Deep Research: Inside OpenAI's New Analysis Tool

OpenAI has launched a new tool that crafts in-depth reports based on multi-step, thorough and precise research on the internet, which it says can equal the output of a research analyst.

The company says the agent, ‘deep research’, can accomplish in tens of minutes what a human can do in many hours. 

Deep research is a new agentic capability that works independently once given a prompt to find, analyse and synthesize hundreds of online resources and create comprehensive reports for people carrying out intensive knowledge work in sectors like science, engineering, finance and more. 

It is powered by a version of the company’s OpenI o3 model optimised for data analysis and web browsing. OpenAI says deep research leverages reasoning to collate and analyse ‘massive amounts of text, images and PDFs’ online, and has the flexibility to pivot in reaction to the information it finds. 

In a launch statement OpenAI said: 'The ability to synthesize knowledge is a prerequisite for creating new knowledge. For this reason, deep research marks a significant step toward our broader goal of developing AGI (artificial general intelligence), which we have long envisioned as capable of producing novel scientific research'.

What does deep research do? 

The short answer: a lot. In practice, the tool independently finds, applies reason to, and consolidates information from the internet – OpenAI trained it on real-world tasks in the same way it did its first reasoning model, OpenAI o1. 

However, while that tool was a dab hand at work in areas like maths, coding and technical output, the complex real-world challenges that industries and businesses face often require more in-depth and extensive information and context gathering from multiple and diverse online sources. 

OpenAI says that deep research expands on its existing capabilities to take on the kinds of problems that many of us face in our day-to-day work and life. That makes it ideal for professionals in information-heavy and complex industries, and for expert-level tasks in areas like science, linguistics and healthcare. 

It’s also equally useful for those professionals in their downtime – OpenAI reports deep research can help ‘discerning shoppers looking for hyper-personalised recommendations on purchases that typically require careful research, like cars, appliances, and furniture’.

The company says the tool will take between five and 30 minutes to complete each task and will cite the source for every claim it makes. It appears as a button in ChatGPT, and users can attach files and spreadsheets to give greater context to prompts and questions. 

OpenAI trained deep research using end-to-end reinforcement learning. This gives the tool the capability to plan and execute a multi-step trajectory to source information and data, including backtracking and reacting in real-time. 

The final output is presented as a report within the chat, with OpenAI setting out plans to add embedded images and data in the future.

Deep research in practice

OpenAI released a demo video, showing how deep research is being used by researchers and analysts at global consulting firm Bain & Company. 

It illustrates how the organisation is harnessing the tool for insight into the global semi-conductor ship shortage, including understanding underlying issues, which industries are most affected and the future outlook. In the clip Reem Anchassi, Director, Research & Data Services at Bain & Co says deep research is “really trying to understand what the requester needs, even down to what geographies and industries you are looking at”, adding that it “summed up rather quickly what the underlying reasons were and provided some context.”

Anchassi said the tool increases the capacity of analysts and researchers, improves efficiency and frees time for other complex tasks across the organisation. 

Deep research is currently available in ChatGPT web, and will be rolled out to mobile and desktop apps over the next month. 

Plotting future development, OpenAI said: “we envision agentic experiences coming together in ChatGPT for asynchronous, real-world research and execution. The combination of deep research, which can perform asynchronous online investigation, and Operator, which can take real-world action, will enable ChatGPT to carry out increasingly sophisticated tasks.”

Explore the latest edition of AI Magazine and be part of the conversation at our global conference series, Tech & AI LIVE. 

Discover all our upcoming events and secure your tickets today.

AI Magazine is a BizClik brand