ChatGPT Agent Revolutionizes Task Automation with Advanced AI Capabilities

The startup OpenAI has launched a new universal AI agent within ChatGPT, designed to carry out a wide array of computer tasks on behalf of its users.

The company claims that this agent can automatically manage user calendars, create editable presentations and slides, as well as execute code.

The ChatGPT agent incorporates several functionalities from earlier agent solutions, including the Operator’s ability to click on websites and the Deep Research feature that aggregates information from multiple sites to provide a concise analytical report.

Users can interact with this tool using natural language through a dialogue with the chatbot.

Initially, the AI agent is available for Pro, Plus, and Team subscribers. To activate it, users must select “agent mode” from the ChatGPT tools dropdown menu.

OpenAI asserts that the new ChatGPT agent significantly outperforms competing solutions. It can leverage ChatGPT connectors to link applications such as Gmail and GitHub for information retrieval and responding to inquiries. It also has access to a terminal and the capability to utilize APIs.

The digital assistant’s abilities include planning and purchasing ingredients for a Japanese breakfast for four, as well as analyzing three competitors and preparing a corresponding presentation.

OpenAI noted that the underlying model demonstrates state-of-the-art results on various benchmarks. In the Humanity’s Last Exam—a highly challenging test with thousands of questions across more than a hundred subjects—the ChatGPT agent achieved a score of 41.6%. This is approximately double the scores of o3 and o4-mini.

In one of the toughest mathematical analyses, FrontierMath, the AI scored 27.4%, surpassing the previous record held by o4-mini at 6.3%.

The startup highlighted that special attention was given to safety during the development of the ChatGPT agent, given its expanded capabilities that could pose risks if misused by malicious actors.

In a report, the model was classified as possessing “high capability” concerning biological and chemical weapons, indicating it could enhance existing means of causing serious harm. While OpenAI emphasizes there is no direct evidence of such threats, they are adopting a preventive approach and implementing additional safety measures. These include:

In July, it was reported that OpenAI reassessed its security protocols to protect intellectual property against corporate espionage amid concerns about theft by Chinese competitors.

Previously, ChatGPT was trained to connect to a greater number of internal sources for real-time contextual information.