Google Opens Gemini Deep Research To Developers Through Interactions API

Google has opened access to its Gemini Deep Research agent, positioning it as a developer-ready system through the newly launched Interactions API. Alongside it arrives DeepSearchQA, an open-sourced benchmark designed to test how well AI agents handle complex, multi-step web research.

The move reflects a broader industry shift: developers and enterprise teams are increasingly looking for systems that do not just answer questions but independently plan, gather, and synthesise information from large and varied datasets. This is the capability Google is now putting into builders’ hands.

The team behind the release, Lukas Haas, Product Manager, Google DeepMind, and Shrestha Basu Mallick, Group Product Manager, Google DeepMind, notes that the system is designed with long-running reasoning tasks in mind and uses Gemini 3 Pro as its core.

How The Deep Research Agent Works

Gemini Deep Research functions as an autonomous research assistant that can run multi-step investigations. It plans queries, reviews results, detects gaps, and continues searching until it reaches a comprehensive answer.

This version improves its ability to navigate deeper into websites, enabling retrieval of niche or layered information. The model reduces hallucinations through multi-step reinforcement learning and is tuned to generate long-form, evidence-linked outputs.

According to Google, the agent now records leading results across internal benchmarks, including Humanity’s Last Exam (46.4%), DeepSearchQA (66.1%), and BrowseComp (59.2%).

DeepSearchQA Expands the Benchmark Landscape

With DeepSearchQA, Google attempts to close the gap between typical fact-based evaluations and how real-world research actually works. The benchmark contains 900 “causal chain” tasks across 17 fields, where each question depends on a previous inference.

Unlike standard QA tests, DeepSearchQA requires exhaustive answer sets, an attempt to measure not just precision but how thoroughly an agent can retrieve and compile information.

Google’s internal comparisons of pass@1 vs. pass@8 also illustrate how allowing more “thinking time” improves output quality, an area the company plans to develop further.

The benchmark assets, dataset, leaderboard, and technical report have also been released to support academic and developer research.

Real-World Use Cases Emerging Across Industries

Early testing shows uptake in sectors where high-precision and context-rich research is key.

Financial Services

Investment teams are using the agent to accelerate due diligence. By pulling together signals across public and proprietary sources, teams are compressing days of foundational research into hours.

As KJ Sidberry, Partner, GV, notes: “Gemini Deep Research agent has been a huge accelerant to our diligence processes, shortening our research cycles from days to hours without loss of fidelity or quality. It feels like having an army of experts ready to go in support of our most ambitious analyses.”

Biotech and Scientific Workflows

Deep research is also being applied in early-stage biomedical analysis. Companies such as Axiom Bio are incorporating it to parse scientific literature and map complex biological mechanisms.

Alex Beatson, co-founder of Axiom Bio, explains: “Gemini Deep Research surfaces granular data and evidence at and beyond what previously only a human researcher could do. We're excited to build on this as a foundation for agentic systems that reason from molecular mechanisms to experimental data and clinical outcomes and empower scientists to develop safer medicines.”

Developer Controls and Integration Options

For builders, the agent supports several features aimed at enabling custom research workflows:

• Unified document and web synthesis (supporting PDFs, CSVs, docs, and web data).
• Control over report structure via prompts, including tables and formatting.
• Detailed citations for each claim.
• Structured JSON schema outputs for downstream applications.

Developers can access the agent through the Interactions API using a Gemini API key from Google AI Studio.

Google says upcoming updates will introduce:

• Native chart generation
• Enhanced Model Context Protocol (MCP) support for tapping into custom data sources
• Integration with Vertex AI for enterprise adoption

Additionally, Gemini Deep Research will soon be integrated into products including Google Search, NotebookLM, Google Finance, and the Gemini app.