ChatGPT Agent is officially online! AI can autonomously operate web pages, whatever you "want" can be achieved.

OpenAI has announced the launch of a brand new upgraded version of the ChatGPT intelligent entity (ChatGPT Agent). This integrated autonomous agent AI system not only understands language and analyzes information but can now also take proactive actions, operate web pages, handle documents, and generate presentations, turning concepts into actual results.

ChatGPT Agent officially launched

ChatGPT is an intelligent entity that is an AI system capable of autonomously selecting tools and possessing the ability to think and act. It is not just a chatbot; it can also operate websites, fill out forms, create presentations, or analyze competitors through a virtual computer, significantly simplifying tedious tasks.

It integrates three major capabilities:

Operator: Web Operation Expert

In-depth research: Multi-step reasoning and information integration tools

ChatGPT Conversational Ability: Natural and Smooth Human-Computer Interaction

Users only need to briefly describe their needs, and ChatGPT will autonomously determine and use the best tools to complete the tasks. For example: "Please summarize my client presentation based on recent news" or "Analyze competitors and convert to PowerPoint."

ChatGPT Agent integration tool for completing complex workflows

ChatGPT is integrated with various online tools, including graphical web browsers, text-based browsers, and modules that can connect directly to APIs. It can switch usage modes based on task requirements:

You can use the API to retrieve data.

The operation of the website simulates clicks and inputs using a browser.

Executing integration tasks in a virtual environment, with complete circulation of background information.

It also supports real-time interaction and correction: during the task process, users can adjust their direction at any time, or interrupt and take over browser operations, providing great flexibility.

ChatGPT Agent sets new industry records in multiple benchmark tests

OpenAI conducted several standardization tests on the ChatGPT model, and the results are impressive:

1. Humanity’s Last Exam (Expert Level Quiz)

ChatGPT has achieved a new high accuracy record of 43.1%, leading other toolset models.

2. DSBench (Data Science Task Testing)

Data analysis accuracy: 89.9%, far surpassing GPT-4o (34.1%) and humans (64.1%)

Data modeling performance: 85.5%, leading overall.

3. SpreadsheetBench (Spreadsheet Operation Capability)

The accuracy of editing Excel spreadsheets reaches 45.5%, which is almost twice that of Copilot.

4. Investment Banking Model Building Task

Performance far exceeds in-depth research tools and OpenAI o3 model

5. WebArena and BrowseComp (Web Tasks and Hard-to-Find Information)

ChatGPT set new records with accuracy rates of 78.2% and 68.9%, leading similar products in the industry.

Whether in business, personal, or educational fields, the ChatGPT intelligent body can demonstrate high practicality. Actual application scenarios include:

Automatically convert dashboard data into presentations

Reschedule itinerary, meetings

Edit and update the financial spreadsheet

Planning travel and booking itineraries

Search and book services, restaurants, and other personal lifestyle matters

You can also schedule tasks to run regularly, for example: automatically generate KPI reports every Monday.

How to activate ChatGPT Agent?

To use the Smart Body feature, simply select "Smart Body Mode" in ChatGPT, and then describe the task. The system will launch the task execution window and display progress and narration in real-time. If necessary, you can:

Terminate the task

Provide new instructions

Take over the operation personally

If you are a Pro, Plus, Team, Enterprise, or Education plan user, access will be gradually opened. Pro users also enjoy nearly unlimited task quotas.

How does the ChatGPT Agent balance security?

ChatGPT has for the first time the ability to "operate websites" in practice. OpenAI has designed multiple security mechanisms to ensure user control and information privacy:

Clear authorization must be obtained before performing operations: such as shopping, making reservations, filling out forms, etc.

Sensitive tasks require "monitoring mode": step-by-step approval of each action

Proactively refuse high-risk actions: such as financial transactions, legal matters.

Prevent prompt injection attacks and abuse behavior

Browsing data is not stored, users can delete cookies and log out at any time.

Currently, while the smart body can handle the production of presentations and task integration, certain functions (such as designing exquisite presentations from scratch) are still in the beta stage, and the formatting and aesthetics may appear somewhat rough.

A new generation of presentation features will be launched in the future, improving formatting, content quality, and template application, while further optimizing data reading and presentation.

OpenAI stated that this is just the first step in integrating autonomous agent systems into ChatGPT, and that it will continue to update in the future, expanding more tools and application capabilities, further turning ChatGPT into a professional, reliable, and efficient digital work partner.

This article announces the official launch of ChatGPT Agent! AI can autonomously operate web pages, allowing you to do what you "want". It first appeared in Chain News ABMedia.