OpenAI has officially released the latest general-purpose AI agent in ChatGPT, which the company stated can complete a range of computer-based tasks on behalf of users.
It can automatically navigate a user’s calendar, create editable presentations and slideshows.
The recently launched ChatGPT agent is a robust tool that blends features from previous agent tools such as Operator and Deep Research.
ChatGPT agent is specifically designed to act instead of just responding; the agent can click through websites, synthesise information from several sources, and interact with apps using natural language prompts.
The agent incorporates ChatGPT connectors, allowing access to apps such as Gmail and GitHub, and features a terminal and API access to perform complex tasks.
The ChatGPT manufacturer said that it can manage real-world activities, such as planning meals or generating slide decks from competitive analysis.
According to OpenAI, the recently launched agent provides cutting-edge performance on several benchmarks.
Benchmark scores show significant enhancements: 41.6% on Humanity’s Last Exam and 27.4% on FrontierMath with tools, double or more than previous models, including o3 and o4-mini.
Due to its improved capability, OpenAI has integrated real-time safety monitors, particularly for biological content, to identify and eliminate potential misuse.
Although performance in real-world settings is being evaluated, the ChatGPT manufacturers see this release as a major step toward realising the full potential of AI agents.
Availability
ChatGPT agent is currently accessible to Pro, Plus, and Team subscribers through the “agent mode” in ChatGPT.
