- OpenAI unveiled Operator, its first AI agent, for ChatGPT Pro subscribers in the US.
- It can complete tasks autonomously, like booking reservations or buying groceries.
- The agent is powered by a new model built in GPT-4o called CUA.
Experts predicted that 2025 would be the year AI agents go mainstream, and OpenAI is delivering on that forecast.
On Thursday, OpenAI unveiled Operator, a system that can use a web browser to do everything from booking travel reservations to buying products.
While chatbots like OpenAI’s popular ChatGPT use generative AI to respond to queries, Operator is an agent that performs tasks autonomously.
Operator will be available Thursday in the United States for ChatGPT Pro users, a $200 monthly plan that gives users access to its latest models, including o1. In the coming months, it will also be made available to ChatGPT Plus subscribers, OpenAI’s $20 monthly subscription tier, and users in other countries.
During a livestream announcing Operator on Thursday, OpenAI CEO Sam Altman called the release an “early research preview,” adding that it will be refined over the coming months. He said OpenAI will also have more agents to launch in the months ahead.
The interface is similar to ChatGPT, in which users simply prompt Operator with a request, like “book a dinner reservation at 7 p.m.” Users can select a specific website through which they want to process the request, like OpenTable in the case of a restaurant reservation, or simply sending the request through a search engine like Google. Operator summarizes its reasoning process in a sidebar so users can quickly identify any step where it might make a mistake, which OpenAI says it’s still prone to do.
Operator is powered by CUA, a new model built on GPT-4o, Reiichiro Nakano, a member of the company’s technical staff, said in the live stream.
“It’s trained to use and control a computer in the same way humans can, by just looking at the screen and using a mouse and keyboard to control it,” he said.
The model bypasses the need for APIs, mechanisms that allow software components to communicate with each other, and “unlocks a whole new range of software we can use that was previously inaccessible,” Nakano said.
He added that the model removes “one more bottleneck in our path towards AGI.”
In a benchmark comparing how AI agents navigate common operating systems, Operator scored 38.1% compared to 72.4% for humans. In another benchmark comparing how AI agents navigate common websites, Operator scored 58.1% compared to 78.2% for humans.
- OpenAI unveiled Operator, its first AI agent, for ChatGPT Pro subscribers in the US.
- It can complete tasks autonomously, like booking reservations or buying groceries.
- The agent is powered by a new model built in GPT-4o called CUA.
Experts predicted that 2025 would be the year AI agents go mainstream, and OpenAI is delivering on that forecast.
On Thursday, OpenAI unveiled Operator, a system that can use a web browser to do everything from booking travel reservations to buying products.
While chatbots like OpenAI’s popular ChatGPT use generative AI to respond to queries, Operator is an agent that performs tasks autonomously.
Operator will be available Thursday in the United States for ChatGPT Pro users, a $200 monthly plan that gives users access to its latest models, including o1. In the coming months, it will also be made available to ChatGPT Plus subscribers, OpenAI’s $20 monthly subscription tier, and users in other countries.
During a livestream announcing Operator on Thursday, OpenAI CEO Sam Altman called the release an “early research preview,” adding that it will be refined over the coming months. He said OpenAI will also have more agents to launch in the months ahead.
The interface is similar to ChatGPT, in which users simply prompt Operator with a request, like “book a dinner reservation at 7 p.m.” Users can select a specific website through which they want to process the request, like OpenTable in the case of a restaurant reservation, or simply sending the request through a search engine like Google. Operator summarizes its reasoning process in a sidebar so users can quickly identify any step where it might make a mistake, which OpenAI says it’s still prone to do.
Operator is powered by CUA, a new model built on GPT-4o, Reiichiro Nakano, a member of the company’s technical staff, said in the live stream.
“It’s trained to use and control a computer in the same way humans can, by just looking at the screen and using a mouse and keyboard to control it,” he said.
The model bypasses the need for APIs, mechanisms that allow software components to communicate with each other, and “unlocks a whole new range of software we can use that was previously inaccessible,” Nakano said.
He added that the model removes “one more bottleneck in our path towards AGI.”
In a benchmark comparing how AI agents navigate common operating systems, Operator scored 38.1% compared to 72.4% for humans. In another benchmark comparing how AI agents navigate common websites, Operator scored 58.1% compared to 78.2% for humans.
American Dollar Exchange Rate