OpenAI Introduces ChatGPT Deep Research Agent for In‑Depth, Cited Research

OpenAI announced a new AI agent called ChatGPT Deep Research , aimed at users who need intensive, accurate research across fields such as finance, science, policy, and engineering, as well as consumers making complex purchasing decisions.

The feature is currently offered to ChatGPT Pro users with a limit of 100 queries per month, with upcoming support for Plus, Team, and later Enterprise accounts; Plus rollout is expected within a month.

To use Deep Research, users select the "Deep Research" option in the chat interface, enter a query, and optionally attach files or spreadsheets. The web‑only experience may take 5–30 minutes per query, after which users receive a notification.

At present, the output is plain text, but OpenAI plans to add image embeddings, data visualizations, and other analytical outputs, as well as connections to subscription‑based and internal data sources.

Accuracy is a concern: the agent runs on a specialized version of OpenAI’s o3 reasoning model, trained with reinforcement learning on real‑world browsing and Python tasks. In internal testing on a 3,000‑question expert exam, the o3 model achieved a 26.6% accuracy rate, outperforming Gemini Thinking (6.2%), Grok‑2 (3.8%), and GPT‑4o (3.3%).

OpenAI acknowledges limitations, noting that Deep Research can still hallucinate, misinterpret sources, and fail to convey uncertainty, sometimes producing formatting errors in citations.

The deep‑cited output is intended to be more trustworthy than typical chatbot replies, though its real‑world impact will depend on whether users verify the information rather than copy‑pasting it.