On February 3, 2025, OpenAI announced the launch of a new AI agent called “deep research,” designed to enhance in-depth research capabilities using its ChatGPT platform. This tool aims to assist users in fields such as finance, science, and policy by providing thorough and reliable research outputs.
- OpenAI introduces deep research AI agent
- Targeted for intensive knowledge work
- Available to ChatGPT Pro users first
- Outputs include citations and summaries
- Uses optimized o3 reasoning AI model
- Similar feature announced by Google recently
The deep research feature is intended for users who require detailed analysis rather than quick answers. It allows users to input queries and attach files, with responses taking between 5 to 30 minutes. Currently, the outputs are text-only, but OpenAI plans to include images and data visualizations in the future. The feature is being rolled out to ChatGPT Pro users first, with additional support for Plus and Team users expected soon.
OpenAI emphasizes the importance of accuracy in deep research outputs, which will be fully documented with citations. The deep research tool utilizes a specialized version of OpenAI’s o3 reasoning model, which has been trained to handle complex tasks involving web browsing and data analysis. Despite achieving an accuracy of 26.6% in a rigorous evaluation, the model still faces challenges, including potential errors in judgment and formatting.
OpenAI’s testing revealed that the deep research model outperformed other AI models in accuracy, but it also acknowledged its limitations. Users are encouraged to critically evaluate the information provided, as the tool may struggle to differentiate between credible sources and misinformation. The company aims to enhance the tool’s capabilities further, including integrating specialized data sources and improving the overall reliability of the outputs.
In summary, OpenAI’s deep research tool represents a significant advancement for users requiring detailed research capabilities. While the tool shows promise in delivering well-cited and thorough information, users should remain vigilant about verifying the accuracy of the outputs.
