Google unveils its deepest AI research agent yet, just hours before OpenAI dropped GPT-5.2

Google

Google has launched a major upgrade to Gemini Deep Research, transforming what was previously a specialised research-report generator into a fully fledged agent designed for long-form reasoning, complex analysis and enterprise-grade investigation. The announcement landed on the same day OpenAI released GPT-5.2 — a timing that raised eyebrows across the AI world.

The new Gemini Deep Research agent runs on Gemini 3 Pro, Google’s most advanced and most “factual” model to date. Google says the system is built to minimise hallucinations, a critical requirement for agentic tasks that span minutes, hours or even days. In such long-running operations, a single incorrect decision can derail an entire workflow.

A key part of the rollout is Google’s Interactions API, which lets developers embed deep-research capabilities directly into their own applications. Instead of simply generating reports, the agent can now be wired into enterprise tools, offering reasoning layers for tasks like due diligence, drug safety analysis, or any workflow requiring enormous amounts of information to be synthesised in context.

Google plans to integrate Deep Research into core products, including Google Search, Google Finance, the Gemini app and NotebookLM. It’s another step toward a future where users don’t manually search at all — their AI agents do the collecting, reading and summarising for them.

To support its claims, Google has introduced a new benchmark, DeepSearchQA, designed to test how well agents perform multi-step information-seeking tasks. It also evaluated Gemini Deep Research on external benchmarks such as Humanity’s Last Exam and BrowserComp. On Google’s benchmark and on Humanity’s, the new agent topped the charts. BrowserComp told a more nuanced story, with OpenAI’s ChatGPT 5 Pro narrowly outperforming Google.

But those comparisons were outdated almost instantly. OpenAI released GPT-5.2 — codenamed Garlic — the same day, claiming strong improvements across standard tests and direct superiority over Google’s models on a wide set of reasoning and accuracy benchmarks.

The timing did not go unnoticed. Both companies are now releasing major AI updates within hours of each other, signalling an intensifying race to dominate agentic AI — the next era where models not only answer but act.