Google has recently released an updated version of its Deep Research Agent, marking the first time this powerful autonomous research feature is made available to developers. Developers can now embed Google's most advanced research capabilities directly into their own applications. The Deep Research Agent was first introduced in the Gemini app at the end of 2024.
The agent uses an iterative research approach, capable of formulating search queries, reading search results, identifying knowledge gaps, and continuing the search until it finds a satisfactory answer. The new version outperforms the previous model and slightly surpasses the web search capabilities of the recently released Gemini 3 Pro.
Google stated that the agent's reasoning core is trained on Gemini 3 Pro and is specifically optimized to reduce errors and improve the quality of reports for complex tasks. Although the deep research system still carries some risk of errors, users can still use it for exploratory information gathering.
In addition to the update to the Deep Research Agent, Google has also launched DeepSearchQA, a new open-source benchmark. Google pointed out that existing benchmarks often fail to accurately reflect complex multi-step web searches. DeepSearchQA includes 900 manually crafted causal chain tasks covering 17 subject areas, with each step depending on the analysis from the previous one. This benchmark evaluates answer completeness, measures search accuracy and retrieval recall, and serves as a diagnostic tool for measuring the benefits of extended thinking time.
Developers can now access a variety of new features, including PDF, CSV, and document analysis, controllable report structures, detailed source citations, and JSON format output. Future updates will add native chart generation and expanded support for custom data sources. The new deep research features will soon be launched in products such as Google Search, NotebookLM, and Google Finance.
Additionally, Google has introduced a new interactive API, a standardized interface that makes it easier for developers to interact with models like Gemini 3 Pro and the Deep Research Agent. The API is currently in public testing and aims to provide developers with the necessary functions for more complex agent applications.
Key Points:
🌟 Google's Deep Research Agent is now available to developers, allowing them to embed advanced research features into their own applications.
📊 The newly launched DeepSearchQA benchmark focuses on evaluating answer completeness for complex web searches.
🛠️ The new interactive API provides developers with more powerful tools to develop complex agent applications.
