Build with Gemini Deep Research (blog.google)

Google has launched an enhanced version of its Gemini Deep Research agent, now accessible via the Interactions API. This advanced agent enables developers to imbue their applications with superior autonomous research capabilities, optimized for nuanced tasks that require extensive context gathering and synthesis. Its core leverages the Gemini 3 Pro model, which is specifically designed to minimize inaccuracies and improve report quality. Notably, the agent uses iterative planning to refine its search, enhancing its capacity to navigate complex information efficiently. The significance of this release lies not only in its advanced performance—evidenced by state-of-the-art metrics on benchmarks like Humanity's Last Exam (HLE) and DeepSearchQA—but also in its open-source benchmark DeepSearchQA, which evaluates agent comprehensiveness across intricate, multi-step web research tasks. By focusing on generating exhaustive answer sets rather than just factual recall, the new benchmark represents a shift towards more realistic assessments of AI capabilities. Early applications in financial services and biotech demonstrate Gemini Deep Research's potential to transform initial research phases, facilitating more profound insights in critical domains like investment due diligence and drug discovery. This positions Gemini Deep Research as a pivotal tool for developers aiming to create the next generation of research automation solutions.