ChatGPT’s New Deep Research Does Research and Analysis for You

This Monday, February 03rd, at an event held in Tokyo in partnership with Softbank, OpenAI announced the launch of the Deep Research, an innovative tool of ChatGPT. This new feature is designed to perform detailed research and analysis in a fraction of the time it would take a human. The introduction of Deep Researchh comes amid growing competition in the chatbot market, highlighted by the boom in the last week of the Chinese chatbot DeepSeek, which became known for its low operating costs and resource efficiency. Check out everything about the new product:

How “deep research” works

A "in-depth research" from OpenAI is a model-powered tool OpenAI o3, optimized for web browsing and data analysis, this tool can work independently to find, analyze e synthesize information from hundreds of online sources. From a simple prompt, in-depth research can create comprehensive reports at the level of a research analyst, saving users time and effort.

One of the main characteristics of in-depth research is its ability to synthesize knowledge, something essential for the creation of new knowledge. This advance marks a significant step in the development of General Artificial Intelligence (AGI), a long-standing goal of OpenAI. Deep search is trained using reinforcement learning on complex navigation and reasoning tasks, allowing it to learn to plan and execute multi-step trajectories to find the data it needs.

Additionally, deep search is equipped to react to real-time information, adjusting its search as needed. It can browse user-uploaded files, generate and iterate graphs using the tool Python, embed graphics and images from websites into your responses, e cite specific phrases or passages from your sources. Thanks to their advanced training, in-depth research reaches new heights in a variety of public assessments focused on real problems.

Tool results and new response formats

The results of Deep Research, initially, impress. In the evaluation “humanity's last exam”, which tests artificial intelligence on a wide range of subjects, the model achieved an accuracy of 26,6%, which may be considered a low rate, but it surpasses all other models evaluated. This assessment includes more than 3.000 multiple-choice and short-answer questions in more than 100 subjects, from linguistics to rocket science, classics and ecology. The biggest gains were seen in chemistry, humanities, social sciences and mathematics.

Compared to other models, the “deep search” of the ChatGPT stands out significantly. While models like GPT-4o, Grok-2 e Sonnet Claude 3.5 obtained accuracies of 3,3%, 3,8% e 4,3%, respectively, the new feature surpassed these numbers by a wide margin. This superior performance can be explained by the model's ability to browse the web and use tools such as Python for data analysis, citing specific sources and integrating graphs and images generated from websites into your responses.

Furthermore, Deep Research da OpenAI also obtained good results in benchmarking public GAIA, which evaluates artificial intelligence on real-world issues. Across three difficulty levels, deep research has demonstrated advanced capabilities in reasoning, multimodal fluency, web browsing e use of tools.

A OpenAI is further expanding the capabilities of Deep Research beyond text, bringing new response formats. Soon, the tool will be able to display embedded images, graphics e data visualizations, making responses richer and more intuitive. These improvements aim to facilitate the understanding of complex information, especially in data analysis. In addition, OpenAI intends to integrate the tool with specialized data sources, including exclusive content and subscription platforms, further expanding the depth and accuracy of the responses provided.

How to use

To use the functionality Deep Research No. ChatGPT, it's simple. First, select the option Deep Research in the message composer. Then enter your detailed query, explaining what you need. Additionally, you can attach files or spreadsheets to provide more context to your question. ChatGPT will then start running the search and a sidebar will appear showing a summary of the steps taken and the sources used.

The process of in-depth research can take from 5 to 30 minutes to be completed, as the tool is dedicated to thoroughly exploring the web. During this time, you can dedicate yourself to other tasks or even step away from the computer, as you will be notified as soon as the search is finished. The search result is delivered in the form of a detailed report in the chat. Soon, OpenAI will also add embedded images, data visualizations, and other analytical results to make reports even clearer and more contextual.

Compared to GPT-4o, which is best suited for real-time, multimodal conversations, “deep search” excels when depth and detail are crucial. For complex, domain-specific queries, deep search’s ability to perform extensive exploration and cite every claim makes all the difference. This results in well-documented, verified answers that can be used as a final product, as opposed to a quick summary that the GPT-4o can provide.

What did you think of the new tool? Deep Research? Tell us about it Comment!

See also:

Sources: TechCrunch, ChatGPT e The Verge.

reviewed by Gabriel Princessval in 03 / 02 / 2025