Gemini Deep Research: Multi-Step Reasoning via the Interactions API
Google has launched an advanced version of Gemini Deep Research as an autonomous agent designed for long-running context gathering and synthesis tasks. Instead of just answering a single prompt, the agent iteratively plans its investigation, formulates search queries, reads results, identifies knowledge gaps, and searches again. It is powered by a specialized reasoning core on Gemini 3 Pro and is designed to run deep into websites to pull specific data points.
Key Features & Technology
- Iterative Planning & Web Search: The agent does not just do a single semantic search. It designs a research plan and executes dozens of sequential queries, adjusting its path based on what it discovers.
- Interactions API: Google has released this agent through the new Interactions API (available via Google AI Studio), allowing developers to build custom deep research workflows directly into their applications.
- Data Synthesis & Report Steerability: It handles massive context windows gracefully and synthesizes web data and user-provided files (PDFs, CSVs, docs) into structured, highly detailed reports with granular citations. Users can specify custom report structures, headers, and data table layouts.
- Performance Benchmarks: Google evaluates the agent using DeepSearchQA (their new open-source benchmark consisting of 900 multi-step "causal chain" tasks requiring deep retrieval recall) and Humanity's Last Exam (HLE), where Gemini Deep Research achieves a state-of-the-art 46.4% on the full HLE set.
Pricing and Business Model
- Developer API: Priced programmatically based on token consumption (input/output) and tool usage under the new Interactions API.
- Consumer/Enterprise Apps: Integrated into Gemini Advanced ($20/month Google One AI Premium plan), NotebookLM, Google Finance, and Google Search. It is also being rolled out to Vertex AI for enterprise environments.
"Gemini Deep Research is an agent optimized for long-running context gathering and synthesis tasks. The agent's reasoning core uses Gemini 3 Pro, our most factual model yet, and is specifically trained to reduce hallucinations and maximize report quality during complex tasks." — https://blog.google/innovation-and-ai/technology/developers-tools/deep-research-agent-gemini-api/
"Deep Research iteratively plans its investigation – it formulates queries, reads results, identifies knowledge gaps, and searches again. This release features vastly improved web search, allowing it to navigate deep into sites for specific data." — https://blog.google/innovation-and-ai/technology/developers-tools/deep-research-agent-gemini-api/