Google Releases Advanced Gemini Deep Research, Competes With OpenAI's Latest Model

JAKARTA Together with OpenAI which launched GPT-5.2, Google also released its newest agent. The agent is Gemini Deep Research which uses the Gemini 3 Pro model.

This agent is claimed to be much more sophisticated than the previous version. With Gemini 3 Pro support, this in-depth reasoning agent is trained to reduce hallucinations. As a result, the agent can carry out complex tasks.

Automatically, Gemini Deep Research can direct users to complex information landscapes with high claimed accuracy. Deep Research can plan investigations, from formulating queries to identifying knowledge gaps.

This improvement makes web search display much better and supports deep navigation to websites to get specific data. For the first time, developers can embed this agent into its application through an interaction API.

Gemini Deep Research recorded the best results on Humanity's Last Exam (HLE) and DeepSearchQA testing. The results show that these agents are shown to be able to produce well-investigated reports at a much lower cost.

Along with this launch, Google also opened the code for a new web research agent benchmark, DeepSearchQA. This benchmark is designed to test agents' completeness on complicated multi-step web research tasks, which have often failed to be captured by other benchmarks.

Developers can already access the API Interaction with the Gemini API key via Google AI Studio. The upcoming update will include creating the original graph for visual analytic reports and expanding connectivity through the support of the Context Protocol Model (MCP).

Google also plans to bring Deep Research to Google Search, NotebookLM, Google Finance, and upgrade it to the Gemini App in the near future.