OpenAi has presented an AI agent developed for Allow ChatGPT users to conduct thorough investigations using chatbot. They have called him Deep Researchand is designed for users who do tasks that require detailed, precise, reliable and exhaustive information in areas such as finance, science, institutional policies or engineering.
The company, which seems to have stepped on the accelerator after the emergence of Deepseek, also points out that it could also be useful to do research in the face of certain types of purchases, usually of products or services that require high investment. For now, yes, it is not available in the United Kingdom, neither Switzerland nor in EU Economic Space. Nor has an approximate date been offered for availability in the area.
Deep Research is therefore designed for users who do not need a short and fast response, or a summary of a text, but for those who need a repossed response, and taken from the search on several websites and other types of sources, and the Subsequent synthesis of the localized information.
For now, this AI agent is only available for Chatgpt Pro subscribers, and limited to 100 questions per month. Openai does not plan to be alone here, since it has already confirmed that within a time it will extend the use of Deep Research to the subscribers of the Plus and Team plans, to later extend it to those of the Enterprise plans.
On the dates, the company calculates that the deployment to the subscribers of the PLUS Plan approximately in one month will be available. They have also confirmed that the limits to the questions that can be asked to this agent will be much less restrictive in the future. In addition, it cannot be used yet from the chatgpt app, and can only be activated from the web. Its integration into the mobile app is planned by the end of this month.
Openai has provided this agent with a special version of its IA O3 reasoning model, trained by learning reinforcement in real world tasks. This type of learning is responsible for “teaching” a model by proof and error until a specific objective is achieved. As the model is closer to the objective, it is receiving rewards, than in theory, make it better and more capable of task as it progresses.
According to the company, the O3 version used for Deep Research is optimized for both web navigation and data analysis. They also point out that it takes advantage of the reasoning to search, interpret and analyze large amounts of texts, images and PDF files on the Internet.
Depending on the information I find, it changes objectives and guidance. In addition, he is able to work with the information provided by the user, and to integrate graphics and images from websites in the answers, in addition to citing phrases and paragraphs of their sources.
To use it you just have to choose the agent in the interface and ask the question. The user has the option to attach files of different types, and after pressing Intro it is only necessary to wait. The agent It takes a while to answer (between 5 and 30 minutes according to OpenAI), depending on the size of the investigation.
Chatgpt will send a notification to the user when you have the answer to your question. At present, it only responds through texts, but Openai has already confirmed that it intends that it can also respond with images, data visualizations and other elements. They also plan to connect to more specialized data sources.
Of course, for now OpenAi has recognized that Deep Research has its limitationssometimes you make mistakes. For example, you can have difficulty distinguishing reliable information from lso rumors. You can also fail when you transmit when you are not sure of something. In addition, you can have format failures in reports and appointments.
