OpenAI has introduced a new ChatGPT agent called “deep research,” designed to provide detailed, analyst-level reports. The tool is part of OpenAI’s Pro plan and can be used for personal research tasks. OpenAI stated that its deep research agent achieved a significantly higher accuracy on Humanity’s Last Exam than DeepSeek-R-1 and GPT-4o. The OpenAI o3 model powers the agent, using its self-fact-checking reasoning capabilities to enhance web browsing and data analysis. The startup cautioned that deep research may occasionally produce inaccurate facts, misinterpret information, or rely on unreliable sources.