OpenAI has introduced a new Deep Research feature. The AI agent will now be able to perform complex research by analyzing data from the Internet. The process of one such study will take from 5 to 30 minutes.
The feature is powered by the o3 model, which is capable of reasoning, and is already available to ChatGPT Pro users in the US with a limit of 100 studies per month. In the future, OpenAI plans to expand its availability to ChatGPT Plus and Teams in other countries, as well as increase the number of queries allowed per month.
today we launch deep research, our next agent.
this is like a superpower; experts on demand!
it can go use the internet, do complex research and reasoning, and give you back a report.
it is really good, and can do tasks that would take hours/days and cost hundreds of dollars.
OpenAI reports that the new feature is intended for "people who do intensive knowledge work in areas like finance, science, policy, and engineering and need thorough, precise, and reliable research." Deep Research can also be useful for those who make "purchases that typically require careful research, like cars, appliances, and furniture."
For now, the results of ChatGPT's in-depth research are text-only. But OpenAI says it plans to add embedded images, data visualization, and other "analytical" results soon. Also on the roadmap is the ability to connect "more specialized data sources," including "subscription-based" and internal resources.
In the final response, ChatGPT provides a full list of references. This is especially important because artificial intelligence can still make mistakes in interpreting data or inventing facts. Perhaps for this reason, OpenAI stated that each result of ChatGPT's in-depth research will be "fully documented, with clear citations and a summary of its thinking, making it easy to reference and verify the information."
OpenAI reported that it tested the new feature using Humanity's Last Exam, a test with more than 3,000 expert questions in various academic fields. The o3 model, which underlies deep learning, achieved an accuracy rate of 26.6%. Although this figure may seem low, it is significantly higher than the results of other models: Gemini Thinking scored only 6.2%, Grok-2 - 3.8%, and GPT-4o - 3.3%.
