Google Enhances Gemini Deep Research Agent with New Features

Google has unveiled an upgraded version of its Gemini Deep Research agent, coinciding with the launch of OpenAI’s GPT-5.2 AI model. This enhanced agent, powered by Gemini 3 Pro, is specifically designed for complex investigative tasks and information synthesis, initially targeting developers followed closely by consumer applications.

The upgraded agent aims to streamline how users gather and synthesize extensive information over prolonged periods. It operates through an autonomous reasoning core that defines its own investigative plans. When prompted, the agent formulates queries, processes the results, identifies knowledge gaps, and conducts further searches. This iterative process enables the system to traverse intricate data landscapes with improved accuracy.

Google’s Strategy to Improve Accuracy

A central feature of this upgrade is its focus on minimizing the risk of misinformation. Google has trained the Deep Research agent to significantly reduce the occurrence of AI hallucinations, thereby lowering the generation of false information. For tasks that require deep reasoning, a minor error at the beginning of a multi-step investigation can render the final report ineffective; thus, maintaining a high level of factual accuracy is essential.

Google has validated the agent’s capabilities by achieving commendable results on advanced reasoning benchmarks. These include the Humanity’s Last Exam (HLE), which assesses general knowledge and complex reasoning, and the open-sourced DeepSearchQA benchmark, developed by Google to evaluate thorough, multi-step web research.

Applications in Diverse Fields

Early testing has demonstrated the high precision of this technology across demanding sectors. In financial services, the agent is proving effective in automating the initial, labor-intensive stages of due diligence. Similarly, in the biotechnology field, it is assisting in speeding up drug safety research by analyzing extensive biomedical literature.

To support external development, Google has introduced the new Interactions API, a unified interface that allows developers to integrate the Gemini models and the Deep Research agent into their applications. This initiative aims to expedite the creation of customized, AI-driven research tools by third parties.

For consumers, Google plans to weave the Deep Research agent’s capabilities into its flagship services, including the main Gemini app, Google Search, and the research application NotebookLM. This integration is expected to transform how users engage with information online, with the AI handling the preliminary research tasks, thereby enhancing user experience and efficiency.