ChatGPT Work explained simply: on July 9, 2026, OpenAI launched something a little different from a typical ChatGPT update. Instead of a smarter chatbot, they shipped an agent that’s meant to actually finish tasks for you rather than just talk about them. We went through exactly what it does, how it works under the hood, who can access it right now, and where it genuinely helps versus where it still needs human oversight.
What ChatGPT Work Actually Is
Regular ChatGPT answers questions and drafts text in a back and forth conversation. ChatGPT Work is different. You give it an outcome you want, and it gathers context across your connected apps and files, breaks the job into smaller steps, and works through them independently, staying on a task for hours if needed. The output isn’t a chat response. It’s a finished document, spreadsheet, slide deck, or even a functioning web app.
OpenAI’s own description of the product puts it simply: it’s built to move ChatGPT “beyond answering questions to getting real work done across web, mobile, and desktop.” It’s a similar philosophy to what Apple is chasing with the agentic ambitions built into Siri AI, just aimed at professional output rather than personal tasks.
The shift here is genuinely significant if you think about how most people have used ChatGPT up to this point. A typical session involves asking a question, getting an answer, maybe asking a follow up, then copying the result somewhere else to actually use it. ChatGPT Work collapses that entire loop. You describe what you need once, walk away, and come back to something that’s already built rather than something you still need to assemble yourself.
What’s Powering It
ChatGPT Work runs on GPT-5.6, OpenAI’s newest model family, which launched alongside the Work agent on the same day. GPT-5.6 actually comes in three versions. Sol is the most powerful and is aimed at the hardest tasks. Terra is built to balance speed and capability for everyday work. Luna is the fastest and lightest of the three, meant for quick, simple jobs.
There’s also a new “ultra” mode inside Sol that lets the system delegate parts of a task to smaller sub-models running in parallel, which is part of how it’s able to stay working on complex, multi-hour projects without losing track of the overall goal. Think of ultra mode as one main coordinator model breaking a big job into smaller pieces, handing those pieces off to helper models, then stitching the results back together into one coherent finished product.
Sam Altman has specifically highlighted efficiency gains with this new model generation, noting that Sol is substantially more token efficient on agentic coding tasks compared to earlier versions. That efficiency matters directly for cost, since more efficient token usage generally translates into cheaper agent runs for the same amount of finished work.
What You Can Actually Ask It to Do
Based on OpenAI’s own rollout details, ChatGPT Work is built around producing finished materials rather than conversation. Common use cases include:
- Turning scattered notes and rough drafts into a polished document
- Building and cleaning up spreadsheets, including formulas and formatting
- Creating full slide presentations from a topic or existing outline
- Generating and hosting simple websites directly through ChatGPT
- Pulling information from connected apps like Google Drive to inform the finished output
The Codex technology that previously powered OpenAI’s separate coding tool has been folded directly into ChatGPT Work, which is part of why it’s capable of handling more technical outputs like building a working web app, not just text documents.
To make this more concrete, imagine you’re a small business owner who just finished a series of customer interviews and has a folder full of rough notes. Instead of manually organizing those notes into a report, you could describe the outcome you want, point ChatGPT Work at the relevant files, and let it draft a structured summary document complete with headers, key findings, and suggested next steps. The same logic applies to building a quarterly budget spreadsheet from receipts and invoices, or turning a rough outline into a polished slide deck ready to present.
How It’s Different From Regular ChatGPT Agents
OpenAI already had agent-style tools before this, including Operator, deep research, and Workspace Agents for enterprise teams. ChatGPT Work essentially consolidates and upgrades this category into one clearer product. Rather than juggling separate tools for separate agent tasks, ChatGPT Work becomes the single place inside ChatGPT where longer, more ambitious projects live.
For businesses already using ChatGPT’s enterprise tools, this also ties into existing integrations like ChatGPT for Excel, ChatGPT for PowerPoint, and Google Drive connections, all of which can now feed into what ChatGPT Work produces, putting it in more direct competition with the office style workflows Apple hopes to eventually support through agentic extensions for Siri AI, and with Google’s own push to fold Gemini into everyday tools, including the coaching layer inside the Fitbit Air.
How It Compares to Microsoft Copilot
It’s worth briefly addressing the elephant in the room here. Microsoft’s Copilot has been pursuing a similar vision inside Word, Excel, and PowerPoint for a while now, deeply embedded into the Microsoft 365 suite most businesses already use daily. ChatGPT Work takes a different approach by staying inside ChatGPT itself and connecting outward to your other tools rather than living inside them directly.
The practical difference comes down to where you’d rather do the work. If your team already lives inside Microsoft 365 for documents and spreadsheets, Copilot’s native integration has an obvious advantage in convenience. If you’re comfortable centralizing your workflow around ChatGPT as the starting point and letting it reach into other apps as needed, ChatGPT Work offers a more flexible, tool agnostic approach. Interestingly, OpenAI’s GPT-5.6 model has also become available inside Microsoft 365 Copilot itself, which shows just how tangled these competing ecosystems are becoming, with the same underlying model sometimes powering multiple competing products at once.
Who Gets Access and When
The rollout is staggered by plan type. ChatGPT Work launched first on web and mobile for Pro, Enterprise, and Edu users, with Plus and Business users gaining access within days of launch. On desktop, it’s bundled into a new unified ChatGPT app that brings Chat, Work, and Codex together in one place, available across every plan, including the free tier, though usage limits vary significantly depending on which plan you’re on.
This desktop consolidation is a meaningful shift on its own. Previously, Codex lived in a separate app from regular ChatGPT. Now it’s folded into a single application, alongside the older “Classic” version of the ChatGPT desktop app, which remains available if you’d rather not switch right away.
How Pricing Actually Works
ChatGPT Work usage is metered through a credit system tied to GPT-5.6 token pricing. Business plan customers had an extended free period for some of these features, with credit-based pricing formally beginning in early July 2026. Pricing for the underlying model itself varies by version, with the most powerful Sol model priced noticeably higher per token than the faster Terra and Luna versions.
In practical terms, this means casual use for smaller tasks stays affordable, while running long, complex, multi-hour agent tasks on the most capable model will draw down credits more quickly, particularly for business and enterprise workspace accounts managing usage across an entire team. Admins on Business and Enterprise plans can view workspace agent activity and usage directly inside an admin console, along with setting safeguards on which specific actions an agent is allowed to take for each connected app, which matters for any organization concerned about an autonomous agent taking unintended actions inside sensitive systems.
Setting Up ChatGPT Work for the First Time
Getting started involves a few basic steps once you have access on a qualifying plan. First, you connect the apps and files you want ChatGPT Work to be able to pull context from, things like Google Drive, existing documents, or specific project folders. Second, you describe the outcome you’re after in plain language, being as specific as possible about format, tone, and any must include details. Third, you let it work, checking in periodically if it’s a longer running task, or simply waiting for a shorter job to complete.
The more specific your initial request, the better the first draft tends to be. Vague requests like “make me a report” produce serviceable but generic results, while requests that specify audience, length, tone, and required sections tend to produce something much closer to a finished, usable product on the first attempt.
How It Fits Into the Bigger AI Assistant Picture
It’s worth stepping back and noticing the timing here. ChatGPT Work launched just about a month after Apple’s own AI assistant overhaul, Siri AI, and roughly around the same period Google rolled its Gemini AI coaching features into products like the Fitbit Air. Every major AI player right now is racing to move from “answering questions” to “actually finishing tasks,” and ChatGPT Work is OpenAI’s clearest step in that direction so far.
If you’re trying to figure out how ChatGPT as a whole compares to Apple’s newly rebuilt Siri, we cover that directly here: ChatGPT vs Siri AI 2026: Which AI Assistant Actually Answers Better? And for the full three way breakdown including Gemini, see our Best AI Assistant Apps of 2026 roundup.
Is ChatGPT Work Actually Reliable Yet
This is the honest caveat worth mentioning. Agentic AI tools across the entire industry, not just OpenAI’s, have a mixed track record so far. Some industry analysis has pointed out that a large share of agentic AI projects at companies end up getting canceled, often because the tools promise more autonomy than they can reliably deliver without a human checking the output closely.
That doesn’t mean ChatGPT Work isn’t useful. It clearly represents real technical progress, and having Codex’s more technical capabilities folded directly into a general purpose tool is a genuine step forward. But treating a multi-hour autonomous agent as fully “set it and forget it” for anything business critical is still risky right now. Reviewing the output before it goes out the door remains a smart habit, at least for the foreseeable future, the same caution worth applying to any of the newer hardware projects covered in our OpenAI Jony Ive device guide.
A useful mental model here is treating ChatGPT Work like a capable junior team member rather than a fully autonomous employee. It can absolutely save you significant time by handling the first draft of a task end to end, but final review before anything goes to a client, gets published, or feeds into an important decision still belongs with a human.
Frequently Asked Questions
What is ChatGPT Work used for? It’s an AI agent inside ChatGPT that completes multi-step tasks and produces finished materials like documents, spreadsheets, presentations, and simple web apps, rather than just answering questions in a chat window.
Is ChatGPT Work free? It’s included across all plans on the new unified desktop app, including Free, but usage limits and access to the most capable models depend on your plan level.
What model powers ChatGPT Work? GPT-5.6, OpenAI’s latest model family, available in three versions called Sol, Terra, and Luna, each balancing power and speed differently.
Is ChatGPT Work the same as Codex? Not exactly. Codex’s technology has been folded into ChatGPT Work, but they share the same underlying usage pool and access depends on your specific plan.
Can ChatGPT Work build a website? Yes. OpenAI has specifically highlighted hosted website creation as one of the new capabilities available through ChatGPT Work.
Does ChatGPT Work replace Microsoft Copilot? Not directly. They solve similar problems in different ways, with Copilot embedded natively inside Microsoft 365 apps and ChatGPT Work operating from within ChatGPT and connecting outward to your other tools.
How long can ChatGPT Work run on a single task? OpenAI has described it as capable of staying on complex projects for hours at a time, breaking larger goals into smaller steps it completes independently before returning a finished result.
