Google Releases Updated Gemini Deep Research

Google has unveiled a "reimagined" version of its Gemini Deep Research agent, offering developers access to the tech giant's most advanced autonomous research capabilities.

The system, powered by Gemini 3 Pro, transforms Deep Research from a specialized report-writing assistant into a fully autonomous research agent designed for long-horizon reasoning and complex analysis.

In essence, the upgraded platform can analyze and synthesize vast amounts of intricate data, delivering reports that are more detailed and accurate than ever before.

As Google explained in a blog post, the redesigned agent is also engineered to minimize hallucinations—a critical improvement for AI systems performing extended research tasks.

"The agent is optimized for long-context retrieval and synthesis tasks," Google stated. "By extending multi-step reinforcement learning to search, the agent can autonomously navigate complex information environments with high precision."

The company added: "Deep Research conducts iterative investigation planning—it formulates queries, reads results, identifies knowledge gaps, and then searches again."

This new release also enables developers to embed Google’s research functionality directly into their own applications via a brand-new Interaction API. The interface is built to support interactions with various agents, giving developers greater control amid the rising trend of agent-based AI systems.

According to Google, early adopters have already applied Gemini Deep Research in fields such as financial services, biotechnology, market research, and drug toxicity safety studies—domains where accuracy and traceability are paramount.

Google also announced plans to integrate this enhanced deep research agent across multiple services, including Google Search, Google Finance, the Gemini app, and NotebookLM.

Concurrent with the launch, Google introduced a new benchmark for complex web search called DeepSearchQA. This framework evaluates agents on "complex, multi-step information-seeking tasks."

Based on these benchmarks, Google reported that performance improves "significantly" when the agent is allowed more time for searching and reasoning—a trajectory the company intends to further explore in future iterations.

The announcement comes on the same day OpenAI launched GPT-5.2, touted as the "most powerful expertise series to date."

The simultaneous releases highlight the intensifying competition between the two tech leaders, both racing to define the next generation of intelligent AI systems.