plan. You can approve or tweak that plan, and the agent gets to work, pulling information from your files, connected tools, and the web.
Along the way, it can pause at checkpoints for your input, refine its outputs, and deliver a finished set of artifacts. The results aren’t just text; ChatGPT Work can generate formatted Excel spreadsheets, Word documents, PowerPoint slides, and even interactive websites through a feature called Sites. These sites can serve as dashboards, project trackers, or prototypes, and they are shareable with colleagues or clients.
The new desktop application—available for Windows and macOS—gives the agent a persistent presence on your machine. It includes a built-in browser so the agent can navigate tabs, log into services, and pull data from web apps as you direct. More than 1,400 plugins extend its reach into tools like Salesforce, Jira, and Notion, making it a cross-application orchestrator rather than a single-purpose bot.
Who Gets Access—and What It Costs
ChatGPT Work is powered by GPT-5.6, which OpenAI released in three tiers:
- GPT-5.6 Sol: The flagship model, tuned for complex, high-stakes tasks. API pricing is $5 per million input tokens and $30 per million output tokens.
- GPT-5.6 Terra: Balanced for everyday use, priced at $2.50 per million input and $15 per million output.
- GPT-5.6 Luna: The lightweight option for high-volume, lower-cost work, at $1 per million input and $6 per million output.
This tiered structure lets organizations assign routine processing to cheaper models while reserving Sol for critical jobs. According to Creative Strategies analyst Max Weinbach, Luna can complete many tasks nearly as well as Sol at one-fifth the cost—a notable advance in cost-efficiency for agentic AI.
ChatGPT Work began rolling out on July 9 to Pro, Enterprise, and Edu subscribers on web and mobile, with Plus and Business users gaining access in the following days. Desktop app availability mirrors these plans, and the agent is included as part of existing ChatGPT subscriptions, though heavy usage may incur additional consumption-based charges.
Why This Matters for Windows Users
For the millions who rely on Windows and Microsoft 365, ChatGPT Work represents both an opportunity and a disruption. The new desktop application sits beside Outlook, Excel, and Teams, able to read files, manipulate data, and even control browser actions. This means you can ask the agent to update a budget spreadsheet with figures from an email, build a PowerPoint deck from a collection of Word docs, or scrape a competitor’s website and compile a summary—all without leaving your primary workspace.
Microsoft Copilot Cowork already offers deep integration with the Microsoft 365 graph, using emails, meetings, and SharePoint data as context. ChatGPT Work takes a broader approach, connecting to any plugin or web service, which may appeal to organizations that mix Microsoft and third-party tools. IT administrators, however, must weigh the governance complexity: an agent that can hop between apps and browser tabs needs careful permission scoping to avoid data leakage or unintended actions.
The desktop app also means OpenAI is no longer just a browser tab; it’s aiming to be a persistent, system-level assistant. For Windows power users, this could replace several automation scripts or macros, but it also demands a new level of scrutiny over what the agent is allowed to do unsupervised.
The Road to Agentic AI
ChatGPT Work didn’t emerge in a vacuum. Anthropic launched Claude Cowork in January 2026, targeting the same “non-coder who needs coding power” audience. Claude Cowork excels at desktop file tasks and local document analysis, while Microsoft made Copilot Cowork generally available as an agentic layer across Microsoft 365, leveraging organizational data. OpenAI’s response stitches together its earlier experiments—Operator, deep research, and Workspace Agents—into a unified service that tries to sit above both office software and web services.
The launch was nearly delayed. In June 2026, The Associated Press and Axios reported that the Trump administration asked OpenAI to restrict access to GPT-5.6 while government agencies assessed national-security and cybersecurity risks. The broad release proceeded on July 9 after extra testing, but the intervention raises questions about how future frontier models will reach commercial markets.
How to Get Started Safely
If you’re a Pro, Enterprise, or Edu subscriber, ChatGPT Work should already be available in your desktop app or web interface. Start small:
- Test with contained data: Give the agent a well-defined task using non-sensitive files to understand its planning and output quality.
- Review every plan: The Plan mode shows you exactly what steps the agent intends to take. Scrutinize these before clicking approve, especially when the task involves web actions or external plugins.
- Set boundaries: In enterprise settings, work with your IT team to restrict plugin access, enforce human approval for sensitive operations, and enable audit logging. Treat the agent’s permissions like you would a new employee who needs time to earn trust.
- Validate outputs: Any generated spreadsheet, document, or website should be compared against source materials. The agent can hallucinate or misinterpret context, especially in long, multi-step workflows.
- Monitor costs: While subscriptions include ChatGPT Work, extensive use of high-tier models can add up. Use Luna for drafts and low-priority work, and reserve Sol for final, high-stakes deliverables.
For organizations already paying for Microsoft 365 Copilot, evaluate whether ChatGPT Work’s cross-platform reach justifies the added license. A mixed environment may benefit from running both, but overlapping AI agents can confuse workflows and inflate costs.
What’s Next
The next few months will test whether Luna and Terra can handle long, autonomous assignments at the economics OpenAI claims. If they can, agentic automation may become accessible to millions of office workers who never learned to code. If not, ChatGPT Work will remain a powerful but human-dependent tool. Either way, the desktop agent era has begun, and Windows users are on the front line.