ChatGPT Agent functionality is now fully available to all Plus, Pro, and Team users. Previously, this feature was gradually rolled out to Pro users starting on July 17, and recently completed full coverage for Plus and Team users. The launch of this new AI tool marks ChatGPT's transition from a traditional conversational assistant to a more autonomous intelligent agent, offering users an unprecedented task automation experience.
ChatGPT Agent: A Breakthrough from Conversation to Action
ChatGPT Agent is the latest unified agent system introduced by OpenAI, integrating the website interaction capabilities of Operator and the web information synthesis capabilities of Deep Research, combined with ChatGPT's conversational intelligence. This system can independently complete complex tasks in a virtual computer environment, such as browsing websites, running code, and generating editable documents. Users only need to describe their needs through natural language, and the Agent will actively select the best tools to execute multi-step workflows from research to operation.
Different from traditional chatbots, ChatGPT Agent not only provides answers but also takes direct actions. For example, it can summarize meeting schedules based on user calendars, analyze competitors for companies and generate presentations, or even help plan trips or online shopping. The introduction of this feature upgrades ChatGPT from a simple knowledge query tool to an intelligent assistant capable of "doing the work" for complex tasks.
Core Features: Multi-Tool Collaboration and User Control
The core advantage of ChatGPT Agent lies in its multi-tool collaboration capability. It is equipped with the following key tools:
- Visual Browser: Simulates human operations, browsing websites through a graphical interface, clicking buttons, and filling out forms.
- Text Browser: Suitable for quick queries and information extraction, more efficient.
- Terminal and API Access: Supports running code and calling third-party APIs, suitable for programming and data processing tasks.
- Connectors: Can connect to applications like Gmail, Google Drive, GitHub, extracting user data to provide personalized responses.
Notably, ChatGPT Agent always emphasizes user control. The system requests user confirmation before performing sensitive operations (such as sending emails or submitting forms) and supports interrupting or taking over tasks at any time. In addition, OpenAI has added multiple protective measures to prevent security risks such as "prompt injection" attacks, including always-on classifiers, training to reject high-risk instructions, and an "observation mode," ensuring user data security.
Phased Rollout and Usage Limits
ChatGPT Agent is currently available only to Plus (20 dollars per month), Pro (200 dollars per month), and Team plan users, while Enterprise and Edu users are expected to gain access in the coming weeks. Pro users can use 400 Agent messages per month, while Plus and Team users have 40, with additional usage requiring credit purchase.
Due to high demand, OpenAI had delayed the launch for Plus and Team users, but as of July 24, all Plus, Pro, and Team users worldwide have unlocked this feature. Users in the European Economic Area (EEA) and Switzerland also gained full access on July 22.
It should be noted that ChatGPT Agent currently does not support memory functions to avoid potential data leakage risks. OpenAI stated that this feature may be reintroduced in the future after improving security measures.
Performance: Potential Beyond Human Capabilities
OpenAI claims that ChatGPT Agent performs well in multiple benchmark tests. For example, in the data science task benchmark "DSBench," the Agent significantly surpassed human performance; in the web navigation benchmark "BrowseComp," it scored 68.9%, an increase of 17.4 percentage points over Deep Research; in handling spreadsheets in the "SpreadsheetBench," it scored 45.5%, far exceeding Microsoft Copilot's 20%. These data indicate that ChatGPT Agent has the potential to match or even surpass human capabilities when handling complex tasks.
Application Scenarios: Multidimensional Support from Personal to Enterprise
ChatGPT Agent has a wide range of application scenarios, covering both personal and enterprise needs:
- Individual Users: Can be used for planning trips, shopping, summarizing documents, or managing schedules. For example, the Agent can arrange wedding schedules based on user emails and calendars, choose gifts, and find nearby hotels.
- Enterprise Users: Can automate the generation of financial models, competitive analysis reports, or presentations, significantly improving work efficiency.
- Developers: Through terminal and API access, the Agent can run code and analyze datasets, supporting programming and data processing tasks.
Although the Agent performs well, its speed and stability still need optimization. In testing, some tasks may take 15 to 30 minutes, and there are limitations in handling high-risk operations such as financial transactions, where it cannot directly perform payments or account operations. OpenAI states that this is intentional to ensure safety and user control.
The launch of ChatGPT Agent is not only a major upgrade in OpenAI's product line but also reflects the trend of the AI industry moving towards "agentization." Compared to traditional generative AI, the Agent is more like a "digital assistant" that understands intent, plans steps, and executes tasks. This ability not only enhances user experience but also opens up new possibilities for enterprise automation processes.
However, the widespread use of AI agents also brings privacy and security challenges. The strict security measures introduced by OpenAI in ChatGPT Agent demonstrate its concern about potential risks. In the future, as the technology matures, AI agents may replace repetitive labor in more fields, but how to balance autonomy with user control remains a key issue for the industry to solve.
AIbase believes that the full launch of ChatGPT Agent marks the transformation of AI from a "responder" to an "executor." Whether for individual users or enterprise teams, the popularization of this feature will drive further improvements in productivity. OpenAI's continuous iteration and optimization will determine whether it can maintain its leading position in the competitive AI market.