OpenAI introduces a new ChatGPT agent for ‘deep research’

OpenAI has launched a new Deep Research mode in ChatGPT, enabling users to conduct multi-step research on the web for complex tasks. This marks OpenAI’s second AI agent, following last month’s release of Operator AI for browser-based tasks.

Deep Research runs on OpenAI’s latest o3 reasoning model, optimized for web browsing and data analysis. It searches, interprets, and analyzes vast amounts of text, images, and PDFs to generate detailed reports, approaching the level of a research analyst, according to OpenAI.

Unlike standard ChatGPT searches, Deep Research queries take 5 to 30 minutes to process. Users receive a notification once their research is complete. Currently, the feature supports text-based results, but OpenAI plans to introduce embedded images, data visualizations, and other analytical outputs in the coming weeks.

Deep Research: Potential and Limitations

Like other OpenAI tools, Deep Research still struggles with hallucination, sometimes generating inaccurate information. OpenAI acknowledges this issue in its blog post. The company also admits that Deep Research has difficulty distinguishing authoritative facts from rumors and often fails to convey uncertainty accurately.

This raises concerns about the reliability of its detailed reports. If the AI produces misleading information, users must verify sources manually, which defeats the purpose of an AI research tool. While OpenAI provides citations, reviewing them adds extra effort.

However, OpenAI plans to enhance this new research tool by integrating specialized data sources, including subscription-based and internal databases. This upgrade aims to improve accuracy and personalize results, making the tool more reliable for in-depth research.