OpenAI has unveiled “deep research,” an advanced agentic feature within ChatGPT designed to autonomously conduct comprehensive, multi-step research tasks on the internet. This capability enables users to input a prompt, upon which ChatGPT will independently search for, analyse, and synthesise information from hundreds of online sources to produce a detailed report comparable to that of a professional research analyst. The process, which traditionally might take a human several hours, is accomplished by deep research in a matter of tens of minutes.
Deep research is fine-tuned on OpenAI’s forthcoming o3 reasoning model, optimised for web browsing and data analysis. This enhancement allows the agent to autonomously navigate and interpret diverse online content, including text, images, and PDFs, thereby generating thorough, well-documented reports with clear citations.
https://openai.com/index/introducing-deep-research/?video=1052827364
images courtesy of OpenAI
Users can activate deep research within the ChatGPT interface by selecting it in the message composer. Research tasks may take between 5 to 30 minutes to complete, depending on complexity, and users receive notifications upon completion. The generated reports are set to include embedded images, data visualisations, and additional analytical outputs in future updates, enhancing the depth and clarity of the information presented.
In terms of performance, deep research has achieved a 26.6% accuracy rate on the “Humanity’s Last Exam” (HLE) benchmark, surpassing previous AI models in expert-level knowledge tasks. This indicates a significant advancement in the model’s ability to handle complex research queries effectively.
Initially, deep research is available to ChatGPT Pro users, with plans to extend access to Plus and Team subscribers in the near future. This phased rollout ensures that a broader user base will soon benefit from this advanced research tool.
Why is this important?
Despite its capabilities, OpenAI acknowledges certain limitations of deep research. The agent may occasionally generate factual inaccuracies or incorrect inferences, struggle to distinguish authoritative information from rumours, and may not always accurately convey uncertainty. Users are advised to critically assess the outputs and verify information when necessary.
The early introduction of deep research represents a significant step forward in AI-driven research assistance, offering users a powerful tool to efficiently gather and analyse information across the web. As OpenAI continues to refine this feature, it has great potential to become an invaluable resource for professionals and researchers seeking comprehensive insights with enhanced efficiency.