ChatGPT's New Deep Research Agent: AI-Powered Analysis & Insights
OpenAI has unveiled an advanced AI agent for ChatGPT called deep research, designed to handle complex inquiries with a multi-step approach. Unlike traditional AI responses, this tool independently gathers data, adjusts based on real-time information, and presents its findings with citations and summaries.
How Deep Research Works
This feature, exclusive to ChatGPT Pro subscribers, enables users to submit queries using text, images, PDFs, and spreadsheets. The AI then takes 5 to 30 minutes to generate a well-researched response within the chat window. Future updates aim to integrate embedded images and charts for a richer research experience.
Instead of merely providing answers, deep research displays a sidebar summary detailing its research process, helping users understand how conclusions were reached. This transparency allows for a better evaluation of the information provided.
Strengths and Limitations
OpenAI claims that deep research operates at the level of a professional research analyst, offering structured and in-depth insights. However, like all AI tools, it has limitations. The model can sometimes:
- Hallucinate or generate inaccurate information
- Struggle with differentiating between authoritative sources and rumours
- Misjudge the certainty of its responses
Despite these challenges, OpenAI is continuously refining its features to improve reliability and accuracy.
Competitive Edge & Future Enhancements
Deep research follows OpenAI's recent launch of Operator, a tool that performs tasks via a web browser. It also mirrors Google's Project Mariner, an AI research prototype revealed in December but not yet publicly available. However, OpenAI has made deep research immediately accessible for Pro users.
The tool allows up to 100 monthly queries for users subscribed to ChatGPT Pro at $200. OpenAI has also hinted at limited access for Plus, Team, and Enterprise users in the near future. Since deep research is computationally intensive, OpenAI plans to release a faster and more cost-effective version for all paid users.
Performance & Accuracy
A press release from OpenAI highlights that deep research achieved a record-breaking score on "Humanity's Last Exam", a benchmark designed to test AI accuracy on expert-level questions. The model, with browsing and Python tools enabled, scored 26.6%, surpassing the previous highest scorer, OpenAI's o3-mini (high) model, which reached 13%.
With this breakthrough, OpenAI continues to push AI capabilities, making research more efficient, accessible, and insightful for users worldwide.