On Thursday, OpenAI unveiled its latest innovation: a smart assistant integrated into ChatGPT, which the company says is capable of performing complex tasks that go beyond conversation and answering questions.
According to the company, the smart assistant can manage schedules, prepare presentations and financial reports, and even assist with life-planning tasks like shopping for ingredients for a Japanese recipe for four people.
Yash Komar, product lead for the new assistant, explained that the goal was to build a product offering practical, wide-ranging solutions for businesses and organizations.
He noted that the new tool has the ability to fill out online forms, use APIs to access services like Google Drive and SharePoint, and operate a command-line terminal to execute commands.
A Fusion of Past Expertise and Advanced Capabilities
The launch merges capabilities from previous tools developed by the company. The new assistant combines the "Operator" tool's feature for visually browsing and interacting with websites with the "Deep Research" tool's ability to process and analyze information from numerous sources to deliver accurate research summaries.
ChatGPT can now do work for you using its own computer.
— OpenAI (@OpenAI) July 17, 2025
Introducing ChatGPT agent—a unified agentic system combining Operator’s action-taking remote browser, deep research’s web synthesis, and ChatGPT’s conversational strengths. pic.twitter.com/7uN2Nc6nBQ
On a technical level, OpenAI reported that the model powering the new assistant has achieved advanced performance on several difficult benchmark tests.
For instance, the model scored 27.4% on the FrontierMath test, one of the most challenging mathematics benchmarks known, when equipped with the necessary tools, compared to just 6.3% for the previous model.
Performance: Theory vs. Practical Application
Despite these results, the company acknowledged that real-world performance is the most important test.
In a practical experiment, research lead Iza Fulford mentioned that the assistant took nearly an hour to complete a large order of donuts, but she described the experience as easier than doing it herself.
Komar added that simple tasks, like planning a dinner appointment, might take five minutes, while complex research tasks, such as preparing a presentation on Nvidia's earnings, could require about 25 minutes.
Safety Aspects of the New AI Assistant
OpenAI has also placed significant emphasis on safety and security. As a precautionary measure, the company has classified the new model as "highly capable" in potentially sensitive areas, such as biological and chemical weapons.
To address these potential risks, the company implemented new safeguards, including a real-time monitoring system to analyze user requests and identify any potentially dangerous content.
Furthermore, the company decided to disable the "memory" feature in the assistant at launch. This decision was made to prevent its exploitation in prompt injection attacks that could lead to sensitive data leaks.
However, the company has not ruled out reconsidering the activation of this feature in the future after further study.
The new assistant will begin rolling out to Pro, Plus, and Team plan subscribers, with access for Enterprise and Education subscribers to follow later this summer.
