New Gemini Deep Research Agent Redefines Autonomous Web Research

Google has rolled out the Gemini Deep Research Agent, its most powerful AI research system built to navigate complex information with high accuracy. The company says the agent brings a new level of depth, precision, and autonomy to long-running research tasks.

A New Era of Deep, Autonomous Web Research

The Gemini Deep Research Agent is designed for multi-step reasoning and long context sessions. It plans every step, forms new queries, identifies gaps, and continues searching. Therefore, it works like a full research assistant that keeps digging until it builds a complete picture.

Google claims it uses the Gemini 3 Pro core to reduce hallucinations and deliver highly factual reports. The system also features a major upgrade in web navigation, allowing it to move deeper into websites and extract intricate data points.

DeepSearchQA Sets a New Benchmark Standard

Google also introduced DeepSearchQA, a benchmark built for evaluating deep research agents. It includes 900 tasks that require multi-step reasoning, cross-linking, and comprehensive answers.
Traditional benchmarks test recall. However, DeepSearchQA tests whether an agent can think across steps and deliver complete results.

Google says the benchmark also highlights the value of longer “thinking time,” where more search steps lead to stronger answers.

Real-World Use Cases Show Immediate Impact

The Gemini Deep Research Agent is already being used in finance, biotech, and market research. Financial firms are speeding up early-stage due diligence by pulling signals from public and private sources. Biotech teams like Axiom Bio are using it to scan scientific literature and uncover insights that previously took days to collect.

Developers Can Build With the Interactions API

Google has made the agent available through the new Interactions API, which lets developers embed advanced research capabilities into their applications. The agent reads PDFs, CSVs, and long documents, then merges them with web data.
It also supports structured output, JSON formats, and detailed citations for easy verification.

Google plans to bring richer outputs such as native charts and expand enterprise support through Vertex AI.