OpenAI has introduced a new feature for ChatGPT called deep research, which enables the AI to conduct autonomous, multi-step research to find and analyze data.
Unlike standard text generation, this tool showcases its research process in a sidebar, complete with citations and a summary of its steps.
Users can input queries using text, images, PDFs, or spreadsheets to provide context, and the AI will take between 5 and 30 minutes to deliver a comprehensive response. In the future, OpenAI plans to enhance the feature by integrating images and charts directly into responses.
Despite its capabilities, OpenAI acknowledges that deep research has limitations. The tool may sometimes generate inaccurate information, misinterpret authoritative sources, or struggle to determine confidence levels in responses.
However, OpenAI claims that deep research functions at the level of a research analyst, making AI-generated insights more valuable and practical.
The tool’s launch follows the introduction of Operator, OpenAI’s web-browsing assistant, and is similar to Google’s Project Mariner, which is still in development. Unlike Google’s tool, deep research is already available, starting with a version optimized for Pro users.
OpenAI is initially offering up to 100 queries per month for Pro users at a $200 monthly fee, with limited access for Plus, Team, and Enterprise users.
The company notes that deep research is computationally intensive, with longer queries requiring more processing power. Over time, OpenAI plans to increase rate limits and introduce a faster, more efficient version.
The AI model behind deep research recently set a new benchmark for accuracy on “Humanity’s Last Exam,” which evaluates AI responses to expert-level questions.
The model achieved 26.6% accuracy with browsing and Python tools enabled, significantly outperforming GPT-4o’s 3.3% and OpenAI’s o3-mini model at 13%. As OpenAI continues refining deep research, it aims to make AI a more powerful tool for in-depth data analysis and decision-making.
