OpenAI commenced the rollout of its artificial intelligence (AI) agent Operator in various countries on Friday. The agentic tool was previously available in preview to ChatGPT Pro users located in the US, following its introduction by the company in January. With a specialized browser, the Operator serves as a general-purpose AI agent capable of autonomously completing online tasks based on prompts provided by the user. Although the AI agent is currently accessible in numerous key regions, the AI firm stated that it will not be accessible in European nations.
OpenAI’s Operator AI Agent Finally Rolls Out
In a message on X (previously known as Twitter), OpenAI’s official account announced the deployment of the AI agent. The company indicated that Operator will now be accessible to ChatGPT Pro users in “Australia, Brazil, Canada, India, Japan, Singapore, South Korea, the UK, and most areas where ChatGPT is available. ” Importantly, the tool is already accessible to individuals residing in the US.
Nonetheless, the company also mentioned that European nations will need to wait a little longer before gaining access to the AI agent. “Still working on making Operator available in the EU, Switzerland, Norway, Liechtenstein and Iceland,” the message added.
At the time of its launch, OpenAI clarified that Operator is powered by the Computer-Using Agent (CUA), which is a unique AI model that derives computer vision abilities from GPT-4o and advanced reasoning capabilities from an unspecified OpenAI model (which could potentially be the o3 model). The AI agent underwent post-training using reinforcement learning.
The CUA possesses agentic abilities and can communicate with graphical user interfaces (GUIs) such as buttons, menus, and text fields displayed on the screen. It also has a specialized browser that enables it to perform actions in a separate environment, allowing the human user to engage in other tasks.
Operator takes both text and images as input. While executing tasks, it analyzes the raw pixel data of the screen and carries out actions using a virtual keyboard and mouse, which are also provided in the sandbox environment. The AI agent is reported to manage multi-step tasks, navigate through errors, and adjust to unforeseen situations.
