OpenAI released its newest AI feature on Thursday. It’s called ChatGPT Agent, and it’s a model that combines the best of ChatGPT, Deep Research, and Operator. It uses natural language, which is the core of ChatGPT, can perform deep internet searches (the foundation of Deep Research), and can navigate the web on the user’s behalf, like Operator.
Unlike Operator, an AI agent that OpenAI released in early 2025, ChatGPT Agent will be available to the more affordable ChatGPT subscription tiers. Operator was reserved for ChatGPT Pro, which costs $200/month. As a ChatGPT Plus user in Europe, I never wanted to pay that much for premium AI use.
Agent changes that, as OpenAI made it available to Plus and Team users from the start. They’ll face lower limits, but they can also purchase additional access if needed.
That said, ChatGPT Agent is proving so popular with Pro users that OpenAI had to delay the rollout for Plus and Team subscribers.
ChatGPT Agent release, price, and limits
OpenAI rolled out ChatGPT Agent to 100% of Pro users, according to a post on X. But that 100% refers to Pro users outside of Europe. The region will have to wait a bit longer for access.
“Due to higher-than-expected demand, Plus and Team users will begin getting access Monday,” OpenAI said. While that’s great news for users outside of Europe, those in the region will have to wait until OpenAI is ready to roll out Agent locally.
You’ll need to pay at least $20/month to use ChatGPT Agent. That’s the cost of a ChatGPT Plus subscription.
As exciting as ChatGPT Agent might sound, access will be limited. ChatGPT Pro users will get 400 messages per month. Plus and Team users will get only 10% of that, or 40 messages. OpenAI said additional usage will be available each month through credit-based options.
While Agent merges the abilities of Deep Research and Operator into one product, only Deep Research will remain as a standalone feature. OpenAI plans to retire Operator in the coming weeks since Agent essentially performs its tasks.
What can ChatGPT Agent do?
ChatGPT Agent works in the same browser tab where you use ChatGPT or the standalone desktop or mobile apps. But if you choose Agent from the Tools menu, you’ll get a different experience.
You’ll open a virtual computer on OpenAI’s servers, where Agent handles tasks by controlling that computer. It can access a visual browser to interact with websites like you would, use a text-based browser for reasoning queries, and access a terminal for coding and APIs. Agent can also connect to your ChatGPT Connectors, like Gmail, for email-related tasks.
Once the user gives the AI a task, Agent gets to work. You’ll see a visual representation of what it’s doing. You can pause a task, change your mind, and restart it. You’ll also stay in control, as the AI will ask for input whenever it needs human approval, like making a purchase or sending an email.
OpenAI demoed ChatGPT Agent during Thursday’s livestream by giving it several tasks. As you’ll see in the clip at the end of this post, Agent had to plan wedding logistics, including finding outfits, nearby hotels, and picking a gift for the newlyweds. At the end, Agent created a report of everything.
Later in the demo, they interrupted the AI to tell Agent to also find shoes for the wedding, and the project continued.
The AI will ask questions when it needs more data from the user, then continue working until the task is done.
Separately, OpenAI engineers gave Agent a task via mobile, asking it to create a sticker and place an order online.
Finally, OpenAI showed an Agent demo where it created a presentation to showcase its own benchmarks. In another demo, they used Agent to plan an MLB road trip.
OpenAI also shared short Agent demos on X, as seen above.
These are just a few tasks Agent can perform, and I already know I can’t wait to try it. ChatGPT already saves me valuable time with many of my online tasks, and Agent could take that even further.
One more thing
Before you get too excited about automating work chores and online shopping with Agent, be aware of the new security risks that come with programs like Agent browsing the web on your behalf.
OpenAI highlighted during the demo and in its blog post that malicious individuals might create websites designed to trick the AI into doing things like handing over credit card details. Prompt-injection attacks are still possible, and while OpenAI has built in safeguards, AI jailbreaks can still happen.
Pay close attention to the kind of data you let the AI handle for you.
OpenAI is also treating ChatGPT Agent as having “High Biological and Chemical capabilities,” so it has strong guardrails to prevent misuse.
To better understand what ChatGPT Agent can do, check the video below and read OpenAI’s blog post at this link.
