I tried in -depth research of Chatgpt. His relationships are worth expected.

• I tried the in -depth research tool, which can perform complex search tasks in a few minutes.
• Propelled by a version of the O3 model from Openai, Deep Research acts independently to work on tasks in several stages.
• I used it to seek the history of prices and their impact on the pricing of consumer goods.

At a time when we are used to obtaining answers to AI requests in a few seconds, I can say with confidence that the in -depth research tool is worth five to 30 minutes.

In addition, it is a fascinating overview of how AI “thinks” thanks to a complex research assignment and executes it.

In -depth research is an agentic AI tool, which means that it can act independently to solve a task in several stages. OPENAI says on his website that the tool “performs in tens of minutes which would take a human of many hours”, and it is not an exaggeration.

Although Chatgpt can provide an in -depth report on a subject if you ask for one, he does not finish the type of research in several steps that in -depth research does. This capacity is not relevant for each request, but it is useful if you want to explore the story or the nuances of a subject. OPENAI says that in -depth research is particularly useful for “niche and non -intuitive information that would require browsing many websites”.

The AI ​​tool is fueled by a version of the OPENAI O3 model to come designed for web navigation and data analysis and uses “reasoning to search, interpret and analyze massive quantities of text, images and PDF on the Internet”.

OPENAI initially rendered the AI ​​agent available for professional users who pay $ 200 per month, but announced its deployment to Chatgpt Plus, Team, Edu and Enterprise users on February 25. PRO users obtain 120 depth research requests per month, while users have access to 10.

How does it work

You can generate an in -depth search report by selecting the “Deep search” button in the chatgpt text bar. Then you write an prompt, which the assistant will clarify to ensure that she is looking for the right information.

Related stories

For example, I asked him to compile a research report on the history of American trade policy concerning the prices and the impact they had on the pricing of consumer goods. In -depth research has responded by asking for more information, especially over the period and the industries that I wanted to covered, the type of analysis and the level of detail.

I told him to seek the 20th century, emphasizing the first mandate of President Donald Trump and the impact on general consumer goods. I also asked him to generate an economic analysis based on data with including tables.

The AI ​​reiterated my instructions and then went to work.

As the assistant compiles the report – which can take about five to 30 minutes depending on the request – you can check their progress and observe their research process. One of the most interesting parts of this process was to see how research has evolved by revealing new information.

Here you can see his reflection process towards the end of his eight -minute research.

In -depth research has sought and analyzed sources, then generated an in -depth report on its results, with online quotes included in the various sections. The central section was dense, but the conclusion gave a solid summation, and the tables were also insightful.

I spent a good part of my 10 uses of research exploration on all long -term effects of various health trends, such as ozempic use. He provided in -depth overviews and included isolated incidents with limited research, which I found interesting. I also used it to investigate the story of my family. He correctly traced the origin of my surname and what I know about the history of my ancestors before immigrating the United States there are generations.

The reports I read were fascinating – and precise on the basis of the links mentioned that I checked. However, he cited Wikipedia a good amount. You can see in the screenshot below of the price request that Wikipedia was one of the sources used to compile the report.

OPENAI said in its advertisement that the model feeding deep research had obtained a score of 26.6% on the last examination of humanity, an assessment designed to assess AI on various subjects. For the context, Chatgpt-4o scored 3.3%.

However, Openai added that the tool “can sometimes hallucinate” or “make incorrect inferences”, although a lower rhythm than that of other chatgpt models.

“It may be difficult to distinguish the information authorizing rumors and currently shows a weakness in the calibration of confidence, having often failed to transmit uncertainty with precision,” said Openai in his announcement.

Almost all the tools of AI that I tried initially impressed me, but most of them failed to be part of my routine. In -depth research is not something that I thought I was losing track of use, but I burned my 10 requests in a few days – and I found myself wanting more.

I do not know if I would pay $ 20 per month for 10 of these requests, since it seems limited enough for a paid service. However, if I had specific legal questions or if I wanted to seek medical treatments, I would probably find the cost.

For the moment, I am delighted to make the most of my next batch of 10 requests.