OpenAI he move again after the unexpected success of Deepseeklow -cost China. After launching the O3-mini model for all chatgpt users, now it has touched the turn of a new agent. Deep Research or Deep Researchcapable of ‘carrying out several steps online research for complex tasks’ and ‘He achieves in tens of minutes what a human being would take many hours’.
As the company explains in his blog, ‘Deep Research is the next OpenAi agent who can do the job for you independently: if you indicate it, ChatGPT will find, analyze and synthesize hundreds of online sources to create a full report at the level of a research analyst‘.
Openai recommends it for work in areas such as Finance, Science, Politics and Engineering that ‘need thorough, precise and reliable investigation.’ It is also useful to obtain ‘hyperpersonalized recommendations on purchases that normally require careful investigation, as cars, appliances and furniture. Each result is completely documented, with clear quotes and a summary of your thinking, which facilitates the reference and verification of the information ‘. Deep Research is also the first example of the next OpenAi language model, O3, here optimized for web navigation and data analysis
Deep Research shows a summary of the process that follows in a sidebar. Users can ask questions using additional texts, images and files such as PDF or spreadsheets To add context, and then take ‘Between 5 and 30 minutes’ Develop an answer that in the future will also include images and graphics.
Among the limitations OpenAi points out for this capacity, remember that Deep Research ‘Sometimes you can hallucinate’ and invent facts, as well as having difficulty differentiate between authorized information and rumors.
https://www.youtube.com/watch?v=ykcdvn3_wiw
This function follows the launch of Operatora tool that can use a web browser to complete tasks by the user and that Openai showed in December under the name of Project Mariner.
OpenAi offers up 100 queries per month for those who pay the monthly fee of 200 dollars of the Pro subscription planand promises a ‘Limited access’ For users of Plus, Team And, eventually, Enterprisequalifying the ability to ‘Very intensive computation’which requires more inference computing the longer it takes to investigate something. He also says that All payment users will obtain higher use limits in the future, when a faster and more profitable version is available.
The company ensures that the model promoted Deep Research obtained a new precision maximum in a benchmark of the so -called ‘The last examination of humanity’which raises expert level questions. Deep Research reached a precision from 26.6 percent with the navigation and python tools enabled, well above the 3.3 percent of GPT-4O. Your model follows O3-mini-High, with 13 percent.