OpenAI just gave ChatGPT a promotion. On July 9, 2026, the company launched ChatGPT Work, a new mode that stops answering questions and starts finishing jobs instead: full spreadsheets, slide decks, documents, and even working web apps, built while you go do something else.
Here’s the thing. Every AI chat tool runs into the same ceiling. You ask, it answers, and then you copy that answer somewhere else and do the actual assembly yourself. That’s the ceiling OpenAI is trying to break with ChatGPT Work. Instead of a wall of text you still have to turn into something usable, you get the usable thing directly: a budget tracker with the formulas already built in, a deck pulled from your own meeting notes, a signup form you can hand off with a link.
What ChatGPT Work Actually Is
ChatGPT Work is an agent mode that sits inside ChatGPT, right next to your regular chat. You give it a goal instead of a question. It gathers the information it needs from your connected apps and files, breaks the job into steps, and works through them on its own, sometimes for hours at a stretch, before handing you back a finished artifact instead of a reply.
It runs on GPT-5.6, OpenAI’s newest model family, which reached general availability the same day. It’s also built directly on top of Codex, OpenAI’s coding agent that already has more than 5 million weekly users. What’s notable is that over 1 million of those users aren’t developers. They were already using a coding agent to get non-coding work done, which is basically OpenAI’s proof of concept for turning that same approach into a product for everyone else.
What It Can Actually Build
The whole pitch of ChatGPT Work is that the output is a finished thing, not a draft you still have to format yourself. Four categories cover most of what it produces:
- Spreadsheets with working formulas and summaries, not just raw numbers dumped into a grid.
- Slide decks built from outlines, meeting notes, or a data room you point it at.
- Documents like proposals, plans, and research briefs, already structured and formatted.
- Sites, a public beta feature that turns your work into an interactive web app, dashboard, or portal you can share through a link.
That last one is a bigger deal than it sounds. A calculator, a live project tracker, or an internal portal used to mean looping in someone who could build it. Now it means describing what you want and waiting.
How It Works Day to Day
ChatGPT Work reaches your actual data through a unified plugin directory. Think Slack, Microsoft Teams, Google Drive, SharePoint, email, calendars, CRMs, and project trackers. You can type “@” followed by an app name to pull its context into a task directly, or just let ChatGPT figure out which plugin is relevant on its own.
Before it runs a long task, ChatGPT Work can show you a step-by-step plan first, something OpenAI calls Plan mode. You approve the plan, set check-ins at points that matter, and it pauses for your sign-off before anything sensitive happens. That’s the control valve that makes handing over a multi-hour task feel less like a leap of faith.
Tasks don’t have to run just once either. Scheduled Tasks let you set something to repeat weekly, trigger off an event, or monitor continuously: reviewing new Slack messages every Monday morning, refreshing a dashboard when new data lands, or updating a deck when feedback comes in by email.
On desktop, there’s a built-in browser so ChatGPT can research the open web, plus a feature called Computer Use that lets it click, type, and move files across your actual apps in the background. OpenAI is also merging the standalone Codex app into a single new ChatGPT desktop app that includes Chat, Work, and Codex on every plan, including Free. The older desktop app gets renamed ChatGPT Classic. In a related move, OpenAI has begun sunsetting its standalone Atlas browser, folding those capabilities into ChatGPT’s built-in browser and an updated Chrome sidebar extension instead.
The Three Models Behind ChatGPT Work
ChatGPT Work runs on three GPT-5.6 variants, and you can pick per task if you’re on a paid plan:
- Sol is the flagship, tuned for the hardest coding, knowledge work, and research tasks.
- Terra is the balanced everyday model, and the one Free and Go users get inside Work.
- Luna is the cheapest option, meant for lighter, routine tasks.
The practical move here is to default to Terra or Luna for routine work and save Sol for anything where accuracy actually matters. Since usage is metered rather than flat rate, that choice affects how fast you burn through your plan’s allowance.
Who Gets Access, and When
| Plan | Access Timing | Models Available |
|---|---|---|
| Pro | Available now, web and mobile | Sol, Terra, Luna |
| Enterprise | Available now, web and mobile | Sol, Terra, Luna |
| Edu | Available now, web and mobile | Sol, Terra, Luna |
| Plus | Rolling out within days of launch | Sol, Terra, Luna |
| Business | Rolling out within days of launch | Sol, Terra, Luna |
| Free / Go | Via the new desktop app only | Terra |
There’s no separate charge for ChatGPT Work itself. It’s included in your existing plan, but usage counts against your plan’s allowance the same way Codex usage does, so a long, complex task can eat a visible chunk of your daily capacity. Enterprise and Edu admins get spend controls in the Admin Console, including workspace defaults, group limits, and individual overrides for people who need more room.
How to Start Using ChatGPT Work: Step by Step
- Check your plan. Pro, Enterprise, and Edu users have access now on web and mobile. Plus and Business are rolling out within days. Free and Go users need the desktop app.
- Download or update the ChatGPT desktop app. Grab it from chatgpt.com/download. If you already have the Codex app installed, updating it turns it into the new unified app automatically.
- Switch to Work mode. Inside the app, use the mode switcher to move from Chat to Work.
- Connect the apps you actually use. Open the plugin directory and connect two or three tools you touch daily: your email, your calendar, your CRM, or Google Drive. Don’t connect everything at once. Start with what you’d use this week.
- Write a brief, not a question. Describe the deliverable, who it’s for, and what “done” looks like. A vague prompt gets you a competent but generic result. The more specific your brief, the sharper the output.
- Review the plan before it runs. Use Plan mode to see the steps ChatGPT intends to take. Adjust anything that looks off before you approve it.
- Set check-ins for anything sensitive. If the task touches customer data, financials, or anything that goes external, add an approval gate so it pauses before acting.
- Let it run, then check the actual output. Review the finished file the way you’d review a report from a new hire: check the numbers against the source, don’t just skim the formatting.
- Set up a Scheduled Task for anything recurring. If you’re rebuilding the same report every week, turn it into a Scheduled Task instead of repeating the brief every time.
Start with something you already understand well. Analyzing a month-end budget variance, prepping for a sales call, or turning source material into a campaign brief are the examples OpenAI itself points to, and they’re good ones, because you’ll know immediately if the output is actually right.
What Early Users Are Reporting
OpenAI shared a handful of early case studies at launch. A marketing lead at Zapier used ChatGPT Work to trace customer touchpoints across the CRM and email, then generate a weekly dashboard that surfaced seven figures in missed pipeline. An R&D manager at RingCentral used it to turn manual monthly launch checks into a repeatable workflow, going from supporting one product manager to roughly fifty. A digital products lead at Virgin Atlantic used it to compare the airline’s passenger experience against competitors, cutting weeks of analysis down to hours. A go-to-market manager at NVIDIA used it to automate conference prep that used to eat up 40 percent of his pre-event time.
Worth saying plainly: these numbers come from OpenAI’s own announcement, not an independent audit. Treat them as a ceiling for what’s possible with a well-built workflow, not a guarantee of what you’ll get on day one.
ChatGPT Work vs. Claude Cowork
If this sounds familiar, it’s because Anthropic shipped a similar idea first. Claude Cowork already lets non-technical users hand off folder-based tasks to Claude and get finished work back. ChatGPT Work is OpenAI’s answer to that same question: what happens when an agent built for developers gets pointed at everyday knowledge work instead.
The short version of how they differ: Claude Cowork leans on a folder-based, file-system approach inherited from Claude Code. ChatGPT Work leans on a plugin directory that plugs directly into the SaaS tools most teams already run: Slack, Teams, CRMs, and so on. Which one fits you better probably comes down to whether your daily work already lives in files on a machine, or spread across a dozen connected apps.
If you’re already using Claude’s connector ecosystem, it’s worth running the same brief through both and comparing what comes back. The gap between them will probably close fast, but right now it’s genuinely useful to test both on a real task before picking one as your default.
Honest Caveats
ChatGPT Work is not magic, and OpenAI isn’t pretending it is.
- It doesn’t know your standards unless you tell it. Without a specific brief, you get a generic version of whatever you asked for.
- It won’t reliably catch its own mistakes. A polished spreadsheet that’s quietly wrong is more dangerous than a rough one, because a clean layout makes you less likely to double-check the numbers.
- Cost adds up fast on long tasks. Metered usage means a multi-hour job can burn through a real chunk of your plan’s allowance, so save the flagship Sol model for jobs where accuracy is worth the spend.
- Access is staggered. If you’re on Plus or Business, you may be waiting a few extra days past the initial July 9 rollout.
- Free tier access is limited to the desktop app, running the lighter Terra model, so mobile-only Free users aren’t getting the full experience yet.
Bottom Line
ChatGPT Work is OpenAI’s clearest move yet toward an AI that finishes jobs instead of just drafting them. If your week involves building the same kinds of decks, trackers, and reports over and over, this is worth fifteen minutes of your time. Connect two apps, write one tight brief, and see what comes back before you decide how much of your workflow it actually deserves.
If half your week disappears into decks, spreadsheets, and status reports, this is the first AI update in a while actually built to hand that work off. Worth trying before your next Monday morning starts.
Sources:
Just Being Resourceful! 
Leave a comment