OpenAI has recently launched a groundbreaking feature called “deep research” in ChatGPT, marking a significant leap forward in AI capabilities. This innovative tool is designed to conduct multi-step research on the internet, accomplishing in tens of minutes what would typically take a human many hours. Deep research represents OpenAI’s next step toward developing Artificial General Intelligence (AGI), which aims to produce novel scientific research.
What is Deep Research
Deep research is powered by a version of the upcoming OpenAI o3 model, optimized for web browsing and data analysis. It leverages advanced reasoning to search, interpret, and analyze massive amounts of text, images, and PDFs on the internet. This capability allows it to pivot as needed in reaction to new information, making it highly effective at synthesizing knowledge from diverse online sources.
Why Was Deep Research Developed?
Deep research is primarily built for individuals who engage in intensive knowledge work, such as those in finance, science, policy, and engineering. It is equally useful for consumers seeking personalized recommendations on significant purchases like cars or furniture. Every output is fully documented with clear citations and a summary of its reasoning process, making it easy to reference and verify the information. This feature is particularly effective at uncovering niche, non-intuitive information that would otherwise require browsing numerous websites.
How Does Deep Research Work?
Deep research independently discovers, reasons about, and consolidates insights from across the web. It was trained using real-world tasks that require browser and Python tool use, employing the same reinforcement learning methods as OpenAI’s first reasoning model, o1. While o1 excels in coding and math, deep research extends these capabilities to handle extensive context and information gathering from diverse online sources.
Using Deep Research
To use deep research in ChatGPT, users simply select the “deep research” option in the message composer and enter their query. They can attach files or spreadsheets to add context to their question. Once initiated, a sidebar appears with a summary of the steps taken and sources used. The research process can take anywhere from 5 to 30 minutes, during which time users can step away or work on other tasks. They will receive a notification once the research is complete, and the final output will arrive as a report within the chat.
Performance and Evaluations.
Deep research has shown impressive performance on public evaluations. On “Humanity’s Last Exam,” a test that assesses AI across a broad range of subjects, the model powering deep research achieved a new high of 26.6% accuracy. This surpasses other models like GPT-4o and OpenAI o1. Additionally, on the GAIA benchmark, which evaluates AI on real-world questions, deep research reached a new state-of-the-art performance.
Limitations and Future Developments.
While deep research offers significant new capabilities, it still has limitations. It may sometimes hallucinate facts or make incorrect inferences, though at a lower rate than existing models. It struggles with distinguishing authoritative information from rumors and often fails to convey uncertainty accurately. OpenAI is working to address these issues and plans to expand deep research’s access to more specialized data sources in the future.
Access and Rollout.
Deep research is currently available to Pro users on ChatGPT web, with plans to roll it out to Plus and Team users soon. It will also be integrated into mobile and desktop apps within the month. OpenAI is working on a faster, more cost-effective version that will provide higher rate limits for all paid users.
What’s Next?
Deep research is available today on ChatGPT web and will be rolled out to mobile and desktop apps within the month. Currently, deep research can access the open web and any uploaded files. In the future, users will be able to connect to more specialized data sources—expanding its access to subscription-based or internal resources—to make its output even more robust and personalized.
Looking further ahead, OpenAI envisions agentic experiences coming together in ChatGPT for asynchronous, real-world research and execution. The combination of deep research, which can perform asynchronous online investigation, and Operator, which can take real-world action, will enable ChatGPT to carry out increasingly sophisticated tasks for users.
This article highlights OpenAI’s advancements in AI research capabilities and provides insights into how deep research can enhance user experiences across various domains. Whether you’re a professional seeking thorough research or a consumer looking for personalized recommendations, deep research is poised to revolutionize how we approach complex tasks.
