
OpenAI has introduced Deep Research, an AI agent that uses reasoning to synthesize large amounts of information across the web and complete multi-step research tasks for you. It's available to Pro users today with Plus and Team users getting access next.
Key Points
- Deep Research independently browses, interprets, and consolidates information from across the web.
- Uses a version of OpenAI’s o3 model optimized for web browsing and data analysis.
- Accomplishes in minutes what might take a human hours or days.
- Scored 26.6% on Humanity’s Last Exam, more than double its predecessor.
Deep Research operates differently from traditional AI chatbots. Rather than providing immediate responses, it takes between 5 to 30 minutes to complete a research task—methodically browsing through hundreds of online sources, analyzing data, and synthesizing findings into comprehensive reports with clear citations.

The agent's performance metrics are impressive. On "Humanity's Last Exam," a benchmark for testing AI systems across various subjects, Deep Research achieved a 26.6% accuracy rate, substantially higher than previous models and almost 3x DeepSeek.
While the model is conducting research, it displays its thought process and browsing activity in a sidebar, offering transparency into how it reaches its conclusions. The tool also allows users to upload files and images for additional context, enabling more nuanced and targeted research outcomes.

Knowledge workers in fields like finance, science, and law can leverage Deep Research to streamline research tasks that typically require hours. The tool shows particular promise in uncovering non-obvious connections and niche information that might be missed in conventional research approaches.
The new capability will first be available to Pro users with a limit of 100 queries per month, the company plans to expand access to Plus and Team users, followed by Enterprise customers. This gradual rollout excludes users in the UK, Switzerland, and the European Economic Area at launch.
The tool's limitations are worth noting. While it demonstrates improved accuracy compared to existing models, it can still produce hallucinations and struggles with distinguishing authoritative information from less reliable sources. Additionally, its current version requires significant computational resources, explaining the restricted access at launch.
Looking ahead, OpenAI plans to enhance Deep Research's capabilities by enabling connections to specialized data sources, including subscription-based and internal resources. This expansion could make the tool even more valuable for organizations requiring comprehensive research across proprietary databases and public information.