OpenAI launches ChatGPT Agent mode, for tasks both easy and tough

The new model will be able to do stuff online, like check for airplane tickets and scan your Gmail, as well as running code, doing research and create Powerpoint, Excel files based on the results.

The ChatGPT Agent should be available in the model selector at the bottom of the ChatGPT window in the app for Pro, Plus and Teams users as of Thursday, but rollout will be «over the next few days,» or over «the coming weeks» for Enterprise and Education users.

EDIT:«We are still working on enabling access for the European Economic Area and Switzerland,» OpenAI says on their website — and as of July 23, they are saying it is on the way.

ChatGPT can now do work for you using its own computer.

Introducing ChatGPT agent—a unified agentic system combining Operator’s action-taking remote browser, deep research’s web synthesis, and ChatGPT’s conversational strengths. pic.twitter.com/7uN2Nc6nBQ

— OpenAI (@OpenAI) July 17, 2025

— ChatGPT can now do work for you using its own computer, handling complex tasks from start to finish, OpenAI says in its presentation.

The best of Deep Research/Operator
The problem was that Operator, OpenAIs agent, couldn’t do research or dive deep into topics, while Deep Research couldn’t interact with websites, click around and do things like refining a query or sharpening results.

«Agent mode» now does both; it clicks, filters and gathers more precise results.

It can connect to Calendar, Gmail and GitHub, using the ChatGPT Connectors system, and it can perform actions on websites — even ones you are logged into.

Doesn’t make important decisions
The agent will not take any consequential action, like actually ordering those airplane tickets on its own — and will instead ask for your permission. You can stop tasks at any point.

It will also ask you for guidance or additional details during operations to ensure the task is aligned with your goals.

ChatGPT in «Agent mode» also scores pretty nicely in Humanity’s Last Exam — hitting 41,6% with tools, which is pretty much in the top tier. The results aren’t public on HLE’s website just yet, so don’t pass judgement until it gets published.

— Today’s launch is just the beginning. We’ll continue to iteratively add significant improvements regularly, making it more capable and useful to more people over time, says OpenAI.

Read more. Launch page at OpenAI, writeups on TechCrunch, MacRumors, and Every.to gives it a run.