Upgraded Deep Research coming to Gemini app, more

Google today announced a “significantly more powerful Gemini Deep Research agent” that will soon be available in consumer apps and is now available for developers.
Today’s announcement is focused on developers. Google has a new Interactions API that serves as its “unified interface for interacting” with models (like Gemini 3 Pro) and agents. Google’s new API reflects the latest model capabilities like “thinking” and advanced tool use that go beyond text generation.
We will expand built-in agents and introduce the ability to build and bring your own agents. This will enable you to connect Gemini models, Google’s built-in agents, and your custom agents using one API.
The first built-in agent is Gemini Deep Research (Preview). Third-party developers can now add “advanced autonomous research capabilities” into their applications.
“Optimized for long-running context gathering and synthesis tasks,” the Gemini Deep Research agent uses Gemini 3 Pro. Google touts how it is “specifically trained to reduce hallucinations and maximize report quality during complex tasks.”
Advertisement - scroll for more content
In response to your prompt, it “formulates queries, reads results, identifies knowledge gaps, and searches again.” There’s also “vastly improved web search, allowing it to navigate deep into sites for specific data.”
By scaling multi-step reinforcement learning for search, the agent autonomously navigates complex information landscapes with high accuracy.
On benchmarks, Google notes state-of-the-art results on Humanity’s Last Exam (reasoning and knowledge), DeepSearchQA (comprehensive web research), and BrowseComp (locating hard to find facts) that surpass Gemini 3 Pro.
Gemini Deep Research achieves 46.4% (versus Gemini 3 Pro’s 43.2%) on the full HLE set, 66.1% (versus 56.6%) on DeepSearchQA and a high 59.2% (versus 49.4%) on BrowseComp: