OpenAI has introduced a new advanced feature for ChatGPT called Deep Research, designed to operate autonomously by planning and executing multi-step processes to gather relevant data.
Unlike traditional AI-generated responses, this tool actively backtracks and adapts in real time to refine its findings.
Instead of just generating text, Deep Research presents a transparent workflow in a dedicated sidebar. This includes citations, summaries, and an overview of the research process, providing users with clear insights into how the information was obtained.
How Deep Research Works
Users can input queries using text, images, and additional files such as PDFs or spreadsheets to provide context. The AI then takes between 5 to 30 minutes to develop a detailed response within the chat window. Future updates promise enhancements like embedded images and charts for better data visualization.
However, OpenAI acknowledges some limitations with Deep Research, including occasional hallucinations (fabricated information), difficulty distinguishing authoritative sources from rumors, and uncertainty in rating the reliability of responses.
AI Agents: The Future of Research and Assistance
OpenAI has positioned Deep Research as an AI-powered research analyst, aligning with its broader vision of AI agents that enhance productivity and justify premium subscription costs. In a demonstration, the tool responded to a request about changes in the retail industry over the past three years, providing a structured summary with bullet points and tables.
This launch follows OpenAI’s recent introduction of Operator, a tool that automates web browsing to complete tasks, and mirrors Google’s Project Mariner, a research prototype revealed in December. While Google’s version is not yet available to the public, Deep Research is rolling out today in an optimized version for Pro users.
Availability and Pricing
OpenAI is offering Deep Research with up to 100 queries per month for Pro users at a $200 monthly fee. Limited access is expected for Plus, Team, and Enterprise users. Due to its high computational demands, OpenAI plans to introduce a faster, more cost-effective version with improved rate limits in the future.
According to OpenAI, the model powering Deep Research achieved record-breaking accuracy on the “Humanity’s Last Exam,” an AI benchmark for expert-level questions. With browsing and Python tools enabled, it scored 26.6% accuracy—significantly higher than GPT-4o’s 3.3% and the o3-mini (high) model’s 13%.
With Deep Research, OpenAI is taking another step toward making AI a powerful research assistant, capable of handling complex inquiries with a structured and transparent approach.
