Google has launched its deepest AI research proxy yet — on the same day it dropped OpenAI GPT-5.2

Google Released Thursday released a “re-imagined” version of its research agent Gemini Deep Research based on The advanced foundation model that has received a lot of hypeGemini 3 Pro.

This new agent is not designed solely to produce research reports, although it can still do so. It now allows developers to include Google’s SATA search capabilities in their own applications. This capability is made possible by a new Google feature Reactions APIwhich is designed to give developers more control in the coming era of agentic AI.

The new Gemini Deep Research tool is equipped to collect mountains of information and handle a large context dump in the router. Google says customers use it for tasks ranging from due diligence to drug safety and toxicity research.

Google also says it will soon integrate this new deep research agent into services including Google Search, Google Finance, the popular Gemini app, and NotebookLM. This is another step toward preparing for a world where humans no longer search for anything on Google, as AI agents do.

The tech giant says Deep Research takes advantage of the Gemini 3 Pro mode as its “most realistic” model that has been trained to reduce hallucinations during complex tasks.

AI hallucinations — where the MBA makes things up — is a particularly critical issue for long-running, deep agent tasks, in which many independent decisions are made over the course of minutes, hours, or longer. The more choices the LLM has to make, the greater the chance that one hallucinogenic choice will invalidate the entire output.

To prove its claims of progress, Google has also created another standard (as if the world of AI needed another standard). The new standard is called DeepSearchQA, and aims to test agents on complex, multi-step information-seeking tasks. Google has open sourced this standard.

It also examined deep research into The Last Test of Humanity, an interesting independent benchmark of general knowledge full of impossible specialized tasks; and BrowserComp, a standard for browser-based proxy tasks.

As you might expect, Google’s new agent outperforms the competition on its own and Humanity’s benchmarks. However, OpenAI’s ChatGPT 5 Pro came in a surprising second all the way and slightly edged out Google in BrowserComp.

But those benchmarks were almost obsolete the moment Google published them. Because on the same day, OpenAI released its highly anticipated version of GPT 5.2 – codenamed Garlic. OpenAI says its latest model outperforms its competitors — especially Google — on a range of typical benchmarks, including OpenAI’s home-grown benchmark.

Perhaps one of the most interesting parts of this announcement is the timing. Knowing that the world was waiting for the release of Garlic, Google dropped some AI news of its own.