Join our daily and weekly newsletters for the latest updates and the exclusive content on AI coverage. Learn more
In case you missed it in favor of Grammy Awards last nightOpenai surprised the world late Sunday evening with the announcement of his new “Deep research”Modality, an AI agent available for users of the Chatgpt Pro subscription plan ($ 200 / month) designed to save humans by seeking, well,” deeply “and largely on the web for given subjects and By compiling professional quality relationships through specialized fields from businesses to science, medicine, marketing and more.
Chatgpt pro users (and soon, Chatgpt Plus, Team, Enterprise and EDU) in the United States will be able to access in-depth research by clicking on the option under the prompt / compose bar at the bottom of the Chatgpt website and Applications.
Sam Altman, CEO of Openai, described the functionality in a series of messages on his Personal account on social network x as “as a superpower; Experts on request! “He added:” It’s really good and can do tasks that would take hours / day and cost hundreds of dollars. “
Deep Research relies on the O series of Openai of reasoning models, by specifically pulling the full O3 model which will soon be published (a smaller and less powerful model, O3-Mini, has just launched on Friday). The complete O3 model can analyze large amounts of information and integrate text, PDF and images into a coherent analysis.
In a Livestream posted on Youtube And available for replay on asking, Mark Chen, head of the research of Openai Frontiers, explained that “deep research is a model that does research in several stages on the Internet. He discovers the content, synthesizes content and reasons on this content, adapting his plan because he discovers more and more information. »»
Chen also stressed the importance of innovation for Openai’s vision: “This is at the heart of our AGE roadmap. Our ultimate aspiration is a model that can discover and discover new knowledge for itself. »»
The launch of The Deep Research marks the second of official Openai agents after the launch of its browser and its cursor control operator earlier this month. And jOshua Amamiam, head of the mission to the alignment of the Stargate Command in Openai wrote on XThe two models can help better define the concept of an “AI agent” – a popular but nebulous term today among companies – far beyond the company or these specific use cases.
“I have the impression that the term” agent “wandered in the desert for a while”, Achaim wrote. “He had no land or examples to point. But agents like the operator or in -depth research give a certain form to this concept. An agent is an AI for general use that makes one or more workflows using tools for you. »»
Openai’s in -depth research obtains a new highest score on the “Last examination of humanity” Benchmark
In -depth research has established new references for precision and reasoning.
Isa Fulford, member of the OPENAI research team, shared in Livestream Youtube that the model reaches “a new summit of 26.6% precision” on “The last examination of humanity“A relatively new AI reference designed to be the most difficult for any model of AI (or human, elsewhere) to complete, covering 3,000 questions on 100 different subjects, such as the translation of old inscriptions on Archaeological discoveries.

In addition, its ability to navigate the web, dynamically reasons and cite sources precisely distinguishes it from the tools of previous AI.
“The model was formed using the end -to -end strengthening learning on the hard navigation and reasoning tasks,” said Fulford. “He learned to plan and execute trajectories in several stages, to react to information in real time and to go back if necessary.”
A remarkable characteristic of in -depth research is its ability to manage tasks that otherwise take human hours, or even days.
During the announcement, Chen explained that “deep research generates results that resemble a complete and entirely cited research document – something that an analyst or an expert in the field could produce.”
Applications and use cases
The use cases for in -depth research are as diverse as possible.
The official OPENAI account on X said that it was “built for people who make intensive knowledge in fields such as finance, science, politicians and engineering and need in -depth and reliable research”.
It also seems precious for consumers looking for personalized recommendations or to carry out detailed research on products, according to examples shared by OPENAI on his official article in depth research announcement blogwhich includes a detailed search assessment of the best snowboard so that someone can buy.
Altman summed up the versatility of the tool, writing: “Try it on your most difficult work task that can be resolved simply using the Internet and see what is happening.”
A personal medical success of in -depth research
Felipe Millon, head of the government of the government of Openai, shared a deeply personal story of the impact of in -depth research. Writing in a series of messages on xHe described his wife’s battle against bilateral breast cancer and how AI’s tool has become an unexpected ally.
“At the end of October, my wife received a diagnosis of bilateral breast cancer. Our world has turned overnight, “wrote Millon.
After a double mastectomy and chemotherapy, the couple was faced with a critical decision: whether or not whether or not to continue radiotherapy. The situation was heavy with uncertainty, because even their specialists provided mixed recommendations. “For his specific case, it’s completely in a gray area,” said Millon. “We felt stuck.”
Having an overview of access to in -depth research, Millon has decided to download the report on his wife’s surgical pathology and to ask if the radiation would be beneficial. “What happened then was breathtaking,” he wrote. “It didn’t just confirm what our oncologists have mentioned – he went further. He cited studies that I had never heard of and adapted when we added details like his age and genetic factors. »»
The specific prompt he used was:
“Read the report on surgical pathology (attached) containing information on bilateral breast cancer. Then, research on if the radiation would be indicated for this patient after 6 cycles of TCHP chemotherapy, based on the type of breast cancer. I want to understand the advantages and disadvantages of radiation for this patient, to what extent it would be likely to reduce the risk of recurrence and if the benefits prevail over potential long -term risks. »»
Millon and his wife checked each study cited by the model, finding them exact and very relevant. “We soon see another specialist, but we already feel more confident about our decision,” he wrote. “It gave us peace of mind when we needed it most.”
Availability and what is the next one?
Deep Research is currently available for professional chatgpt users, with plans to develop at more and team levels, followed by business and education markets.
As Chen warned, “it is always possible that it hallucinates, so when you make reports, be sure to check the sources yourself.”
The capacity of the model to think independently for long periods also makes it at high intensity of resources, and Openai is currently working to optimize its performance for wider accessibility.
OPENAI also alluded to future integrations with personalized data sets, which would allow organizations to take advantage of the owner research tool.
For Millon, the impact of in -depth research is already clear. “We often speak internally in Openai moments when you” feel the act “, and it was one of them,” he wrote. “This thing will change the world.”