From shopping to scheduling—ChatGPT's new agent handles your digital tasks
What's the story
OpenAI has launched a new general-purpose AI agent, called ChatGPT agent, into its ChatGPT platform. The company claims this innovative tool can perform a wide range of computer-based tasks for users. These include automatically managing calendars, online shopping, creating editable presentations and slideshows, and even executing code. The launch is part of OpenAI's effort to make ChatGPT more than just a Q&A tool by enabling it to take actions on behalf of users.
Tool integration
A blend of functions from Operator and Deep Research tools
The ChatGPT agent is a combination of various functions from OpenAI's earlier tools, such as Operator and Deep Research. It can navigate websites and synthesize information from multiple sources into a concise report. Users can interact with this new AI agent by simply typing natural language prompts into ChatGPT. The tool is now available for subscribers of OpenAI's Pro, Plus, and Team plans.
Enhanced features
It can create slide decks and even plan group meals
The ChatGPT agent also comes with access to ChatGPT connectors, letting users link apps like Gmail and GitHub. This way, the AI can find relevant information for your prompts. OpenAI claims the new agent can plan meals for groups or analyze competitors and create slide decks. It also has terminal access and can use APIs to connect with specific apps.
Benchmark scores
ChatGPT agent model outperforms previous models on several benchmarks
The model behind the ChatGPT agent has achieved state-of-the-art performance on several benchmarks, OpenAI said. The company highlighted that the ChatGPT agent model scored 41.6% on Humanity's Last Exam (pass@1), a difficult test made up of thousands of questions across more than one hundred subjects. That's roughly double what OpenAI's o3 and o4-mini scored on the test.
Safety measures
The model is classified as 'high capability' in weapon domains
OpenAI has developed the ChatGPT agent with an emphasis on safety, owing to its new capabilities that could potentially be misused. The company has classified the model as "high capability" in biological and chemical weapon domains. To mitigate these risks, OpenAI has introduced new safeguards such as a real-time monitor that runs a classifier across every prompt entered into ChatGPT agent.