Is Gemini 3.5 Flash Really That Powerful? Google Turns Its Search Box into an AI Agent

At Google I/O the company announced a strategic move toward "24‑hour nonstop AI agents," noting that monthly token consumption grew from 9.7 trillion two years ago to 480 trillion last year and now exceeds 3.2 quadrillion—a seven‑fold increase.

Google reports more than 8.5 million developers using Gemini to build new applications, and 13 of its products have over 1 billion users, with five exceeding 3 billion. The AI‑driven Search experience already serves over 1 billion monthly active users, and the Nano Banana image model has generated over 50 billion pictures.

The keynote introduced the Gemini 3.5 series, highlighting the lightweight Gemini 3.5 Flash as the most advanced agent‑and‑programming model to date. Google claims it combines frontier intelligence with the series’ characteristic speed.

Benchmark results show Gemini 3.5 Flash achieving 76.2% on Terminal‑Bench 2.1, 1656 Elo on GDPval‑AA, and 83.6% on MCP Atlas, all surpassing Gemini 3.1 Pro. On the CharXiv Reasoning test it scores 84.2% and generates output tokens four times faster than competing models.

The model’s cost is reported to be less than half that of other frontier models, and Google states that users no longer need to trade off quality for latency.

Integrated with the updated Antigravity execution framework, Gemini 3.5 Flash can orchestrate collaborative sub‑agents. Demonstrations include synthesizing the AlphaZero paper and creating a fully playable game within six hours, as well as generating new city‑scape visuals.

Google also emphasized safety improvements: the Frontier Safety Framework reduces the likelihood of harmful content generation and lowers false safety rejections.

A new AI agent called Spark, built on Gemini 3.5, is deeply integrated with Gmail, Docs, and Slides, allowing users to automate tasks, set recurring workflows, and generate reports with minimal prompting.

The Search box received its biggest overhaul in 25 years, embedding Gemini 3.5 Flash into the AI Mode. The upgraded box supports multimodal input (text, image, video, files, Chrome tabs) and can be assigned continuous monitoring tasks, such as scanning news sites for specific criteria.

Google unveiled Gemini Omni Flash, the first model in the Gemini Omni series, capable of generating high‑quality video from any combination of image, audio, video, and text inputs, and allowing interactive video editing via dialogue.

Gemini 3.5 Flash is now globally available through the Gemini App, Google Search AI Mode, the Antigravity platform, Google AI Studio, and Android Studio. Gemini Omni Flash is being rolled out to Google AI Plus/Pro/Ultra subscribers, the Gemini App, Google Flow, YouTube Shorts, and YouTube Create, with an API release planned for developers and enterprises.

As DeepMind CEO Demis Hassabis remarked, “When we look back on this period, we’ll realize we were standing at the foot of the singularity.”