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

On Thursday, Google introduced an upgraded version of its research agent, Gemini Deep Research, built on the cutting-edge foundation model, Gemini 3 Pro. This revamped agent extends beyond generating research reports; it empowers developers to incorporate Google’s advanced research capabilities into their applications through the new Interactions API. This initiative aims to provide developers with enhanced control as we transition into an era dominated by intelligent agents. Gemini Deep Research is designed to efficiently synthesize vast amounts of information while managing large context inputs. Google reports that clients are utilizing this tool for a variety of applications, including due diligence and drug toxicity safety assessments. Furthermore, the tech giant plans to integrate this deep research agent across multiple platforms such as Google Search, Google Finance, the Gemini App, and the widely used NotebookLM. This marks a significant stride toward a future where AI agents handle inquiries on behalf of users, eliminating the need for traditional search methods. A notable feature of this advancement is its reliance on Gemini 3 Pro, which is touted as Google's most reliable model, specifically engineered to reduce inaccuracies, often referred to as AI hallucinations. These hallucinations present a critical challenge during intricate tasks requiring prolonged reasoning, where a single erroneous decision can compromise the entire output. To validate its claims of progress, Google also launched a new benchmark named DeepSearchQA, aimed at evaluating agents on complex, multi-step information retrieval tasks. Additionally, the tool was assessed using the independent benchmark, Humanity’s Last Exam, which includes a range of obscure knowledge tasks, as well as BrowserComp for web-based agent tasks. As expected, Google’s new agent excelled in its own benchmark and in Humanity’s Last Exam. Nevertheless, OpenAI’s ChatGPT 5 Pro closely followed, even outperforming Google in the BrowserComp test. Coincidentally, the announcement coincided with OpenAI's launch of GPT 5.2, codenamed Garlic, which OpenAI claims surpasses its competitors, particularly Google, across a variety of standard benchmarks. The synchronization of these releases highlights the competitive landscape of AI development, with both companies striving for superiority in a rapidly evolving field.