Google Search said it is now utilizing the latest Gemini exemplary named 3.5 Flash-Lite. It is being utilized for agentic hunt experiences, but besides apt for Google AI Overviews and AI Mode (I assume). Google wrote, "3.5 Flash-Lite is besides rolling retired successful Google Search," connected the institution news blog.
Google said wherever it is being utilized successful Google Search by saying "we’re besides releasing Gemini 3.5 Flash-Lite, designed for some low-latency tasks and tasks wherever precocious throughput is captious for developers workflows, for illustration agentic hunt and archive processing." Google specifically said "agentic search," arsenic an example.
Google announced information agents and agentic hunt features astatine Google I/O successful May pinch Gemini 3.5.
But Google is apt besides utilizing 3.5 Flash-Lite arsenic an updated exemplary for Google AI Overviews and AI Mode, if not now, astir apt soon.
Robby Stein from Google later said connected X, "It offers stronger instruction pursuing and amended understands personification intent, truthful conversations travel overmuch much seamlessly." Where exactly? In AI Mode? AI Overviews? Just agentic search? I americium not 100% sure.
3.5 Flash-Lite is Google's "fastest, astir cost-effective 3.5-class model, delivering 350 output tokens per 2nd according to the Artificial Analysis Index, besides importantly outperforming anterior Flash-Lite generations successful agentic workflows."
Google added:
3.5 Flash-Lite enables businesslike scaling for agentic systems. Across reasoning levels, the exemplary importantly outperforms 3.1 Flash-Lite. Depending connected the workload, developers tin configure the exemplary to prioritize low-latency, low-cost execution for high-volume tasks pinch the minimal and debased reasoning levels, aliases prosecute higher reasoning levels to process multi-step subagent workloads. The exemplary now besides has machine usage arsenic a built-in instrumentality to reliably support these agentic tasks crossed surfaces.It’s a important measurement up successful coding and agentic tasks arsenic seen successful Terminal-Bench 2.1 (54% vs 31%), agelong discourse arsenic seen successful GDM-MRCR v2 (72.2% vs. 60.1%), and real-world task execution arsenic seen successful GDPval-AA v2 (1140 vs. 642).
In fact, connected galore agentic and coding evals, 3.5 Flash-Lite moreover outperforms 3 Flash, including connected SWE-Bench Pro (54.2% vs. 49.6%) and OSWorld-Verified (74.0% vs. 65.1%), making it a faster & much tin action for workloads connected some 2.5 and 3 Flash.
Here are immoderate capacity charts Google shared:
So possibly Google Search will get a spot faster arsenic well?
Search accusation agents are coming. This lays the instauration for that IMO. At I/O, Google said Search agents would motorboat this summer. Well, it's summer. :)
From I/O: "Information agents will motorboat first for Google AI Pro & Ultra subscribers this summer." https://t.co/Jgi6Qrg7IL pic.twitter.com/mpByMSz3xo
Great momentum from @GoogleDeepMind pinch today’s motorboat of 3 caller Gemini models.
Excited to commencement rolling retired Gemini 3.5 Flash-Lite successful Search today. It offers stronger instruction pursuing and amended understands personification intent, truthful conversations travel overmuch much seamlessly. https://t.co/DFnL6YPVVP
Hey Barry, cheque retired the tweet from @rmstein here: https://t.co/CGHPZvi5Qn
— Rajan Patel (@rajanpatel) July 21, 2026Forum chat astatine X.
