OpenAI has launched Deep Research, a new capability within ChatGPT designed to conduct multi-step research on the internet for complex tasks. It utilizes an optimized version of the upcoming OpenAI o3 model, which specializes in web browsing and data analysis. Deep Research can autonomously search, analyze, and synthesize information from a vast range of sources, delivering detailed reports with citations. This tool aims to enhance productivity by handling time-intensive investigations efficiently. The research process, which can take between 5 and 30 minutes, provides users with a structured breakdown of sources and reasoning steps.
Deep Research builds upon OpenAI o1, the company’s first reasoning model, which has demonstrated strong performance in coding, mathematics, and technical problem-solving. Unlike o1, which employs structured reasoning, Deep Research extends these capabilities by independently gathering and synthesizing information. The model has achieved high accuracy on evaluations including Humanity’s Last Exam and GAIA, reflecting improvements in real-world problem-solving. While the tool introduces significant advancements, it is still being refined to address limitations including occasional factual inaccuracies and confidence calibration issues. Initially available to Pro users, Deep Research will eventually expand to Plus and Team users, with broader access planned in the coming months.
