OpenAI unveils a new ChatGPT agent for ‘deep research’

OpenAI has introduced a new AI “agent” called deep research, powered by ChatGPT, to assist users in conducting thorough and complex research tasks.

Designed for individuals involved in intensive knowledge work across various fields such as finance, science, policy, and engineering, deep research aims to provide precise and reliable research results. It can also be beneficial for those making significant purchases that require detailed research, like cars, appliances, and furniture.

Unlike quick summaries, ChatGPT deep research is ideal for situations where information from multiple sources needs careful consideration.

OpenAI is rolling out deep research to ChatGPT Pro users initially, with plans to expand to Plus, Team, and Enterprise users in the future. The feature is currently limited to 100 queries per month and will include support for various user levels in the coming weeks.

Users can access ChatGPT deep research by selecting the “deep research” option in the composer, entering a query, and attaching files or spreadsheets. The feature is currently available on the web only, with mobile and desktop app integration planned for later this month. Responses from deep research may take between 5 to 30 minutes, and users will receive a notification once the search is complete.

While current outputs from ChatGPT deep research are text-based, OpenAI is working on incorporating embedded images, data visualizations, and other analytical outputs in the future. The company also aims to connect to more specialized data sources and enhance the accuracy of the research results.

To ensure the precision of deep research, OpenAI is utilizing a specialized version of its o3 “reasoning” AI model, trained through reinforcement learning on real-world tasks involving web browsing and data analysis. This model is optimized for browsing and analyzing vast amounts of text, images, and PDFs from the internet, adapting as needed based on encountered information.

During testing, the o3 model achieved an accuracy of 26.6% in Humanity’s Last Exam, surpassing other AI models in challenging academic questions. However, OpenAI acknowledges that ChatGPT deep research is not flawless and may occasionally produce errors or misinterpret information.

While deep research offers well-cited and detailed outputs, users should remain vigilant of potential inaccuracies and formatting issues. OpenAI’s continuous improvements aim to enhance the reliability and effectiveness of the deep research feature.

Despite similarities with Google’s recent AI feature, ChatGPT deep research stands out for its focus on thorough research and detailed analysis, providing users with a more comprehensive approach to information retrieval.