Google Unveils Advanced Gemini Deep Research for Safer, Smarter AI Agents and Apps: 5 Things to Know
Google has made its advanced Deep Research AI agent accessible to developers for the first time. Under the hood, the Advanced Agent runs on Google’s most advanced multimodal model, the Gemini 3 Pro, which powers the agent’s “reasoning core.”
Google is taking a big leap toward making its most capable AI research tools more accessible and making the way they think a little more human. The company has introduced a new and more powerful version of its Deep Research Agent, which is now open to developers for the first time, as well as a new benchmark designed to test how well AI systems handle complex, multi-step web searches. Think of it as giving developers their own mini research assistant, powered by Gemini, that knows how to dig deeper, double-check their work, and even admit when it doesn’t have all the answers.
Here are five key things to know about Google’s latest AI push.
Deep Research Graduation from Gemini App
First introduced inside the Gemini app in late 2024, Deep Research is now taking to the world. Developers can finally embed Google’s most advanced autonomous research features directly into their products and workflows. The agent doesn’t just pull search results, it thinks about them.
Its workflow resembles how a careful human researcher might work: It formulates queries, reads the results, identifies gaps, and then refines its findings again. This process is repeated until it reaches a conclusion it considers satisfactory. Google says this iterative approach helps the system provide richer, more complete insights than simple quick-response models.
Built on Gemini 3 Pro
Under the hood, the Advanced Agent runs on Google’s most advanced multimodal model, the Gemini 3 Pro, which powers the agent’s “reasoning core.” The company says the model has been fine-tuned to reduce hallucinations, an ongoing challenge for large language models, while improving the accuracy and quality of long-term research summaries.
In internal testing, this version of Deep Research reportedly outperformed the Gemini 3 Pro’s web search capabilities. While Google acknowledges that users should not take every answer as gospel, it argues that thorough research is invaluable for gathering exploratory information, especially when dealing with unfamiliar topics or cross-domain analysis.
A new open-source benchmark for complex web queries
Along with Agent, Google is introducing DeepSearchQA, an open-source benchmark that aims to reflect how real research work unfolds online. The company argues that existing benchmarks often test isolated facts, not the step-by-step logic that humans use to connect the dots across multiple sources.
DeepSearchQA contains 900 hand-crafted “causal chain” tasks across 17 subject areas ranging from history and policy to climate science and health. Each task builds on the previous task, making it a tougher and more realistic test of how well the AI can maintain reasoning over time. Rather than just checking factual correctness, the benchmark measures the completeness of the answer, whether the model fully captures the nuances and dependencies of a question.
To support the community, Google is also releasing a dataset, leaderboards, and technical reports to allow developers and researchers to benchmark their own systems against Deep Research.
Developer-Friendly Features (And What’s Coming Next)
Developers will have access to a number of powerful tools through the Deep Research API. These include PDF, CSV and document parsing, structured report templates, granular source citations, and JSON schema output that make integration easy.
Future updates are set to add native chart generation, allowing the agent to visualize data on its own, and expanded support for the Model Context Protocol (MCP), which lets developers plug in custom data sources. Advanced Deep Research will also soon be introduced on Google Search, NotebookLM, and Google Finance, making its capabilities more widely available to end users.
A new API for the era of “thinking” models
To tie everything together, Google has also unveiled the Interaction API, a new standard for connecting AI models like Gemini 3 Pro and agents like Deep Research. Available in public beta through Google AI Studio, it replaces the simple request-response style of the old GenerateContent interface with something more dynamic and stateful.
The new API supports server-side session management, nested message structures, background execution for long-running tasks, and built-in MCP compatibility. Google describes it as a necessary step toward more autonomous, persistent AI systems that can remember context, plan multi-step logic and handle complex workflows without constant human supervision.
Google’s latest upgrade signals a change in the way the company views AI agents: not just as text generators, but as active researchers who can help build smarter, safer, and more reliable systems. With Deep Research now open to developers and a benchmark to keep them honest, Google is clearly betting on a future where AI doesn’t just answer questions, but learns how to ask better ones.