Google kicked off I/O 2026 on May 19 — and this time it wasn’t an incremental update fest. The announcements had real substance.
Gemini 3.5 Flash: Frontier intelligence at Flash prices
The new flagship is called Gemini 3.5 Flash, and it’s a statement. Google says it’s 4x faster than comparable frontier models in output tokens per second — while beating Gemini 3.1 Pro on coding, agentic, and multimodal benchmarks. That’s remarkable because Flash models are usually the budget compromise. Here, the fast variant becomes the best one.
Gemini 3.5 Flash is rolling out now in the Gemini app, Search, Antigravity 2.0, and the API. Gemini 3.5 Pro — the heavier model — is already in use internally but won’t be available to everyone until next month.
Gemini Spark: The agent that runs on its own
Spark is Google’s new proactive AI agent. It can reason across connected apps and handle tasks in the background — under your supervision, but without you needing to approve every step. The beta starts next week for trusted testers and AI Ultra subscribers.
This is Google’s answer to Anthropic’s Managed Agents and Dreaming. The approach is similar: agents that automate workflows and operate independently. The real question is how well it works in practice.
Omni: Google’s world model
Perhaps the most ambitious announcement: Omni is a new model that combines Gemini’s intelligence with generative media models. It understands the world multimodally — text, image, video — and can generate and edit cinematic video. Omni is rolling out now for Gemini Plus, Pro, and Ultra subscribers.
What this means for Claude users
Google showed they’re serious about the agent space. Between Spark, Antigravity 2.0, and the new models, the competition for best AI platform just got a lot more intense. For Anthropic, this means the lead they’ve built with Managed Agents and Cowork isn’t safe. Google has the infrastructure, the user base, and now the models to compete for real.
Sources: