OpenAI has launched a new feature called Deep Research for ChatGPT Pro users. This tool is designed to assist professionals in fields such as finance, science, policy, and engineering by providing precise and reliable information. It can also help everyday users make informed decisions on matters like purchasing a car or home appliances.
Unlike quick answer tools, Deep Research is intended for users who need to analyse multiple sources carefully. Users can select the 'deep research' option in ChatGPT's interface, enter a query, and attach files or spreadsheets if necessary. The AI takes between 5 to 30 minutes to generate a response, notifying users when it is ready.
Currently, the Deep Research feature is available only on the web version of ChatGPT, with plans to integrate it into mobile and desktop apps later this month. At present, responses are text-only, but OpenAI intends to add images, data visualisations, and other analytical outputs soon.
The feature is being rolled out starting today for ChatGPT Pro users, with a limit of 100 queries per month. OpenAI plans to extend access to Plus and Team users next, followed by Enterprise customers. However, the feature is not available in the UK, Switzerland, or the European Economic Area, with no release timeline provided for these regions.
There are concerns about the accuracy of AI-generated research due to the potential for hallucinations or misleading information. To mitigate this, OpenAI states that each Deep Research response will include citations and a summary of its reasoning to help users verify the information.
Deep Research is powered by a specialised version of OpenAI's new o3 model, designed for web browsing and data analysis. This model employs reinforcement learning, improving through trial and error. It is capable of searching, interpreting, and analysing large amounts of text, images, and PDFs from the internet, adapting its approach based on the information encountered, processing user-uploaded files, generating and refining graphs using Python, and embedding visuals from websites in its responses.