.png){kind=logo}
Google has launched an updated version of its research agent, Gemini Deep Research, that uses Google’s latest flagship AI model, Gemini 3 Pro. It’s yet another instance of Google and OpenAI’s rivalry as the reveal happened on the same day OpenAI launched GPT-5.2 Garlic.
Google has enabled Gemini Deep Research to be incorporated into third-party apps. It is made possible via the company’s newly launched Interactions API. According to Google, the upgraded version provides developers with more control as it transitions to an ‘agent-first’ AI paradigm.
Deep Research can process large windows of context and synthesize knowledge from multi-step reasoning. It is already being employed at enterprises for due diligence analysis, drug toxicity research, and more.
The company will integrate Deep Research into various consumer and enterprise products, including Google Search, Google Finance, the Gemini app, and NotebookLM. This shows the tech firm’s plan to build an ecosystem in which users allow the AI agent to search and analyze on their behalf.
Google also emphasizes that the upgrade benefits from Gemini 3 Pro, its ‘most factual’ model yet. The Pro model focuses on learning ways to suppress hallucinations, which is an ‘essential challenge’ for any agent making decisions on its own for an extended period.
To validate its claims, Google introduced a new benchmark task set, DeepSearchQA, focused on highly complex, multi-step information-seeking problems. It also compared its model on Humanity’s Last Exam, an extremely tough independent benchmark, and BrowserComp, which assesses performance on browser-based agentic tasks.
As anticipated, Google’s agent performed well on Google’s benchmark and all other tests. However, OpenAI’s ChatGPT 5 Pro followed closely and sometimes beat Google on BrowserComp.
Also Read: Google AI Plus Plan Launches in India With Gemini 3 Pro at Rs. 399
It couldn’t have been more apt. While the industry was looking forward to OpenAI’s GPT-5.2, Google stole the spotlight with its launch. However, that brief moment soon came to an end when GPT-5.2 proclaimed itself better on a new set of benchmarks.
