The short answer: Chat in the browser is for talking. ChatGPT Work on the web is for delegating from anywhere; it can use your online tools like Gmail and Google Drive, but it cannot touch anything saved on your computer. The desktop app is the full agent: the only surface where it can use your local files, your installed apps, and computer use to click through screens on your behalf. If you run a business, the desktop app is the one that changes your operations, and the other two become its companions rather than its replacements.
I walk through the chatgpt.com Chat and Work toggle, and why I tell owners to get the desktop app, in the full tutorial:
The Three Front Doors, Side by Side
OpenAI now routes you three ways. On chatgpt.com there is a toggle between Chat and Work, and both pages offer you the desktop app download. Here is what actually differs:
| Chat in the browser | Work on the web | The desktop app | |
|---|---|---|---|
| What it is | The ChatGPT you have always known, at chatgpt.com | The agent, reachable from any browser and your phone | The merged app (formerly Codex): Chat, Work, and Codex in one |
| Shape of the output | Answers, drafts, ideas | Finished documents, spreadsheets, presentations, reports, Sites | Everything Work on the web makes, plus work done inside your own machine and apps |
| Reaches your connected apps (plugins) | Limited connector use in chat | Yes: Gmail, Drive, Notion, Slack, CRMs, and the rest | Yes, the full directory |
| Reaches your local files and folders | No (uploads only) | No, cloud files only | Yes |
| Uses your installed apps and clicks your screen (computer use) | No | No, it has its own hosted browser instead | Yes |
| Voice mode | Yes: the new GPT-Live voice is rolling out on web and mobile | No | No, not as of mid-July 2026 |
| Deep Research and custom GPTs | Yes | No | No, those stayed in the browser and ChatGPT Classic |
| Availability | Every plan | Rolling out from July 9: Pro, Enterprise, and Edu first, Plus and Business over the following days | App downloads on every plan (Mac and Windows); the Work agent inside follows the same paid rollout |
Two things in that table surprise most owners. The desktop app, the most powerful surface, is the one without voice mode right now: OpenAI's new GPT-Live voice launched July 8 on web and mobile only. And Work on the web is real work, honestly closer to what I would call a Claude Cowork: it connects to your tools, runs long tasks, and hands back finished files. It simply cannot touch anything that lives on your computer.
When Chat in the Browser Is the Right Tool
The browser did not become obsolete on July 9. It kept three jobs.
Voice. If you think out loud, the browser and the mobile app are currently the only places the new voice mode lives. Talking through a strategy on a walk still happens here.
Deep Research and your custom GPTs. Both stayed behind in the browser and ChatGPT Classic. If you built GPTs for your business, they did not move into the new app; use them where they are until OpenAI migrates them.
Quick questions from any machine. No install, no setup, any computer. That convenience is permanent.
What the browser Chat cannot do is the thing this whole agent era is about: it describes work instead of finishing it. You can get website copy out of it; you cannot get a deployed website.
When Work on the Web Earns Its Place
Work on the web is the same agent as the desktop app, but it can only reach things that live online: your connected apps, your files in Drive or Dropbox, the web itself. Anything saved on your computer is invisible to it. Within that boundary, there are three moments it shines.
You are not at your computer. Start a task from your phone in the school pickup line, review the draft at dinner, and pick the same project up on the web tomorrow. The task keeps moving because it runs on OpenAI's side, not your laptop. This is also why I tell every owner to set up the ChatGPT mobile app on day one: it is how you direct your agent when you are away from your desk.
Your work lives entirely in the cloud. If your business runs on Gmail, Drive, Notion, and a web CRM, web Work can reach all of it through plugins and produce finished deliverables, including hosted Sites.
You are on a locked-down or borrowed machine. Company laptop that will not allow installs, a hotel business center, a client's office. The agent is a login away.
The ceiling: no local folders, no installed apps, no clicking your screen. The twenty years of files on your machine, your desktop accounting software, the legacy tool with no API that needs computer use to operate: all invisible to it.
Why the Desktop App Is the One That Changes Your Operations
Everything in my setup tutorial assumes the desktop app, and this is why.
It sees your actual business. Your local folders, the proposals from eight years ago, the downloads you never filed. The agent can read, organize, and build from all of it without you uploading a thing.
It works your machine, with your permission. Computer use means it can open apps, click through settings you do not understand, and operate software that has no plugin, in the background while you do something else. The permissions you toggle (full access, computer use, locked use, prevent sleep) are what turn it from a chatbot into something closer to an employee.
It carries the builder's toolkit. The built-in agent browser, the Codex lane for anything code-shaped, scheduled tasks that run on your machine. One warning that comes with that power: locally running work depends on your computer being awake, which is exactly why the Prevent sleep setting exists. Turn it on, or your Drive-cleanup agent falls asleep when you go to the gym.
And if your computer is too old to run the desktop app, I will say the unpopular thing: that is your sign to get a new computer. This is the tool your operations will run on. Eventually you will want to explore local and open-source models too, and the same machine carries you there.
How I Actually Run All Three
This is my honest setup, and what I recommend to clients:
1. Desktop app as headquarters. All real work happens here: the agent folder, the connected plugins, the projects. This is the surface you configure once and properly.
2. Phone and web Work as the remote control. Kick off tasks, answer the agent's questions, review drafts from anywhere. Same account, same projects.
3. Browser Chat for voice, Deep Research, and your old GPTs. It is the annex where the features that have not moved yet still live.
The mistake to avoid is treating the three as interchangeable and configuring none of them well. The owners getting $65,000-assistant value from a $100 plan did the desktop setup once, correctly, and then used the web and phone as windows into it.
Common Questions
Is ChatGPT Work on the web the same agent as the desktop app? Yes. It is one agent with one brain, and you are just talking to it through two different windows. Log in on the web and you will see the same projects and the same conversations you started in the desktop app, and work done in one place shows up in the other. Both draw from the same monthly usage allowance, which works like the data plan on your phone: every task uses some of it, and it refills on a schedule. The difference is what the agent can touch from each window. From the web, it can only use things that live online, like your Gmail, your Google Drive, and your connected apps. From the desktop app, it can use all of that plus everything on your actual computer: the folders on your desktop, your downloads, and the programs you have installed.
Do I lose anything by staying browser-only? You lose the agent's ability to see your computer. Practically, that means it can read a contract stored in Google Drive but has no idea about the folder of old proposals saved on your desktop. It cannot open the programs installed on your machine, and it cannot click through screens on your behalf the way the desktop app can. If years of your business history live on your computer, or you depend on an older program that does not connect to anything, a browser-only agent is working with one eye closed.
Where is voice mode? In your browser at chatgpt.com and in the ChatGPT app on your phone, and only there for now. OpenAI released its new voice experience, called GPT-Live, on July 8, 2026, and it has not been added to the new desktop app, Work, or Codex yet. So if you like talking to ChatGPT instead of typing, keep using the browser or your phone for that part; your desktop app cannot do it yet.
I'm on Plus and don't see Work on chatgpt.com. Nothing is wrong with your account, and you do not need to fix anything. OpenAI is turning Work on for groups of customers in waves: the Pro, Enterprise, and Edu plans got it first starting July 9, and Plus and Business accounts are being switched on over the days that follow. If the Work option has not appeared for you yet, your wave simply has not arrived. Check back in a day or two.
Does the desktop app cost extra? No. The app itself is a free download for everyone, on every plan, Mac or Windows. What costs money is the same thing that always costs money: your ChatGPT subscription, and the usage allowance that comes with it. Tasks you give the Work agent draw that allowance down whether you started them from the desktop app or the web, so the app does not add a separate bill. The pricing guide explains how the allowance works and how to keep from burning through it.
Sources
- OpenAI: the ChatGPT Work announcement (July 9, 2026)
- ChatGPT Learn: Get started with Work and What's new (OpenAI's official docs; source for web vs. desktop reach and rollout order)
- OpenAI Help: Voice mode FAQ and 9to5Mac on the GPT-Live rollout (voice on web and mobile only)
- PCWorld and Neowin launch coverage (desktop computer use and built-in browser)
Bottom Line
Chat talks, web Work delegates, the desktop app operates. Set up the one that operates, and let the other two orbit it.