Highlights

Google recently unveiled its innovative AI agent, Gemini Deep Research, built on the cutting-edge Gemini 3 Pro foundation model. A standout feature of this agent is its ability to integrate Google’s SATA-model for developers' applications via the Interactions API. The tool is set to revolutionize tasks like due diligence and drug research. Notably, the timing of this release coincided with OpenAI's debut of GPT-5.2, sparking competitive comparisons.

Sentiment Analysis

• The sentiment around Google's new release is largely positive. The debut of Gemini Deep Research represents a significant leap in AI capabilities, particularly in terms of reducing hallucinations.
• Developers are enthusiastic about the Interactions API offering more control in AI-driven applications.
• Some mixed reactions arise from the coincidental timing with OpenAI's new model, raising debates on the rivals' capabilities.

Article Text

On the same day as OpenAI's anticipated release of GPT-5.2, Google presented its reimagined AI research agent, Gemini Deep Research, which is founded on the Gemini 3 Pro model. This agent is more than just a research tool; it offers developers the ability to embed Google’s advanced research capabilities into their applications through the new Interactions API. This represents a shift toward more autonomous, AI-driven tasks where AI agents take a central role in information synthesis and application development.

The new agent's standout features include its ability to process large amounts of information and handle extensive context, making it suitable for complex tasks such as due diligence and drug development. Google is poised to integrate this technology into various services such as Google Search, Google Finance, the Gemini App, and NotebookLM, marking a move towards a future where AI agents handle much of the information retrieval previously performed by humans.

An interesting aspect of this development is the focus on minimizing AI hallucinations. The Gemini 3 Pro's advancements in factual accuracy are critical to ensuring reliability in long-term, deep reasoning tasks where one error can undermine entire outputs. Google also introduced the DeepSearchQA benchmark to evaluate agents on multi-step information tasks, allowing comparisons to other models.

The release timing drew attention, particularly as Google's agent outperformed others in benchmarks, yet faced close competition from OpenAI’s ChatGPT 5 Pro. Consequently, OpenAI announced their own impressive feat with GPT-5.2, nicknamed Garlic, claiming superiority over Google's model on various tests. This rivalry emphasizes the rapid advancements in AI technology.

Key Insights Table