- New AI tool can generate comprehensive reports by analyzing hundreds of online sources.
- Powered by OpenAI’s cutting-edge o3 model, it completes tasks in minutes that would take humans hours.
- Aimed at professionals in finance, science, engineering, and other fields.
- Available initially to US Pro tier users with usage limits.
OpenAI has announced “Deep Research,” a new artificial intelligence capability designed to create comprehensive reports by analyzing vast amounts of online data. The company claims this tool can match the output of a human research analyst, completing in minutes what would take hours for a person. This announcement comes as OpenAI faces growing competition from rivals, particularly China’s DeepSeek, prompting the company to accelerate its product development cycle.
Deep Research leverages a version of OpenAI’s latest and most powerful model, o3, to sift through text, images, and PDFs from hundreds of sources. It then synthesizes this information into a structured report, complete with citations for each claim. OpenAI demonstrated the tool’s capabilities by showcasing its analysis of the translation app market.
While OpenAI positions Deep Research as a significant step towards artificial general intelligence (AGI), the tool is initially targeted at professionals in fields like finance, science, and engineering. It can also be used for research on consumer purchases like cars or furniture. The tool will be accessible as a button within ChatGPT.
The release follows OpenAI’s recent unveiling of other AI agents, including Operator, which can perform tasks like booking restaurant reservations. The company also recently released o3-mini, a free, slimmed-down version of the o3 model. The full o3 model’s capabilities have been highlighted by AI safety experts, who have noted its impressive reasoning abilities and potential implications for AI risks.
Deep Research will be available to US users of OpenAI’s Pro tier, which costs $200 per month. Due to the computational intensity of the tool, usage will be capped at 100 queries per month. It is not currently available in the UK or Europe.
Experts like Andrew Rogoyski from the University of Surrey’s Institute for People-Centred AI have cautioned about the potential for users to rely too heavily on the output of such tools without proper verification. He noted the challenge of confirming the accuracy of AI-generated analyses, which could require significant time and effort from human experts.
