Google's I/O announcements: new models, a cloud agent that never sleeps, and a redesigned Gemini app

Details

Google's I/O 2026 was anchored by Gemini 3.5 Flash. CEO Sundar Pichai cited Artificial Analysis data showing it as the only model in the upper-right quadrant of intelligence versus speed. At Antigravity-optimized inference, Google claims 12x faster throughput than the baseline. Costs are reported at roughly a third to half of comparable frontier models. Pichai's enterprise pitch: companies shifting 80% of their workloads to a 3.5 Flash/Pro mix could save over $1 billion annually. Gemini 3.5 Pro is expected next month. Gemini Omni is a Gemini-architecture model trained multimodal from scratch — not a standalone video generator like Veo, but a unified model that takes any input and produces any output. Video ships first; it can use its own generated output as input for iterative editing, and users can swap characters or change video style. Omni Flash launches Tuesday across Google AI+, Pro, and Ultra subscriber tiers. Gemini Spark is the 24/7 cloud personal agent vision made concrete: a continuously running assistant that works without user interaction. The Gemini app receives a significant visual redesign. For developers, Antigravity gains multi-agent management tools and SynthID watermarking is expanded to more surfaces.