OpenAI has released a new agent for its flagship artificial intelligence product ChatGPT called “deep research,” which can trawl the internet for information to create a report “at the level of a research analyst.”
OpenAI said in a Feb. 2 blog post that deep research was “built for people who do intensive knowledge work in areas like finance, science, policy, and engineering and need thorough, precise, and reliable research.”
The company added that it’s also useful for research before buying big-ticket items like cars or appliances, and the outputs — which can take anywhere from 5 to 30 minutes — are “fully documented, with clear citations and a summary of its thinking.”
An example of ChatGPT’s “deep research” mode showing its progress and citations. Source: OpenAI
The latest agent follows OpenAI’s Jan. 23 launch of Operator, a ChatGPT agent that can use the internet to complete tasks like ordering groceries and booking holiday tours.
It also comes about a week after the AI space and US tech stocks were rocked by a new AI model from the China-based DeepSeek that reportedly performed as well as ChatGPT but was developed for a fraction of the cost.
Microsoft and OpenAI are reportedly probing if data from ChatGPT’s API was improperly obtained by a group linked to DeepSeek.
OpenAI said its deep research agent scored a new high on the AI evaluation called Humanity’s Last Exam, which has 3,000 expert-level questions on over 100 topics, achieving an accuracy of 26.6% compared to a score of 9.4% for DeepSeek-R-1 and 3.3% for its own GPT-4o model.
Related: DeepSeek privacy concerns raise international alarm bells
The agent is powered by the OpenAI o3 model but “optimized for web browsing and data analysis.” OpenAI o3 is the firm’s latest “reasoning model,” which attempts to essentially fact-check itself to avoid getting facts wrong or generating false information.
OpenAI warned that deep research “can sometimes hallucinate facts in responses or make incorrect inferences” and can “struggle with distinguishing authoritative information from rumors.”
Last month, Google announced it was rolling out a similar feature, also called “Deep Research” for its AI model Gemini in early 2025, while OpenAI said its agent is now available on its $200-a-month Pro plan, limited to 100 queries a month.