OpenAI Launches ‘Deep Research’ for Better Insights
OpenAI has announced a new AI feature called Deep Research, designed to help users conduct in-depth, complex investigations using ChatGPT. Unlike the chatbot’s usual quick summaries, Deep Research is tailored for people who require thorough, precise, and reliable information across fields like finance, science, policy, and engineering. It also aims to assist consumers with major purchase decisions that require careful comparisons, such as cars, appliances, and furniture.
A New Level of AI-Assisted Research
Deep Research is meant for instances where a simple answer isn’t enough—where users need to gather, analyze, and verify information across multiple sources. OpenAI is rolling out this capability to ChatGPT Pro users today, with Plus and Team users next in line, followed by Enterprise customers. Initially, users will be limited to 100 queries per month, though OpenAI expects to increase this limit significantly soon. The launch is geo-targeted, with no set timeline for availability in the U.K., Switzerland, and the European Economic Area.
How Deep Research Works
To use Deep Research, users can select the feature in the ChatGPT composer, input a query, and even attach files or spreadsheets. Unlike instant AI responses, Deep Research takes its time—5 to 30 minutes—to gather and analyze data before delivering a detailed report. Users receive a notification once the process is complete.
For now, responses are text-only, but OpenAI plans to add embedded images, data visualizations, and other analytic outputs soon. Future updates will also integrate specialized data sources, including subscription-based and internal resources.
Accuracy and Limitations
One of the biggest concerns with AI-powered research is accuracy. ChatGPT, like all AI models, is prone to hallucinations and misinformation. To address this, OpenAI ensures that every Deep Research response is fully documented, with clear citations and a summary of the reasoning behind the results.
To boost reliability, OpenAI is using a specialized version of its o3 AI model, optimized for web browsing and data analysis. This model leverages reinforcement learning, improving itself by completing real-world research tasks using browsing tools and Python-based data analysis.
According to OpenAI’s internal testing, the Deep Research model achieved 26.6% accuracy on Humanity’s Last Exam, a challenging benchmark designed to test AI reasoning abilities. While this may not seem high, it significantly outperforms competitors like Gemini Thinking (6.2%), Grok-2 (3.8%), and GPT-4o (3.3%).
However, Deep Research isn’t perfect. OpenAI admits that it can misinterpret data, struggle to differentiate authoritative sources from unreliable ones, and occasionally fail to indicate uncertainty in its findings. Formatting errors in reports and citations also remain an issue.
The Future of AI-Driven Research
For researchers, students, and professionals, a well-cited, in-depth AI-generated report is a step above simple chatbot answers. The key question remains: Will users critically analyze AI-generated research, or will they simply copy-paste the outputs without further verification?
Interestingly, OpenAI’s Deep Research launch follows a similar announcement from Google, which introduced an AI feature with the same name just two months ago. As AI-driven research tools continue to evolve, competition will likely intensify, shaping the future of knowledge discovery.
