At a critical juncture in the global AI narrative moving towards the &#34;Agent era,&#34; Google chose a rather dramatic release timing.
On Thursday, coinciding with OpenAI's highly anticipated launch of GPT-5.2 (internal code name Garlic), Google simultaneously released the most powerful deep research-type Agent to date—Gemini Deep Research &#34;remastered,&#34; claiming it is based on its most advanced Gemini 3 Pro model. On the same day, DeepMind also announced the establishment of the first automated research laboratory in the UK, utilizing AI and robotics to accelerate materials science experiments.
This is not a &#34;collision,&#34; but rather a carefully orchestrated strategy: while competitors focus global attention, Google responds with a product of greater strategic significance—pushing Agent towards operating system-level capabilities.
<h2>From &#34;writing reports&#34; to &#34;embedding applications&#34;: Google attempts to define the infrastructure-level entry of Agents</h2>
The new Gemini Deep Research is no longer a traditional tool for &#34;automatically writing research reports&#34;; it is positioned as:
<ul>
<li>
A deep research Agent capable of handling larger-scale contexts and digesting &#34;information mountains.&#34;
</li>
<li>
An &#34;AI research kernel&#34; that can be directly embedded into apps by developers.
</li>
<li>
A long-chain reasoning Agent capable of continuously executing multi-step reasoning tasks for several minutes or even hours.
</li>
</ul>
These capabilities generally point to a trend: Google is attempting to turn Deep Research into the &#34;underlying capability of search engines&#34; for the future AI era.
To make it easier for developers to embed Deep Research into their applications, Google launched a brand new Interactions API. This is equivalent to packaging &#34;search + multi-step reasoning + evaluation&#34; into an operating system-level service.
Google also announced that Deep Research will gradually be integrated into Google Search, Google Finance, the Gemini app, and NotebookLM.
In other words: the future is not about users &#34;googling,&#34; but rather your Agent googling everything for you.
Why Deep Research? Google aims to address the biggest pain point after AI transitions to Agents: hallucination rates.
Google claims that Deep Research benefits from the &#34;higher factuality&#34; of Gemini 3 Pro, which can reduce the distortion of results caused by the accumulation of hallucinations in long-chain reasoning tasks. Such tasks may last for dozens of minutes or even hours, posing significant risks.
<h2>Google sets a new benchmark: DeepSearchQA, BrowserComp, Humanity’s Last Exam</h2>
To validate performance, Google released a new DeepSearchQA benchmark for testing multi-step information retrieval and open-sourced this benchmark
In Google's own benchmarks and the &#34;Last Exam for Humanity&#34; benchmark tests, the new Agent outperformed its competitors, but OpenAI's ChatGPT 5 Pro performed closely and slightly surpassed Google in the BrowserComp test.
However, these benchmark comparisons were almost outdated the moment Google released them. Because on the same day, OpenAI released the highly anticipated GPT-5.2, claiming to have the strongest agent coding, surpassing human experts. OpenAI claims that the model outperformed its competitors in a series of typical benchmark tests, particularly Google.
From the results, this day became a &#34;face-off&#34; between Google and OpenAI in direct competition.
Google chose to simultaneously announce Deep Research on the day of the GPT-5.2 release, which is hard not to be seen as a proactive competitive stance—both a response to OpenAI's new model and an attempt to seize the developer entry point amid the rapid development of Agentization.
As AI Agents are about to become the new generation of &#34;search boxes,&#34; the competition between Google and OpenAI has shifted from a model war to who can become the infrastructure for future information access methods

GOOG

Google has launched the most powerful deep research agent to date—Gemini Deep Research's "remastered version," attempting to define the infrastructure-level entry point for agents. This agent can handle larger-scale contexts, can be directly embedded by developers into the app's "AI research kernel," and can continuously perform multi-step reasoning tasks for several minutes or even hours

- On the same day OpenAI launched GPT-5.2, Google introduced its advanced deep research agent, GeminiDeep Research, based on the Gemini 3 Pro model.  
- This new agent aims to redefine the role of AI in research by enabling long-chain reasoning tasks and integrating seamlessly into applications through a new Interactions API.  
- As Google and OpenAI compete for dominance in AI infrastructure, the focus has shifted from model performance to establishing foundational capabilities for future information access.

Google's Yangmou: Launching the "Deepest" Research Agent in History on the Release Day of GPT-5.2