Date:
February 03, 2025
OpenAI launches Deep Research, an AI tool that autonomously analyzes and synthesizes web data for in-depth reports, promising transparency and accuracy.
At the onset of DeepSeek’s rise, OpenAI introduced Deep Research, a new AI-driven research assistant designed to conduct in-depth, multi-step investigations across the web. Unlike traditional ChatGPT responses, this tool autonomously searches analyzes, and synthesizes information from multiple sources, producing detailed, citation-backed reports.
Currently available for ChatGPT Pro users, the feature aims to assist professionals in fields like finance, science, policy, and engineering. These are the areas where accuracy and deep analysis are essential. It could also benefit everyday consumers researching high-stakes purchases, such as cars or appliances.
"Instead of simply generating text, it shows a summary of its process in a sidebar, with citations and a summary showing the process used for reference," said Richard Lawler of The Verge.
A More Intelligent and Transparent AI Researcher
Users can activate OpenAI's new tool, Deep Research, by selecting the feature in ChatGPT's message composer and entering a query. The AI then conducts an extensive web search over five to 30 minutes, analyzing text, images, PDFs, and user-uploaded files. A live sidebar provides a breakdown of the research steps, complete with citations.
Deep Research is powered by a specialized version of OpenAI’s o3 reasoning model to enhance accuracy, which has been optimized for web browsing and data analysis. The version of o3 was trained using reinforcement learning, where the AI learns through trial and error, receiving virtual “rewards” for accurate research behavior. As a result, it doesn’t just fetch information - it strategically searches, interprets, and synthesizes data, adjusting its approach when necessary.
To validate its effectiveness, OpenAI tested Deep Research against Humanity's Last Exam, a rigorous benchmark featuring over 3,000 expert-level questions across disciplines like linguistics, rocket science, and ecology. The model scored 26.6% accuracy, far outperforming OpenAI's own GPT-4o (3.3%) and competitors like Gemini Thinking (6.2%) and Grok-2 (3.8%).
"The model powering deep research showcased a human-like approach by effectively seeking out specialized information when necessary," OpenAI wrote in its launch announcement.
A Great AI Researcher But Not Perfect Yet
Despite its advanced capabilities, OpenAI's new tool, Deep Research, is not perfect. OpenAI acknowledges that it can still hallucinate (generate incorrect information), struggle to distinguish authoritative sources from rumors, and occasionally misform citations. These issues, however, are expected to improve over time.
For now, OpenAI's new tool, Deep Research, is available with a cap of 100 queries per month. OpenAI plans to expand access to Plus and Team users next, followed by Enterprise customers. However, the rollout currently excludes users in the U.K., Switzerland, and the European Economic Area, with no confirmed timeline for those regions.
The Race to Build Smarter AI Research Tools
The launch of OpenAI's new tool, Deep Research, signals a growing competition in the AI-driven research space. Google recently announced a similar Deep Research feature, though it has yet to be made public.
With AI shaping how people access and interpret information, the question remains: Will users critically evaluate AI-generated research or simply copy-paste results without scrutiny?
