OpenAI today announced a new ChatGPT agent called deep research, designed to conduct thorough, multi-source research tasks that take between five and 30 minutes. The feature, available immediately to ChatGPT Pro users at a limit of 100 queries per month, uses a specialized version of the o3 reasoning model trained via reinforcement learning on browser and Python tool use.

Deep research is intended for complex knowledge work in finance, science, policy, and engineering, as well as high-stakes purchases like cars and appliances. Users select the mode in the ChatGPT composer, optionally attach files or spreadsheets, and receive a notification when the report is complete. Outputs are currently text-only, with embedded images and visualizations promised soon.

OpenAI claims the agent scored 26.6% on Humanity’s Last Exam, far ahead of Gemini Thinking (6.2%) and GPT-4o (3.3%), but acknowledges limitations including hallucinations, difficulty distinguishing authoritative sources from rumors, and failure to convey uncertainty. The company said each output will include clear citations and a summary of its reasoning. The launch was preceded by Google’s announcement of a similarly named feature in December 2024.

Around the industry, the debate already heating up is whether users will actually verify the citations or simply copy-paste the professional-looking reports, and whether the measure of ‘deep research’ can meaningfully replace the slower, messier human process of reading and synthesizing.