OpenAI has launched Deep Research, a groundbreaking tool designed to revolutionise the way complex tasks are carried out on the internet. With the ability to independently search, analyse, and synthesise vast amounts of online information in just a fraction of the time it would take a human, this new capability marks a significant leap towards artificial general intelligence (AGI).
Deep Research leverages a version of OpenAI’s O3 model, which has been specially optimised for web browsing and data analysis. The tool is able to sift through massive volumes of text, images, and PDFs, applying reasoning to interpret and consolidate data from hundreds of online sources. It is capable of pivoting as needed in response to the information it uncovers, ensuring a thorough and comprehensive analysis. OpenAI believes that this ability to synthesise existing knowledge is key to creating new knowledge, a crucial step in their vision of AGI that can produce novel scientific research.
The tool has been specifically designed to assist those engaged in intensive knowledge work, including professionals in sectors like science, finance, engineering, and policy, who require precise and thorough research. It is also aimed at helping shoppers seeking hyper-personalised recommendations for high-consideration purchases such as cars, furniture, and appliances.
Deep Research is trained on real-world tasks using end-to-end reinforcement learning, a methodology similar to the one used for OpenAI’s initial reasoning model, O1. However, while O1 excelled at technical tasks like coding and mathematics, the real-world challenges faced by professionals often require in-depth research across diverse sources. This is where Deep Research comes in. It can handle complex, multi-faceted inquiries by independently gathering and contextualising information. It also boasts a unique ability to backtrack and adjust its approach based on new data, ensuring that the research remains relevant and up to date.
For users, Deep Research can be accessed through the ChatGPT platform, where it functions as an agent that can perform comprehensive investigations across a wide range of topics. Once a user inputs a query, the tool embarks on a multi-step journey to find, analyse, and synthesise information. Each task typically takes between five and 30 minutes to complete, and once finished, the results are presented in the form of a detailed report. This report includes clear citations and an outline of the steps the model took during the research, making it easy for users to verify the sources and reasoning behind each claim.
The tool’s impact has already been showcased in a demo video, where it was used by researchers at Bain & Company to investigate the global semiconductor shortage. The team at Bain found that Deep Research quickly summarised the root causes of the crisis, identified the industries most affected, and even provided context for future developments. According to Reem Anchassi, Director of Research & Data Services at Bain, the tool has increased the capacity of researchers, improved efficiency, and freed up valuable time for other complex tasks.
OpenAI’s CEO, Sam Altman, has emphasised that this tool is just the beginning of a larger vision. The company is planning to expand the capabilities of Deep Research to enable more advanced agentic experiences, where ChatGPT can carry out not just research, but also real-world actions. As OpenAI continues to refine its model, the potential applications of Deep Research are vast, ranging from helping researchers tackle difficult questions to supporting businesses with insights that are critical for making informed decisions.
Currently available via ChatGPT’s web platform, Deep Research will be rolled out to mobile and desktop apps in the coming months. OpenAI’s long-term vision includes further enhancing this tool, adding features like embedded images and data visualisations to improve clarity and context. As the tool evolves, it will play a crucial role in advancing the field of AGI, enabling machines to not only perform tasks more efficiently but also to contribute to the generation of new, valuable knowledge.
