OPENAI has revealed another new agentic functionality for chatgpt called Deep Research, which, according to her, can operate independently to “plan and run a trajectory in several steps to find the data it needs, go back and react to information in real time if necessary. “
Instead of simply generating text, it shows a summary of its process in a side bar, with quotes and a summary showing the process used for reference.
Users can ask questions using text, images and additional files such as PDFs or calculation sheets to add context, then it will take “5 to 30 minutes” to develop a response provided in the cat window , with promises that in the future will also be able to include on -board images and graphics. OPENAI also notes the limitations for in -depth research, saying that it can “sometimes hallucinate” and invent facts, have trouble making the difference between information and rumors, and recording how it should be an answer .
Developing means for generative AI tools to be more useful and deserves to be paid is future companies like OpenAi have promised for agents, and he says that in -depth research is capable of operating at the level of an analyst of research. The demonstration video included here begins with a request for information on changes in the retail industry in the past three years, with an answer that includes fleas and tables.
This feature closely follows the launch of Operator of Openai, a tool that can use a web browser to perform tasks for you, and is similar to the research prototype of Mariner Research Project that Google showed in December. Google’s tool is not yet available for the public, but in -depth searches are launching “with an optimized version for professional users today”.
OPENAI offers up to 100 requests per month for those who pay the monthly costs of $ 200 and “limited access” promised for more, the team and, possibly, business users, calling “very intensive” capacity, requiring more calculation inferences longer to seek something. It also indicates that all paid users will obtain higher rate limits in the future when a faster and more profitable version is available.
A press release indicates that the model feeding Deep Research has obtained a new summit for the precision on a reference on AI nicknamed “Last Exam’s Humanity’s Last”, which requests answers to questions at the expert level. The Openai Deep Research model reached 26.6% precision with activated navigation and python tools, well above 3.3% of GPT-4O, and the top scorer, its model O3-mini (high) evaluated only on the text, at 13%.