OpenAI has released a new artificial intelligence tool "DeepResearch" designed to perform time-consuming online research tasks for users, covering many fields from complex scientific problems to car recommendations. The new tool further adds to the startup’s lineup of AI agents.
DeepResearch can take anywhere from 5 to 30 minutes to complete its job - digging deeper into the web while you're away or working on other tasks. The final output appears in the chat in the form of reports, and soon users will also see embedded images, data visualizations and other analysis output for enhanced clarity. This time has greatly improved its capabilities. For example, on "Man's Last Exam", DeepResearch achieved an impressive 26.6% accuracy - far exceeding its nearest competitor DeepSeekR1, which scored less than 10%. This leap highlights the system's iterative optimization and structured synthesis capabilities, proving that in the rapid development of artificial intelligence, some tasks do take time.
In a blog post on Sunday, OpenAI announced that the service, called "DeepResearch," will be available to select paying customers through its ChatGPT online chatbot. The tool can search text, images, PDF files and user-uploaded files online based on user instructions, and then generate a detailed report. OpenAI likens this capability to that of a research analyst, stressing that it can complete in "tens of minutes" what would normally take a person "hours" to complete.
However, OpenAI also reminds users that "DeepResearch" is still in its early stages and may mistake fictional information for fact, and may have difficulty distinguishing rumors from accurate information. In addition, this research tool is "very computationally intensive" and initial users can only submit 100 queries per month.