The deep research tool launching today for ChatGPT Pro subscribers is an AI agent that shows its steps
OpenAI has revealed another new agentic feature for ChatGPT called deep research, which operates autonomously to plan and execute a multi-step trajectory to find the necessary data. Instead of generating text, it offers a summary of its process in a sidebar, complete with citations and a reference process.
How does it work?
Users can ask questions using various formats like text, images, PDFs, and spreadsheets for context. It takes anywhere from 5 to 30 minutes to develop a response, with future capabilities to include embedded images and charts. However, OpenAI notes some limitations, such as potential factual errors and struggles with distinguishing authoritative info from rumors.
Advancements in AI tools
Companies like OpenAI are focused on making generative AI tools more useful and valuable. Deep research is positioned as capable of operating at the level of a research analyst, providing detailed and comprehensive responses.
The launch of Operator, a web browser tool for task automation, precedes the introduction of deep research. Google's Project Mariner research prototype is a similar concept, though not yet available to the public. However, deep research is launching with an optimized version for Pro users today.
Subscription and Access
OpenAI offers up to 100 queries per month for Pro users at a $200 monthly fee, with limited access for Plus, Team, and future Enterprise users. The tool is computationally intensive, with inference compute increasing based on the complexity of the research task.
With a new accuracy milestone on the "Humanity’s Last Exam" benchmark, the deep research model outperforms previous iterations like GPT-4o and o3-mini in expert-level question responses.
© 2025 Vox Media, LLC. All Rights Reserved
