Please note: Most, if not all, of the articles published at this website were completed by Chat GPT (chat.openai.com) and/or copied and possibly remixed from other websites or Feedzy or WPeMatico or RSS Aggregrator or WP RSS Aggregrator. No copyright infringement is intended. If there are any copyright issues, please contact: bicycledays@yahoo.com.
Google launched on Thursday a “reimagined” model of its analysis agent Gemini Deep Analysis based mostly on its much-ballyhooed state-of-the-art basis mannequin, Gemini 3 Professional.
This new agent isn’t simply designed to supply analysis studies — though it may possibly nonetheless do this. It now permits builders to embed Google’s SATA-model analysis capabilities into their very own apps. That functionality is made doable by way of Google’s new Interactions API, which is designed to provide devs extra management within the coming agentic AI period.
The brand new Gemini Deep Analysis software is an agent geared up to synthesize mountains of knowledge and deal with a big context dump within the immediate. Google says it’s utilized by clients for duties starting from due diligence to drug toxicity security analysis.
Google additionally says it can quickly be integrating this new deep analysis agent into providers, together with Google Search, Google Finance, its Gemini App, and its common NotebookLM. That is one other step towards getting ready for a world the place people don’t Google something anymore — their AI brokers do.
The tech large says that Deep Analysis advantages from Gemini 3 Professional’s standing as its “most factual” mannequin that’s educated to attenuate hallucinations throughout complicated duties.
AI hallucinations — the place the LLM simply makes stuff up — are an particularly essential problem for long-running, deep reasoning agentic duties, wherein many autonomous choices are remodeled minutes, hours, or longer. The extra decisions an LLM has to make, the better the prospect that even one hallucinated alternative will invalidate your complete output.
To show its progress claims, Google has additionally created one more benchmark (as if the AI world wants one other one). The brand new benchmark is unimaginatively named DeepSearchQA and is meant to check brokers on complicated, multi-step information-seeking duties. Google has open sourced this benchmark.
Techcrunch occasion
San Francisco | October 13-15, 2026
It additionally examined Deep Analysis on Humanity’s Final Examination, a way more apparently named, impartial benchmark of basic information full of impossibly area of interest duties; and BrowserComp, a benchmark for browser-based agentic duties.
As you would possibly count on, Google’s new agent bested the competitors by itself benchmark, and Humanity’s. Nevertheless, OpenAI’s ChatGPT 5 Professional was a surprisingly shut second all the best way round and barely bested Google on BrowserComp.
However these benchmark comparisons have been out of date nearly the second Google printed them. As a result of on the identical day, OpenAI launched its extremely anticipated GPT 5.2 — codenamed Garlic. OpenAI says its latest mannequin bests its rivals — particularly Google — on a collection of the everyday benchmarks, together with OpenAI’s homegrown one.
Maybe one of the crucial attention-grabbing elements of this announcement was the timing. Realizing that the world was awaiting the discharge of Garlic, Google dropped some AI information of its personal.
