Google has released a “reimagined” version of its Gemini Deep Research agent, offering developers access to the tech giant’s most advanced autonomous research capabilities.
Powered by Gemini 3 Pro, the updated system shifts Deep Research from a specialized, report-writing assistant to an autonomous research agent designed to conduct long-form reasoning and complex analysis.
In essence, the renewed system is able to analyze and condense huge amounts of complex data and provide more detailed and accurate reports than before.
Google said in a blog post that the revamped system is also designed to reduce hallucinations, a critical requirement for agents operating over extended periods of time.
“[This] agent is optimized for long-running context gathering and synthesis tasks,” Google said. “By scaling multi-step reinforcement learning for search, the agent autonomously navigates complex information landscapes with high accuracy.”
“Deep Research iteratively plans its investigation — it formulates queries, reads results, identifies knowledge gaps, and searches again,” the tech giant added.
The release also allows developers to embed Google’s research capabilities directly into their own applications via the company’s new Interactions API. This interface is designed to allow interaction with different agents and give developers greater control amidst a rise in agentic AI.
According to Google, early users are already applying Gemini Deep Research to areas such as financial services, biotech and market research, as well as drug toxicity safety research, where precision and traceability are essential.
Google also said it will be integrating its new deep research agent across its services, including Google Search, Google Finance, its Gemini App and NotebookLM.
Alongside the launch, Google also unveiled a new benchmark for complex web searches. Dubbed DeepSearchQA, the framework evaluates agents on “intricate, multi-step information-seeking tasks.”
Using the benchmark, Google said it observed “significant” performance gains when agents were given more time to perform searches and reasoning steps. This, the company said, is something it plans to further develop in future releases.
The news came on the same day that OpenAI unveiled GPT-5.2, described as its “most capable series yet for professional knowledge work.”
The simultaneous releases underscore the intensifying competition between the two companies as they race to define the next generation of agentic AI systems.