Google launches next-gen AI research tool amid GPT-5.2 rollout
What's the story
Google has unveiled a revamped version of its research agent, Gemini Deep Research. The new tool is built on the company's advanced foundation model, Gemini 3 Pro. Unlike its predecessor, this updated agent can do more than just generate research reports. It also lets developers integrate Google's SATA-model research capabilities into their own applications via a new Interactions API. Google's launch coincides with OpenAI unveiling GPT-5.2 model.
Advanced capabilities
A tool for complex information synthesis
The new Gemini Deep Research tool is an agent that can synthesize large amounts of information and manage a big context dump in the prompt. Google says customers use it for tasks ranging from due diligence to drug toxicity safety research. The tech giant also plans to integrate this deep research agent into its services like Google Search, Google Finance, Gemini App, and NotebookLM.
Model precision
Deep Research leverages Gemini 3 Pro's accuracy
Google says that Deep Research takes advantage of Gemini 3 Pro's status as its "most factual" model. The model is specifically trained to minimize hallucinations during complex tasks. AI hallucinations, where the LLM fabricates information, are especially problematic for long-running, deep reasoning agentic tasks with multiple autonomous decisions over extended periods.
Benchmarking progress
Google introduces new benchmark to test AI agents
To prove its progress claims, Google has introduced a new benchmark called DeepSearchQA. The benchmark is designed to test agents on complex, multi-step information-seeking tasks and has been open-sourced by the tech giant. It was also tested on Humanity's Last Exam, an independent general knowledge benchmark filled with niche tasks; and BrowserComp, a benchmark for browser-based agentic tasks.
Performance comparison
Deep Research outperforms competitors in benchmarks
Google's new agent outperformed its competitors on its own benchmark and Humanity's Last Exam. However, OpenAI's ChatGPT 5 Pro was a close second across the board and slightly outperformed Google on BrowserComp. The performance comparisons were made public just as OpenAI launched its highly anticipated GPT 5.2 model, codenamed Garlic, which reportedly outperforms rivals including Google on a range of benchmarks.