OpenAI Releases Operator AI Agent, Ushering in a New Era of Paid Consumer Services
OpenAI recently launched a new AI agent named Operator, the company’s first paid subscription service aimed at consumers. This agent can autonomously execute complex tasks such as online shopping, booking restaurants, and purchasing plane tickets by interacting with a browser, and is built on the GPT-4o model. Operator is currently in a research preview phase, accessible only to Pro users.
Core Features and Technical Foundation of Operator
Operator is designed as a computer-use agent capable of operating a browser like a human, including actions like clicking, typing, scrolling, and waiting for pages to load. It learns user preferences by observing past behaviors, such as prioritizing a specific hotel chain or payment method when booking accommodations. Users can simply describe a task in natural language, such as “Book me a business class ticket from New York to Los Angeles,” and Operator can independently complete the entire process.
The agent is based on the GPT-4o model and is trained on a computer-use dataset, supporting asynchronous task processing. Users can monitor its progress in a dedicated application and intervene or stop the operation at any time. OpenAI states that Operator represents a significant advancement toward more autonomous AI agents.
Access and Pricing Model
Operator is being launched as a standalone application, currently available only to Pro users in a few countries, including the United States and Canada, with a weekly usage quota of 50 tasks. Once the quota is exceeded, users must wait for it to reset or upgrade their subscription. OpenAI plans to introduce a Operator Pro paid plan, offering unlimited tasks, high-priority access, and more advanced features. The specific pricing has not yet been announced, but it will be the company’s first purely consumer-paid service, distinct from enterprise-level subscriptions like ChatGPT Plus.
Potential Risks and Safety Measures
While Operator enhances productivity, OpenAI emphasizes that it is not perfect and can still make mistakes or produce unintended consequences, such as incorrect bookings or information leaks. Therefore, the company requires users to supervise the process at all times and provides a one-click stop button in the interface. To mitigate privacy risks, Operator will not access emails or files and is limited to browser operations.
According to OpenAI’s VP of Research, Kevin Weil, this launch marks a shift for AI agents from being assistive tools to independent executors. Multiple media outlets, including The Verge and Reuters, have confirmed that Operator has begun small-scale testing, with user feedback indicating a success rate of over 85% for simple tasks. This technology is poised to reshape daily workflows and drive the adoption of AI in the consumer sector.
