OpenAI Launches Advanced AI Agent in ChatGPT

OpenAI has unveiled a new general-purpose AI agent within ChatGPT, designed to perform a wide range of computer-based tasks for users. Dubbed the ChatGPT agent, the tool combines capabilities from OpenAI’s previous agentic tools, enabling users to automate tasks like navigating calendars, generating presentations, and running code.
Users can interact with the agent using natural language prompts, and it integrates with apps like Gmail and GitHub through ChatGPT connectors. The agent can also access terminals and APIs, allowing it to perform complex tasks such as analyzing competitors or planning a Japanese breakfast for four.
The ChatGPT agent is rolling out to subscribers of OpenAI’s Pro, Plus, and Team plans on Thursday. Users can activate it by selecting “agent mode” in ChatGPT’s tools menu. OpenAI emphasizes that this launch marks its boldest attempt to make ChatGPT an agentic tool capable of taking actions and offloading tasks, rather than just answering questions.
Performance benchmarks highlight the agent’s capabilities: it scores 41.6% on Humanity’s Last Exam and 27.4% on FrontierMath when using tools, significantly outperforming previous models. However, OpenAI acknowledges potential risks, designating the model as “high capability” in domains like biological and chemical weapons. To mitigate risks, OpenAI has implemented safeguards, including real-time monitoring of prompts related to biology and disabling the memory feature to prevent misuse.
While the ChatGPT agent shows promise, its real-world effectiveness remains to be seen. OpenAI claims it has developed a more capable model, but agent technologies have historically struggled with complex, real-world interactions.

Published: 7/18/2025