Google Unveils its Most advanced AI Research Agent Yet, on the Same Day OpenAI Released GPT-5.2

Udaipur, Dec 12, 2025: Google on Thursday introduced a “reimagined” version of Gemini Deep Research, a powerful AI research agent built on its flagship Gemini 3 Pro foundation model.

Unlike earlier versions that focused primarily on generating research reports, the updated Gemini Deep Research agent is now designed to be embedded directly into third-party applications. This leap is powered by Google’s new Interactions API, which gives developers far greater control as they build next-generation agentic AI experiences.

A deeper, more capable research engine

Google says the upgraded agent can synthesize massive volumes of information, process extremely large prompts, and deliver insights across complex domains. Customers are already using it for everything from financial due diligence to drug toxicity and safety research.

In the coming months, Google plans to integrate this agent across its ecosystem—including Google Search, Google Finance, the Gemini app, and NotebookLM—signaling a future where AI assistants may handle information-seeking tasks on behalf of users.

According to Google, Deep Research benefits from Gemini 3 Pro’s reputation as its “most factual” model, specifically trained to reduce hallucinations during long, multi-step reasoning tasks.

Addressing the hallucination problem

Hallucinations remain a significant challenge for agentic systems that operate autonomously for extended periods. When an AI must make hundreds of sequential decisions, even a single incorrect step can compromise an entire output.

To showcase its progress, Google introduced a new benchmark—DeepSearchQA—designed to test AI agents on complex, multi-stage information retrieval tasks. The company has open-sourced the benchmark for researchers.

Google also evaluated Deep Research on:

• Humanity’s Last Exam — an independent and notoriously difficult general-knowledge benchmark full of obscure tasks.
• BrowserComp  a benchmark designed to test browser-based agentic workflows

Unsurprisingly, Google’s agent scored highest on its own test and performed strongly on Humanity’s Last Exam. However, OpenAI’s ChatGPT 5 Pro remained close behind, even outperforming Google slightly on BrowserComp.

Timing tells the story

Those benchmark comparisons quickly became outdated. On the very same day, OpenAI launched GPT-5.2 (codenamed “Garlic”), claiming superior performance across major benchmarks including many where Google had just touted its own wins.

The overlapping announcements highlight the escalating rivalry between Google and OpenAI. With industry expectations swirling around OpenAI’s “Garlic” release, Google strategically timed its Deep Research reveal to stay in the spotlight.

 #GoogleAI #GeminiDeepResearch #Gemini3Pro #OpenAI #GPT52 #AIAgents #AIResearch #TechNews #UdaipurNews #RajasthanUpdates #AICompetition #GoogleVsOpenAI #DeepSearchQA #AgenticAI