Google just supercharged search! Meet Gemini deep research

In an official blog post, the company said: “For the first time, developers can embed Google’s most advanced autonomous research capabilities directly into their own applications.”

Google has also open-sourced a new web research agent benchmark, DeepSearchQA, designed to evaluate an agent’s comprehensiveness and accuracy in complex web research tasks.

According to Google, Gemini Deep Research is engineered for long-running context gathering and synthesis, making it ideal for tasks that demand deep reasoning and meticulous information processing.

The company noted that the agent’s reasoning engine is powered by Gemini 3 Pro, described as Google’s “most factual model yet.” It has been specifically trained to minimize hallucinations and maximize report accuracy during advanced, multi-stage research workflows. By scaling multi-step reinforcement learning for search, the agent can autonomously navigate intricate information ecosystems with high precision.

Google added that Deep Research employs an iterative investigation approach:
It plans each step, formulates queries, reviews results, identifies knowledge gaps, and refines its searches. The latest release includes significantly enhanced web search capabilities, enabling the agent to dig deeper into websites to extract highly specific data.