The ChatGPT agent was OpenAI’s way to let a generative AI chatbot do a task instead of only describing it, and ChatGPT’s agent mode was the switch. As of October 2026, though, the help page for the agent opens with a plain line: “ChatGPT agent is no longer available.” It points to ChatGPT Work for longer, multi-step tasks and to a cloud tool for web workflows.
This guide covers what the agent was, its plans and limits, what replaced it and how to start, and sets ChatGPT beside the agentic tools from Claude and Grok. Every fact comes from the makers’ own pages, read on 22 October. DigitaVision is a separate company, not affiliated with OpenAI, Anthropic or xAI.
Key takeaways
- The agent launched in July 2025. Its help page now says it is no longer available and recommends ChatGPT Work.
- That page still lists monthly message limits: Plus 40, Pro 400, and Business and Enterprise 40.
- Work runs on web and mobile for Plus, Pro, Business, Enterprise and Edu, and the cloud tool comes with paid plans except Free and Go.
- Claude has Cowork, and Grok lists agent requests and cloud agents. They are not the same product.
What is the ChatGPT agent, and what is agent mode in ChatGPT?
The ChatGPT agent is a feature of ChatGPT that works in a virtual browser on the agent’s own virtual computer. The launch post says it joins Operator’s skill at interacting with websites, deep research’s skill at synthesizing information and the chatbot’s conversational fluency. So what is agent mode in ChatGPT? It is the switch that turned a chat into a job the AI carried out while you watched. The post says it handles complex tasks from start to finish.
The post is now marked “outdated”, and the help page says Operator was folded into the agent. The Operator website is no longer accessible.
Key components of ChatGPT agents: browser, API and use cases
The launch post lists a toolbox: a visual browser that works through a graphical interface, a text-based browser for a simple query, a terminal for code execution and direct access to APIs. Connectors such as Gmail and GitHub let it read information that fits a request. Using multiple tools, it was built for complex tasks, and you could ask the agent to find a restaurant or build a report. It could:
- Interact with websites to complete tasks using the site itself: filter results and fill out forms, with a log-in handed to you through takeover.
- Run code, analyze a dataset and build slide decks and spreadsheets, which suits research, reporting and light coding.
- Plan a trip, a bit like a travel agent, or compare competitors, the real-world tasks OpenAI gave as examples.
- Repeat a finished task daily, weekly or monthly, a simple automation and a form of automating tasks, so you can automate a routine report.
The post cites benchmark scores for the model but admits limits: slide outputs could feel rudimentary, and mistakes happen. Typical use cases were research, trip planning and meeting prep. It calls the risk profile higher than a normal chat because the agent works directly with your personal data. Agents can access sensitive data such as emails and files once you sign them in.
Which plans got ChatGPT agent mode, and what were the usage limits?
At launch it reached Pro, Plus and Team, then Enterprise and Edu in August 2025, per the release notes. The help page still lists the limits below even though the agent is gone, so read them as legacy figures.
| Plan | Monthly agent messages |
|---|---|
| Plus | 40 |
| Pro | 400 |
| Business and Enterprise | 40 |
| Business and Enterprise, flexible pricing | 30 credits per message |
Only your first request counted, not clarifications or sign-in steps, and scheduled runs counted too.
Why OpenAI now points to ChatGPT Work
The successor is ChatGPT Work. Its help article calls Work an agent designed for longer, multi-step work and finished deliverables. The Work page says this AI tool uses your apps, files and tools to create documents, decks, sheets and reports, run recurring jobs and check in when judgment matters. A Plan mode shows a step-by-step plan first.
As of October 2026, per OpenAI:
- Work runs on web and mobile for Plus, Pro, Business, Enterprise and Edu.
- Free and Go get limited Work in the desktop app.
- The cloud tool is on paid plans in supported regions, excluding Free and Go.
Codex stays the tool for coding, and Plus plans include GPT-6 among the OpenAI models, per the pricing page. The help page says some new models, such as GPT-6.1 Sol, are for Work and Codex only, not regular chats. Work follows the same usage structure as Codex, so there is no fixed message count, and its ability to handle long jobs depends on the allowance. A long task uses more of the allowance than a short one, and optional credits cover extra use.
Benefits of using a ChatGPT tool like Work
Using a ChatGPT agent was meant to be simple: agents operate on live sites, and agents handle several steps without a new instruction for each one. No technical skills are needed, which makes ChatGPT easier to use. You can interrupt or stop a task at any point.
How to use ChatGPT agent today: start a task in Work
The old route was the tools menu or typing /agent, and the help page still describes it. Because the page says the agent is gone, the working route is Work:
- Open ChatGPT on the web or in the mobile app and select Work.
- Start a chat or open a project, and add the files and context to use.
- Describe the outcome in natural language, with the website and any constraints.
- Approve website access if asked, and use the secure sign-in form when a site needs a log-in.
- Follow progress, confirm anything important and check the result and its sources.
You do not pick the cloud tool yourself; ChatGPT’s own routing chooses when to use it. If a task gets stuck, you can take control of the browser.
Safety controls, prompt injection and what the chatbot cannot do
The help page lists confirmations before high-impact actions, refusal patterns, prompt injection monitoring and a watch mode on some sites. The cloud tool asks before visiting a new website, and you can tailor the permissions: Always ask, Auto approve or Always allow, the last “not recommended”. It asks you to confirm things like a booking or a payment, and a password typed into the secure form is not visible to the model. Human oversight stays part of the design. Enterprise owners can configure which apps and sites are allowed, and the old page says the agent could not read synced apps such as Google Drive directly.
The maker says these measures do not remove every risk, and the limits are real:
- Some websites block automated visitors, even when the page works for you normally.
- It is not a customer support bot or a custom agent built for one business.
- Work on web and mobile cannot reach files on your computer; the desktop app can, with permission.
- Prompt injection, where a page tries to trick the AI agent, remains a risk for agentic AI systems generally. Avoid sensitive logins and vague requests such as “handle my email”.
ChatGPT Work versus Claude and Grok AI agents
These are the closest products on the makers’ pages. They differ, so compare the jobs you need done, not the names.
| ChatGPT Work | Claude Cowork | Grok | |
|---|---|---|---|
| What it is | Multi-step work and deliverables | Hands off tasks across your files and tools | Chat with agent requests, Grok Bot and cloud agents |
| Web tasks | Cloud tool, paid plans except Free and Go | Built-in browser in the desktop app on Pro, Max and Team | Not described on the pricing page |
| Entry plan | Plus | Pro, $17 a month billed yearly or $20 monthly | Premium $30 lists Grok Bot and cloud agents |
Grok’s Free card lists “limited agent requests”, and xAI recommends Super for daily agent users. For ChatGPT plan choices, see our ChatGPT plans guide.
Where DigitaVision fits
DigitaVision does not sell the retired agent, and no listing restores the old feature. The catalog does list paid ChatGPT accounts, and the maker puts Work and the cloud tool on paid plans.
ChatGPT Plus Personal costs $17.99 for 30 days. It is a private login of your own, set up by the DigitaVision team, and the maker lists Work on Plus for desktop, web and mobile. Setup usually takes 2 to 4 hours, up to 24 hours in rare cases.
ChatGPT Business Personal costs $16.99 per 30 days. It is a Business seat sent as an invite to your personal account. The Business overview lists Work for both seat types, so ask support which one your invite carries.
You can pay in USD or BDT, and support runs every day from 8 AM to 3 AM Bangladesh time. To judge the value first, read our Plus worth-it guide, or see the ChatGPT free trial guide for the free plan.
Is ChatGPT agent worth it in 2026? Who it suits and who it does not
The old agent is gone, so the question is whether Work and its cloud tool are worth a paid plan. They suit people with repeat research, reports and data jobs who want a draft waiting for review and will check results first. They do not suit anyone who wants a hands-off robot for payments or critical tasks, or who needs a fixed message count. For light use of ChatGPT AI as a plain assistant, the free plan is enough.
FAQ
Is ChatGPT agent still available?
No. As of October 2026, the help page says it is no longer available and points to ChatGPT Work and the cloud tool.
What is agent mode in ChatGPT?
It was a tools-menu option, also started with /agent, in which the AI used its own virtual computer to finish a task. Work now covers that job.
Which plans include ChatGPT Work?
OpenAI lists Work on web and mobile for Plus, Pro, Business, Enterprise and Edu, with limited desktop access on Free and Go.
Is ChatGPT agent free?
The help page says agent mode was only for paid plans, and the cloud tool also excludes Free and Go.
Can I get ChatGPT Work from a DigitaVision account?
The Plus and Business listings are paid ChatGPT accounts, and the maker’s pages decide what each plan includes.
