Google has released an enhanced version of its Gemini Deep Research agent to developers through a new programming interface, the company announced Wednesday.
The autonomous research tool can now be embedded directly into third-party applications for the first time, according to the blog post. The release comes alongside DeepSearchQA, an open-source benchmark designed to evaluate agents performing complex web research tasks.
According to Google, the Gemini Deep Research agent is optimized for lengthy information-gathering and synthesis tasks. The agent uses Gemini 3 Pro as its reasoning core, which the company describes as its most factual model. The system has been specifically trained to minimize errors and maximize report quality during complex tasks.
The agent operates by iteratively planning its investigation. It formulates queries, reads results, identifies knowledge gaps, and conducts additional searches. The article states that this release features substantially improved web search capabilities, allowing the agent to navigate deeply into websites for specific data.
Google reports that the new agent achieves state-of-the-art results on several benchmarks. It scored 46.4 percent on the full Humanity's Last Exam set, 66.1 percent on DeepSearchQA, and 59.2 percent on BrowseComp. The company says the agent is also optimized to generate reports at lower cost.
New Benchmark for Research Agents
The company is releasing DeepSearchQA to address what it sees as limitations in existing benchmarks. According to the post, traditional tests often fail to capture the complexity of real-world, multi-step web research.
DeepSearchQA features 900 hand-crafted tasks across 17 fields, where each step depends on prior analysis. Unlike traditional fact-based tests, the benchmark measures comprehensiveness, requiring agents to generate complete answer sets. This evaluates both research precision and retrieval recall, the article states.
The benchmark also serves as a diagnostic tool for the benefits of extended processing time. Internal evaluations showed significant performance gains when the agent performed more searches and reasoning steps, according to Google.
Early Applications
The article cites early testing in several industries. Financial firms are reportedly using Gemini Deep Research to automate initial stages of due diligence. The agent aggregates market signals, competitor analysis, and compliance risks from web and proprietary sources.
In scientific research, Axiom Bio, which builds systems to predict drug toxicity, found that Gemini Deep Research provided unprecedented depth across biomedical literature, according to the company's feedback cited in the post.
Developer Features
For developers, the agent offers several capabilities. It analyzes user-provided documents including PDFs and spreadsheets alongside public web data. Users can control output through prompting, defining structure and formatting requirements. The system provides detailed citations for claims and supports structured outputs for easy parsing.
Developers can access the agent through the new Interactions API using a Gemini API key from Google AI Studio, according to the documentation.
Google plans future updates including native chart generation for visual reports and expanded connectivity through Model Context Protocol support for custom data sources. The company is also working to bring Gemini Deep Research to Vertex AI for enterprise users.
The Deep Research agent will soon be available in Google Search, NotebookLM, Google Finance, and an upgraded version in the Gemini App, according to the announcement.
I truly appreciate you spending your valuable time here. To help make this blog the best it can be, I would love your feedback on this post. Let me know in the comments: How could this article be better? Was it clear? Did it have the right amount of detail? Did you notice any errors?
If you found it valuable, please consider sharing it.