Want to immerse yourself in a topic but don’t have time to read reports and papers? OpenAI is launching Deep Research, a new AI tool that explores for you.
OpenAI announced in a blog Deep Research, a new AI tool specializing in deep research. According to OpenAI, the tool is an “AI agent” that autonomously searches for relevant resources on a topic you want to learn more about. Consequently, the responses are also much more comprehensive than when you ask ChatGPT a question.
You ask the Deep Research assistant a question in text, which you can optionally accompany with PDF files or spreadsheets to provide more context. Then the AI tool autonomously sets out to find the requested information. Deep Research plans and executes multiple paths simultaneously and is able to “go back and respond to real-time information” when needed, OpenAI says in the blog.
The underlying model is the o3-mini which is generally available as of today. This model is capable of performing more complex tasks and reasoning. Deep Research will therefore provide you with a comprehensive answer that includes not only information but also an overview of the research process. The answer may take five to 30 minutes to arrive.
Research Assistant
The Deep research tool specializes in Web browsing and complex data analysis. The model uses reasoning ability to search, interpret and analyze vast amounts of text, images and other data on the Internet, and then convert what it learns from that analysis into human language. In the future, OpenAI plans to add the ability to convert information to visual material to the model.
In the demo below, OpenAI shows how Deep Research does a comprehensive analysis of the retail sector over a period of several years. To do this, the model has to look up and interpret a lot of different sources.
OpenAI describes Deep Research as a handy research assistant. The model can sift through large amounts of source material much faster than human researchers. Yet the model is not free of errors, warns OpenAI. In the Humanity’s Last Exam-benchmark it achieves a score of 26 percent, where ChatGPT does not get beyond three percent.
Deep Research will be offered exclusively to users who subscribe to the most expensive Pro subscription ($200 per month), at a maximum of 100 requests per month. Plus subscribers will have a “limited version” available. Free users should expect nothing: the model simply requires too much computing power to just give it away.