Openai launches a modern AI of general purpose in ChatGPT, which according to the company can perform a wide range of computer tasks on behalf of users. Opeli claims that the agent can automatically navigate the user calendar, generate editable presentations and slides, and run the code.
A tool called CHATGPT Agent combines several capabilities of previous OPENAI agency tools, including the operator’s ability to click websites, as well as the ability to Deep Research to synthesize information from dozens of websites in a concise research report. Opeli claims that users will be able to interact with the agent, simply supporting chatgpt in natural language.
On Thursday, Agent Chatgpt is introduced to the Pro, Plus and team plans. To activate the tool, users can choose the “agent mode” in the drop -down chatgpt tool menu.
The launch of the ChatgPT agent represents the bravest OpenAI attempt, which will still transform ChatGPT into an agency product that can take action and discharge tasks for users, and not just answering questions. In recent years, companies from the Silicon Valley, including Opeli, Google and embarrassment, have revealed dozens of AI agents who promised to do it. However, these agents and the early version proved that they are struggling with convoluted tasks and seem less convincing as products than Ultimate Vision managers around AI agents.
Having said that, Opeli claims that Agent Chatgpt is much more talented than his previous offers.
The modern agent of the company can access CHATGPT connectors, enabling users to connect applications such as Gmail and Github so that the agent can find the right information from your prompts. OPENAI claims that Agent ChatgPT has access to the terminal and can apply API interfaces to access some applications.
Opeli suggests that users can touch the chatgpt agent to “plan and buy ingredients to make a Japanese breakfast for four”, as well as “analyze three competitors and create a deck of slides.” These types of possibilities require a chatgpt agent to analyze websites, plan the course of action and apply tools – much more complicated tasks than OpenAi previously tried to solve with agents.
TechCrunch event
San Francisco
|.
October 27-29 2025
According to OPENAI, the model underlying the CHATGPT agent offers the most contemporary performance on several comparative tests.
The company claims that the CHATGPT agent model evaluates 41.6% for the last examination of humanity (application@1), a tough test of thousands of questions in over a hundred subjects. This is about twice, what O3 and O4-Mini OpenAI won during the test.
On FrontieRmath, one of the most tough known mathematics reference points, OPENAI claims that Agent ChatgPT evaluates 27.4%when he has access to tools such as a code terminal. The previous most contemporary result comes from O4-Mini, which obtained only 6.3%.
Opeli notes that he has developed a chatgpt agent with security in mind, mainly because the product presents modern possibilities that can make him more hazardous in the hands of a bad actor. OpenAI has previously warned that agency models could be more hazardous opportunities.
IN Safety report For an agent Chatgpt, Opennai claims that he was designated as “high abilities” in the domains of biological and chemical weapons, which is defined as part of OpenAI readiness as a model with the possibility of “strengthening existing paths to serious damage.” Opeli notes that there is no direct evidence of this, but he decided to apply a caution approach and activate modern security to reduce these risks.
The modern security of the CHATGPT agent includes a monitor that works in real time when users interact with the product. Opeli claims that he launches the classifier in each poem introduced to the ChatgPT agent, determining whether the request is related to biology. If so, OpenAI launches the answer of the ChatgPT agent via a second monitor, which determines whether the content can be used to cause a biological threat.
Opeli also claims that it has turned off the function of CHATGPT memory for this agent to prevent improper apply. In other parts of CHATGPT, the OpenAI memory function allows chatbot to refer to information from previous user chats. However, OpenAI claims that bad actors could apply this function in the ChatgPT agent to exclude confidential data by rapid injection attacks. The company claims that it can add this function again in the future again.
Although Agent Chatgpt sounds impressive, it turns out how capable it is in the real world. Until now, agent technology has proved to be relatively breakable during interaction with the real world. To say that, Opeli claims that he has developed a more talented model that can perform the promise of AI agents.