New ChatGPT Agent Is Here — What to Know

OpenAI has launched ChatGPT Agent — an AI that not only reasons, but takes action using a virtual computer. It can browse the web, run code, use tools, and deliver finished work like presentations or spreadsheets.

What It Does

ChatGPT Agent handles full tasks, like:

• Researching a topic
• Creating a slideshow
• Booking ingredients online
• Summarizing your inbox
• Planning meetings or events

It can open websites, log in (with your help), download files, analyze data, and ask questions when needed — all in one flow.

How It Works

The system combines:

• Operator: web interaction
• Deep Research: info synthesis
• ChatGPT: reasoning and conversation

It uses tools like:

• A visual browser
• A terminal for code
• APIs and connectors (like Gmail or GitHub)

It picks the best tool for the job and switches fluidly between them.

You Stay in Control

You can:

• Approve actions (like purchases)
• Take over the browser
• Pause, stop, or steer the task
• Clear browsing data anytime

Real Results

ChatGPT Agent sets new records:

• Humanity’s Last Exam: 44.4% pass@1
• FrontierMath: 27.4%
• DSBench (data science): ~90%
• SpreadsheetBench: 45.5% (.xlsx)
• WebArena: 78.2% real-world task accuracy

In investment banking tasks, it beat both GPT-4 and human baselines in key metrics.

Safety First

With great power come strict guardrails:

• Requires your approval for real-world actions
• Refuses risky tasks like bank transfers
• Resists prompt injection attacks
• Keeps login inputs private
• Lets you wipe all data with one click

It's classified under high biosafety protocols and tested by domain experts.

How to Use It

Available to:

• Pro (400 messages/month)
• Plus & Team (40 messages/month + credits)

Find "Agent mode" in the tools menu. Just describe your task — the agent takes it from there.

What’s Next

Slide generation will improve. Workflow speed and polish will get better. Less oversight will be needed over time — but safety will remain.

This isn’t just smarter AI. It’s your new digital teammate.