OpenAI launched Deep Research, a new artificial intelligence feature that applies ChatGPT. The system allows users to research different topics through extensive investigations, offering different functionality from typical chatbot functionality by analyzing multiple sources while delivering detailed results.
OpenAI states that this tool provides significant advantages to professionals working in finance, science, policy and engineering industry sectors. The system helps buyers determine suitable choices when buying items such as cars, furniture and household appliances.
Users who subscribe to ChatGPT Pro can now access the new functionality, which allows 100 monthly queries. OpenAI planned to add deep research functionality first for its Plus and Team users and then for Enterprise customers. The geographical release targets specific regions, while the U.K., Switzerland, and the European Economic Area wait for undisclosed dates to receive the update.
Deep Research functionality allows users to access it through the ChatGPT composer by choosing the option and then entering their search query. Users can upload files or spreadsheets to Deep Research to analyze complex data sets. Research tasks require five to thirty minutes, while the system sends notifications when analyses are completed. The launch will introduce Deep Research to the web platform first, but upcoming plans include mobile and desktop support during the current month.
The present version of Deep Research generates text results. OpenAI intends to add visual elements and analytical functions, including images, as part of its upcoming development. Future upgrades of the tool will establish connections between specialized data sources that span internal resources and subscription-based platforms.
The o3 "reasoning" AI model received a specific version from OpenAI to boost accuracy. The model optimization meets web browsing and data analysis requirements by enabling it to search, interpret, and analyze vast amounts of text alongside images and PDFs. The system can examine user-submitted files and produce data charts with Python assistance, followed by graphical elements from original content and external sources.
OpenAI tested deep research reliability with Humanity’s Last Exam benchmark, which contains expert-level questions from different academic disciplines and totals over 3,000. With its performance rate of 26.6%, the o3 model outpaced competing models Gemini Thinking (6.2%), Grok-2 (3.8%), and GPT-4o (3.3%) from OpenAI.
OpenAI acknowledges that the tool maintains certain constraints despite the recent progress made. Authentic information and rumors present challenges for Deep Research to distinguish between them. The platform struggles with inconsistent behavior by failing to signal doubt and producing incorrect document formatting in research reports. All OpenAI research outputs contain citations and explications of the reasoning steps to enable users to access information verification easily.