OpenAI Introduces Deep Research in ChatGPT, an AI Agent That Can Conduct Multi-Step Research

OpenAI introduced a new artificial intelligence (AI) agent in ChatGPT on Sunday that can independently research on the Internet. Dubbed Deep Research (not to be confused with Google's Gemini-powered AI agent with the same name), it can conduct multi-step research on complex topics. The AI firm stated that the AI agent is powered by one of the o3 AI models and is currently available to the ChatGPT Pro users, the company's highest subscription tier. Notably, the announcement came just days after Google rolled out its Deep Research agent to the Gemini Android app.

OpenAI Introduces Deep Research AI Agent

In a blog post, the San Francisco-based AI firm detailed its latest agentic capability in ChatGPT. Aimed at research-oriented tasks, OpenAI says it can complete tasks that would take a human multiple hours in tens of minutes. Deep Research joins the company's other AI agent dubbed Operator which can autonomously perform tasks online such as booking tickets, reserving a table in a restaurant, or buying a product.

Deep Research in ChatGPT
Photo Credit: OpenAI

Highlighting when the AI agent might be useful, the company said that Deep Research is built for users who do intensive research-based work in areas such as finance, science, policy, and engineering. It is also aimed at users who prefer “hyper-personalised recommendations” on high-ticket purchases that generally require research such as cars, furniture, and home appliances.

ChatGPT Pro users can find the Deep Research button underneath the text field next to the Search icon. Users can type their query and tap on the icon to activate the AI agent. The agent will respond with a follow-up request, asking for details about the users' preferences and more details about their requirements. Users can also add a file to provide it with more context.

After responding to the follow-up request, the AI agent will begin researching, with a thin bar showing the progress. A sidebar will also appear with a summary of the steps taken and sources used. The output can take between five to 30 minutes, and the user will get a comprehensive report once the research is done.

Powered by a version of the o3 model, the AI agent will use citations wherever information is taken from a particular source. It will also use tables and sub-headings to make the report readable. OpenAI says in the next few weeks, the company will add support for images, data visualisation, and other analytic output in these reports.

Currently, it is only available to ChatGPT Pro subscribers, but the company said it will soon be added to the Plus and Teams subscriptions. Notably, the Pro subscription is priced at $200 (roughly Rs. 17,500) a month.