Google Unveils Upgraded Gemini Deep Research Agent

Google has upgraded its Gemini Deep Research agent, expanding access to the system and positioning it for long, multi-step research tasks that require gathering and synthesizing information from across the web.

The upgrade gives developers access to one of Google’s most advanced autonomous research systems through its Interactions API, as the company steps up competition in AI-powered research and analysis tools.

The updated agent is built to handle extended investigations rather than short factual queries. Google said it can plan research step by step, generate search queries, review results, identify gaps and continue searching until it has sufficient information to produce a coherent report.

The system runs on Gemini 3 Pro, which Google describes as its most factual model so far and says is trained to reduce hallucinations during long-form research tasks.

“This release features vastly improved web search, allowing it to navigate deep into sites for specific data,” Google DeepMind product managers Lukas Haas and Shrestha Basu Mallick wrote in a post announcing the update.

Google said the upgrade significantly improves web navigation, enabling the agent to extract information from deeper sections of websites and handle layered or complex page structures more effectively.

Early benchmark results suggest stronger performance. The agent scored 46.4% on Humanity’s Last Exam, 66.1% on DeepSearchQA and 59.2% on BrowseComp. Google also said the system produces detailed research outputs at a lower computational cost than earlier versions.

Alongside the agent update, Google released DeepSearchQA, an open-source benchmark designed to evaluate how well AI systems handle multi-step research tasks that depend on information gathered earlier in the process. 

The benchmark includes 900 tasks across 17 fields and emphasizes completeness rather than isolated fact retrieval. Internal testing showed clearer gains when the agent was given more time to search and reason, Google said.

The upgraded Gemini Deep Research agent is expected to roll out across Google Search, NotebookLM, Google Finance and the Gemini app over time.

The release comes as competition in AI research tools intensifies. Google’s benchmarks show the updated agent outperforming several rivals, though OpenAI’s ChatGPT 5 Pro placed close behind and scored higher on BrowseComp. 

OpenAI announced its own model update, GPT-5.2, hours after Google’s release, underscoring how quickly the race for professional and research-oriented AI systems is moving.