OpenAI has introduced a groundbreaking addition to ChatGPT with the launch of its general-purpose AI agent. This advanced tool is designed to automate a broad spectrum of computer-based tasks, empowering users to delegate complex workflows like calendar management, presentation creation, and code execution. Building on previous innovations like Operator’s web navigation and Deep Research’s information synthesis capabilities, the ChatGPT agent aims to transform how users interact with AI by enabling seamless task completion through natural language prompts.
Available starting Thursday for Pro, Plus, and Team subscribers, the agent mode can be activated via a dropdown menu in ChatGPT. Unlike earlier AI agents from OpenAI and competitors, this iteration promises enhanced reliability for intricate tasks. Users can connect apps like Gmail and GitHub through ChatGPT connectors, granting the agent access to APIs and terminal functions. Examples of practical applications include planning multi-course meals, conducting competitive analyses, and generating editable slide decks—all executed autonomously after simple user instructions.
ChatGPT can now do work for you using its own computer.
Introducing ChatGPT agent—a unified agentic system combining Operator’s action-taking remote browser, deep research’s web synthesis, and ChatGPT’s conversational strengths. pic.twitter.com/7uN2Nc6nBQ
— OpenAI (@OpenAI) July 17, 2025
Performance benchmarks highlight significant improvements over previous models. The ChatGPT agent achieved 41.6% on the rigorous Humanity’s Last Exam (double earlier scores) and 27.4% on the FrontierMath benchmark when using tools like code execution—a stark jump from the prior 6.3% record. These metrics underscore OpenAI’s focus on advancing the model’s reasoning and problem-solving abilities.
However, the increased capabilities come with heightened safety considerations. OpenAI classified the agent as “high capability” in biological and chemical domains under its Preparedness Framework, prompting new safeguards like real-time monitoring for biological threats and disabled memory features to prevent data exfiltration. While the company emphasizes caution, it acknowledges potential future iterations may reintroduce memory functionality with improved security measures.
Despite the technical leap, real-world effectiveness remains to be tested. Past AI agents have struggled with unpredictable environments, but OpenAI asserts its latest model addresses these limitations. As the AI industry races toward fully autonomous agents, ChatGPT’s new tool could mark a pivotal step in bridging the gap between theoretical potential and practical utility.
