OpenAI unlocks ChatGPT Agent for Plus, Pro, and Team tiers

OpenAI’s dream of an AI that browses the web, writes code, and actually gets complex tasks done just arrived for the masses. ChatGPT Agent is now live for all Plus, Pro, and Team subscribers, including those in Europe.

The feature was technically unveiled last week as a surprise launch, but Pro users hammered the system so hard that OpenAI had to pump the brakes on the rollout for Plus and Team accounts. Now the dust has settled, and anyone paying $20 a month can take the new virtual operator for a spin.

A live stream of AI at work

ChatGPT Agent isn’t just an updated prompt window. It effectively pairs Operator—the tool OpenAI launched earlier this year for $200-per-month Pro users—with the deep dive research abilities of Deep Research.

Under the hood, Agent spins up a virtual machine loaded with two browsers (one visual, one text-only), a terminal, APIs, and ChatGPT connectors. When you set it loose, a live window pops up right inside the ChatGPT interface. You can watch the AI browse websites, click links, and jump into the terminal to parse messy information, basically like watching a coworker share their screen.

In early testing, one reviewer asked Agent to plan a late-fall running trip across Europe, demanding a half-marathon and a full marathon within two weeks of each other, in mild weather, connected by train routes. The agent took 23 minutes to crawl the web, crunching transit options and race dates into a full itinerary of eight half-marathons and seven full marathons. When told it forgot ticket links and sold-out statuses, it fixed the table in 60 seconds.

Strict limits and cautious guardrails

Agent intelligence is notoriously expensive to run, and OpenAI’s pricing structure reflects that. Plus users get just 40 Agent messages per month. Pro subscribers paying $200 per month get 400 messages, while Team limits depend on account setups. OpenAI will let you buy more credits once you burn through your quota.

Safety-wise, the model won’t handle your credit card or book hotels on its own yet. It stops short of sensitive actions and requires user confirmation before finalizing anything transactional.

Because every prompt burns one of your 40 monthly credits, users will need to treat prompts like formal project briefs. Sloppy one-liners that require follow-up corrections will eat your quota fast.

Why it matters

We think this is the clearest signal yet that AI is shifting from conversational text generators to autonomous execution engines.

For knowledge workers, researchers, and power users, ChatGPT Agent turns hours of manual link-hopping and tab-juggling into a background task. If you frequently compile market comps, scrape niche directories, or coordinate multi-leg travel logistics, this alone easily justifies the $20 monthly fee.

OpenAI was smart to put Agent inside a dedicated virtual desktop with visual feedback. Watching the agent click and type builds user trust and de-mystifies failure points, which is a massive upgrade over opaque black-box tools. But the friction is real: a 20-minute wait time rules it out for quick queries, and forty prompts a month means you have to be deliberate about when to call in the machine.