OpenAI launches ChatGPT agent to automate tasks

OpenAI has launched a powerful new AI tool—ChatGPT agent—marking a significant leap in the company’s ongoing evolution from chatbot to action-taking digital assistant. This latest offering, now available to Pro, Plus, and Team subscribers, aims to automate a wide range of computer-based tasks traditionally handled by humans.
From Assistant to Agent
Unlike earlier versions of ChatGPT, which mainly responded to user queries, the ChatGPT agent is designed to act—navigating calendars, generating presentations, writing and executing code, and even clicking through websites. Users can interact with the agent through simple natural language prompts, selecting “agent mode” from the tool dropdown menu in ChatGPT.
The release blends technologies developed in OpenAI’s prior tools: Operator, which could perform actions like navigating websites, and Deep Research, which could compile multi-source information into comprehensive reports. This fusion allows ChatGPT agent to not only understand tasks but also take meaningful actions on the user’s behalf.
Deep Integration and Broad Capability
OpenAI says the agent now supports ChatGPT connectors, enabling integration with popular apps such as Gmail and GitHub. It can pull relevant data from those platforms, use a terminal for code execution, and access APIs for third-party software.
This connectivity, combined with OpenAI’s cutting-edge model architecture, delivers unprecedented performance. For example:
- On Humanity’s Last Exam, a benchmark comprising thousands of complex questions across 100+ subjects, ChatGPT agent scores 41.6% pass@1—more than double the performance of previous models like o3 and o4-mini.
- On FrontierMath, one of the most difficult math benchmarks, it achieves 27.4% when equipped with tools like a coding terminal—compared to just 6.3% from o4-mini.
These scores suggest that ChatGPT agent doesn’t just automate simple tasks—it brings real cognitive capabilities to demanding intellectual problems.
The Road to Real AI Assistants
The launch is part of a broader industry push to develop agentic AI—systems that don’t just answer but act. While companies like Google, Perplexity, and OpenAI have all pitched this vision, early versions of AI agents often struggled to deliver. OpenAI, however, believes ChatGPT agent is a major leap forward in reliability and usefulness.
Still, the company remains cautious. With new power comes new risk, and OpenAI says it has prioritized safety in the agent’s design, especially considering its advanced abilities and potential for misuse.
What Comes Next?
The launch of ChatGPT agent could signal the beginning of a new era of AI tools that are not just assistants but true digital collaborators—handling tasks, managing information, and integrating across systems. Whether this version fully delivers on that vision remains to be seen, but it is undoubtedly OpenAI’s most ambitious move yet toward that goal.
For professionals, developers, and teams eager to offload digital busywork or supercharge productivity, the ChatGPT agent could soon become an indispensable part of the workflow.