Google has launched an upgraded version of its Gemini Deep Research Agent in late 2025, opening it to developers for the first time. Powered by Gemini 3 Pro, the tool is designed to enable safer, more reliable AI agents capable of complex, multi-step web research and reasoning.

BulletsIn

• Google opens Gemini Deep Research Agent to developers for the first time
• Tool evolved from Gemini app (launched late 2024) to full developer API
• Agent performs iterative research: search, read, identify gaps, refine queries
• Built on Gemini 3 Pro, Google’s most advanced multimodal AI model
• Improved reasoning core reduces hallucinations, boosts long-form accuracy
• Outperformed standard Gemini 3 Pro web search in internal tests
• Google introduces DeepSearchQA, an open-source benchmark for real-world research
• Benchmark includes 900 causal-chain tasks across 17 subject areas
• Developers get tools like PDF/CSV parsing, citations, structured reports, JSON outputs
• New Interactions API enables stateful, multi-step, autonomous AI workflows