OpenAI launched a new agent that is capable of performing complex online research in multiple steps, including scientific studies and personalized bike recommendations.
Deep Research is powered by an optimized version of OpenAI’s o3 reasoning models for web browsing and data analytics. It can search through and analyze large amounts of text, images and PDFs in order to create a thorough report.
Openai””https://openai.com/index/introducing-deep-research/””> claims that the tool represents an important step towards its overarching goal to develop artificial general intelligence (AGI), which matches (or exceeds) human performance. It claims that what the tool does in “tens” of minutes would take an average human many hours.
Deep Research can search the internet, analyze the data it finds, and then compile a detailed analysis that cites the sources. It can also draw from files uploaded to the system by users.
OpenAI developed Deep Research using the same “chain of thought” reinforcement-learning methods it used to create its o1 multistep reasoning model. While o1 was designed primarily to focus on mathematics, coding or other STEM-based activities, Deep Research can handle a much broader range of topics. It can also adapt its responses to new data that it finds during its research. This does not mean that Deep Research will be immune to the pitfalls of other AI models. OpenAI claims that the agent can sometimes mislead its users and give them incorrect information. However, this happens at a “significantly” lower rate than ChatGPT. Deep Research is very computationally intensive because it can take up to 30 minutes to answer a single question. The longer it takes to research an inquiry, the more computing power it requires.
Despite this, Deep Research is available to OpenAI Pro subscribers at no additional cost and will soon be rolled out to Plus, Team, Enterprise users.
