Operator is a web application designed to perform simple online tasks within a browser, such as booking concert tickets or placing online grocery orders. The app is powered by a new model called the Computer-Using Agent (CUA), built on OpenAI’s multimodal large language model GPT-4o. Currently available at operator.chatgpt.com for ChatGPT Pro subscribers in the U.S.—a premium service costing $200 per month—OpenAI plans to expand access to other users in the future.
OpenAI claims that Operator surpasses competing tools, such as Anthropic’s Computer Use (based on Claude 3.5 Sonnet) and Google DeepMind’s Mariner (a browsing agent using Gemini 2.0). Operator aims to automate tasks like booking travel accommodations, making restaurant reservations, and online shopping. Its interface organizes tasks into categories such as shopping, delivery, dining, and travel, each enabling specific automation functions.
When activated, Operator opens a dedicated web browser within a small pop-up window to execute tasks, providing explanations of its actions as it works. Users retain full control of their screen since Operator operates within its own browser.
Operator’s capabilities stem from the CUA model, which integrates the vision capabilities of GPT-4o with advanced reasoning from other OpenAI models. Trained to interact directly with website interfaces, the CUA can navigate menus, press buttons, and fill out forms without relying on developer-facing APIs. OpenAI has partnered with companies like DoorDash, eBay, Instacart, Priceline, StubHub, and Uber to ensure Operator adheres to their terms of service.
This is cool and scary at the same time.