OpenAI Releases ‘Operator’ AI Agent That Can Perform Tasks for You

openai operator ai agent homepage
Image Credit: OpenAI via YouTube
In Short
  • OpenAI has finally released its "Operator" AI agent. It can perform actions and accomplish tasks for you.
  • It connects to a cloud web browser where it performs actions. You can ask Operator to order groceries, book a table at a restaurant, buy tickets, and more.
  • Operator is currently being rolled out to ChatGPT Pro users in the US, starting today.

After months of anticipation, OpenAI has finally released its first “Operator” AI agent that can perform tasks for you on the web. It’s the first AI agent that is being rolled out to consumers. So far, we have seen AI models generating texts, images, videos, and audio clips. However, with the agentic AI model, OpenAI has demonstrated that AI systems can perform actions and accomplish tasks as well.

OpenAI’s Operator AI agent is currently in early research preview and it’s rolling out to ChatGPT Pro users in the US, starting today. The Pro subscription plan costs $200 per month. That said, OpenAI says Operator will be available to ChatGPT Plus users in more regions in the coming months. You can access the Operator AI agent at operator.chatgpt.com.

operator ai agent buying grocery on instacart
Image Credit: OpenAI via YouTube

As for how the Operator AI agent works, it basically connects to a web browser in a cloud environment where it performs tasks. It takes a screenshot of the screen and analyzes the visual information as to where to click next. It works similar to Anthropic’s Computer Use tool.

For deeper integration, OpenAI has partnered with many services including Uber, Instacart, OpenTable, DoorDash, eBay, and more. For example, you can ask Operator in natural language to reserve a table at a restaurant, book tickets, order snacks, buy groceries, and much more.

Operator then opens the website and starts performing tasks. You can monitor the progress and also take control of the cloud browser. For critical tasks like making a payment or booking a ticket, Operator asks the user for final confirmation. Note that you have to enter your credentials and payment information into the cloud browser.

operator ai agent benchmark
Image Credit: OpenAI via YouTube

OpenAI says that the Operator AI agent can “make mistakes” and it’s not perfect since it’s in early research preview. However, the ChatGPT maker put out some benchmarks for the Operator AI agent. In the OSWorld benchmark, OpenAI’s Operator scored 38.1% and in the WebArena benchmark, it achieved 58.1%. Both scores are lower than average human scores.

Nevertheless, Operator has kickstarted the agentic era. Google is also working on Project Mariner which can perform tasks for you in the Chrome browser, however, it’s still in development. From the progress so far, it seems 2025 is going to be the year of AI agents.

#Tags
Comments 0
Leave a Reply

Loading comments...