Google I/O: Agent-Optimized Gemini 3.5 Flash Model Released Google used its I/O 2026 keynote to shift attention from its delayed flagship AI model to a smaller, faster system it says could power a new era of autonomous agents.

Timeline: From anticipation to a strategic pivot

In the run-up to I/O, expectations centered on Gemini 3.5 Pro, billed as Google’s next “most powerful” AI model. Instead, CEO Sundar Pichai told the packed audience that Gemini 3.5 Pro “isn’t coming out yet,” asking developers to “give us until next month,” a delay that “drew groans” from attendees.

Attention quickly turned to what Google did launch on May 19: Gemini 3.5 Flash. Ars Technica described it as a highly efficient model that “supposedly offers frontier-level intelligence” while outputting tokens nearly four times faster than prior Pro models, potentially saving heavy enterprise users “a billion dollars per year” in compute costs.

Google simultaneously unveiled a redesigned Gemini app built around a new “Neural Expressive” visual language and announced that Gemini 3.5 Flash would become the default model in Gemini and AI Mode, with the company calling it its “strongest” system yet for coding and long-horizon agentic tasks.

Agents, not chatbots

TechCrunch framed the release as a strategic turn “on agents, not chatbots,” reporting that Gemini 3.5 Flash can independently manage research projects and, in internal tests, “build an operating system entirely from scratch.” DeepMind’s Koray Kavukcuoglu said the model “outperforms our latest frontier model, 3.1 Pro, on nearly all the benchmarks,” and that an optimized version runs up to 12x faster for agent workloads.

Developer advocate Addy Osmani promoted Flash as “fast, great for building rich UIs + agents and… stronger from coding to long-horizon tasks and multi-step workflows.” Pichai emphasized that Flash is already “available today for everyone in @antigravity and across our products and APIs,” claiming it beats 3.1 Pro “across almost all benchmarks with huge progress in coding.”

DeepMind CEO Demis Hassabis called Gemini 3.5 Flash “amazing,” citing performance “better than 3.1 Pro on coding & agentic tasks” and speeds of “800 tokens/sec… often at less than half the cost,” while stressing that “Pro [is] to come.”

Competing views on performance

Outside Google, the reaction was more measured. A widely shared CursorBench leaderboard placed Gemini 3.5 Flash 10th with a 49.8% score, below several rival models; Elon Musk amplified that comparison with a terse recommendation to “Try Composer 2.5,” highlighting a competitor that scored 63.2% at far lower cost.

Taken together, I/O 2026 presented Gemini 3.5 Flash as Google’s near-term workhorse: a cheaper, faster model designed to live inside always-on agents like the new Gemini Spark and the Antigravity development environment, while the more powerful Gemini 3.5 Pro remains on the horizon.

Continue reading https://foxvector.com