OpenAI launched ChatGPT Work on July 9, 2026, and it is the clearest sign yet that the company wants ChatGPT to finish jobs, not just answer questions. ChatGPT Work is an agent, powered by the new GPT-5.6 models, that can run for hours across your apps and files, then hand back finished documents, spreadsheets, presentations, live dashboards, and even working web apps. OpenAI framed it as a tool for your most ambitious work, and for once the framing tracks with what shipped.
For anyone who builds things, from a solo creator assembling a launch page to a small studio producing client decks, the interesting part is not the chat. It is that the agent now owns the whole loop: gather context, plan the steps, execute across tools, and produce a real deliverable you can ship. Here is what ChatGPT Work does, how it compares to what you already use, and how to put it to work today.
What ChatGPT Work Actually Is
ChatGPT Work is a persistent agent built into ChatGPT with Codex wired in underneath. Instead of returning a single response, it translates a broad goal into a sequence of steps and completes them across the web, mobile, and desktop. It can pull context from your connected apps and files, use a new built-in browser to reach websites and online tools, and lean on a plugin system that connects outside apps and systems into the workflow.
The agent is designed to work independently. You can schedule it, kick off a task from your phone, and let it run in the background while it assembles the output. According to MacRumors, it can operate continuously for hours on complex projects, which is the difference between a chatbot that drafts a paragraph and an agent that produces a finished report.
The launch also introduced a Sites beta, a feature for generating interactive sites and web apps, including reports and live dashboards, directly from a prompt. That is the piece that pushes ChatGPT Work out of the assistant category and into the build category.
What You Can Build With It
The scope test that matters for creators is simple: does it help you make or ship something? ChatGPT Work clears it. The agent is built to produce finished materials rather than drafts you then have to rebuild:
- Documents and reports with sourced context pulled from your files and the live web.
- Spreadsheets with real data and formulas, not just a table of placeholder numbers.
- Presentations where GPT-5.6's improved design judgment shows up in cleaner layouts.
- Interactive sites and web apps through Sites beta, including live dashboards that update.
GPT-5.6 is the reason the output quality jumped. OpenAI says its flagship GPT-5.6 Sol model produces tasteful, ergonomic, and functional interfaces from high-level direction, and delivers better accuracy in presentations, documents, and spreadsheets. The same model family is already showing up in creative build tools elsewhere, including Figma Make, where it improved first-pass design-to-code quality on July 9.
ChatGPT Work vs the Tools You Already Use
ChatGPT Work does not exist in a vacuum. It sits against the older ChatGPT agent, the standalone Codex workflow, and rival agents from Anthropic and Google. Here is how the pieces line up for someone who wants finished output.
| Capability | ChatGPT Work | ChatGPT Classic | Standalone Codex |
|---|---|---|---|
| Runs for hours unattended | Yes, with scheduling | No, single-turn | Partial, code-focused |
| Builds web apps and dashboards | Yes, Sites beta | No | Code output only |
| Built-in browser | Yes | Limited | No |
| Connects your apps and files | Yes, plugins | Limited | Repo-scoped |
| Model | GPT-5.6 Sol | Prior GPT-5 | Codex line |
| Start from phone | Yes | Yes, chat only | No |
The practical read: ChatGPT Classic still wins for quick back-and-forth, but ChatGPT Work is the one you reach for when the deliverable is a multi-part artifact. If you have used Anthropic's agent tooling through MCP servers for Claude and ChatGPT, the mental model is similar: give the agent tools and context, then let it execute a plan.
The Codex and Desktop App Merger
Alongside ChatGPT Work, OpenAI merged the Codex app and the ChatGPT app into a single desktop application for Mac and Windows. The existing ChatGPT desktop app was renamed ChatGPT Classic, and Codex app users can update into the new unified desktop version. Per Thurrott, this consolidation means the coding agent and the general agent now live in one place, so context handoffs between writing code and assembling a deliverable stop being a copy-paste chore.
This matters more than a UI cleanup. When the code environment and the document environment share the same desktop app, ChatGPT Work can build a web app in Sites beta, wire in data, and produce the report that explains it, all inside one session.
How to Start Using ChatGPT Work
If you are on a qualifying plan, here is a workflow to get a real deliverable out of it on day one:
- Pick a multi-step job, not a question. Good first tasks: "build a live dashboard of my newsletter metrics" or "turn these three files into a client-ready deck."
- Connect your sources. Use the plugin system to link the apps and files that hold your context, and let the built-in browser reach any live data.
- Give high-level direction, not micro-steps. GPT-5.6 Sol is tuned to take a broad goal and make the design and structure calls itself.
- Schedule or background it. Start the task, then let it run. Check in from your phone rather than babysitting the screen.
- Review the artifact, then iterate. Treat the first output as a strong draft. Point the agent at specific fixes rather than restarting.
What this enables: the payoff is time. A dashboard or a client deck that used to eat an afternoon becomes a scheduled task you review once. That is the creator outcome, and it is why the agent framing is worth taking seriously this time. Reporting from Yahoo Finance notes the tool is aimed squarely at replacing hours of manual assembly work.
Availability and Pricing
ChatGPT Work is available immediately on Pro, Enterprise, and Edu plans, and OpenAI said it will expand to Plus and Business plans within days. The GPT-5.6 models rolled out the same day across ChatGPT, Codex, and the OpenAI API. The flagship Sol variant reaches Plus, Pro, Business, and Enterprise users, while the balanced Terra model is available to Free and Go tiers. OpenAI has not published specific per-seat pricing beyond the existing plan structure.
Frequently Asked Questions
What is ChatGPT Work?
ChatGPT Work is an OpenAI agent, launched July 9, 2026 and powered by GPT-5.6, that runs multi-step tasks across your apps and files and returns finished documents, spreadsheets, presentations, dashboards, and web apps.
How is ChatGPT Work different from regular ChatGPT?
Regular ChatGPT, now called ChatGPT Classic, answers in a single turn. ChatGPT Work plans and executes a sequence of steps over minutes or hours, uses a built-in browser and plugins, and produces a complete deliverable rather than a reply.
What can ChatGPT Work build?
Documents, spreadsheets, presentations, reports, and through the new Sites beta, interactive sites and web apps including live dashboards. GPT-5.6 Sol handles the design and structure decisions from high-level direction.
Which plans include ChatGPT Work?
It is live on Pro, Enterprise, and Edu plans immediately, with Plus and Business plans getting access within days of the July 9 launch.
Do I still need the Codex app?
No. OpenAI merged Codex into a single ChatGPT desktop app for Mac and Windows. Codex users update into the unified app, and the old ChatGPT desktop app is now ChatGPT Classic.
What are GPT-5.6 Sol, Terra, and Luna?
They are the three GPT-5.6 variants: Sol is the flagship for coding and knowledge work, Terra is the balanced everyday model, and Luna is the cost-efficient option. ChatGPT Work runs on Sol.
Related deep dives: GPT-5.6 Sol, Terra, and Luna explained and OpenAI GPT-Live full-duplex voice.
