Google just unloaded an entire year’s worth of announcements at I/O 2026: Gemini 3.5 Flash (half to one-third the price of comparable frontier models), Gemini 3.5 Pro (coming next month), Gemini Spark (general-purpose AI agent, now in beta), Omni (world model for simulating physical environments), and a $100/month AI Ultra subscription tier. Oh, and AI Mode just hit 1 billion monthly active users. The agentic Gemini era isn’t coming — it’s here.
🔍 THE BOTTOM LINE
Google is betting everything on agents. Gemini 3.5 is the cheap engine, Spark is the hands, Omni is the eyes, and AI Ultra is the business model. The web giant isn’t just adding AI to search — it’s replacing the way you interact with information entirely.
Gemini 3.5 Flash: The price war escalates
Gemini 3.5 Flash is now the default model in Search and the Gemini app, and it costs half to one-third of comparable frontier models. This isn’t a mid-tier model being positioned as flagship — Google is claiming frontier-level performance at commodity pricing. Deep Think mode pushes performance further for harder tasks.
For developers, this is the continuation of the great AI price collapse. Every time Google or OpenAI cuts prices, the floor drops for everyone. If you’re building on AI APIs, your margins just got better — or your competitors’ did.
Gemini Spark: The agent that acts
Gemini Spark is Google’s general-purpose AI agent, now in beta. It doesn’t just chat — it takes action across apps. Book a restaurant, manage your calendar, buy tickets, coordinate across services. This is Google’s answer to OpenAI’s operator and Anthropic’s computer use — but with a crucial advantage: Google owns the ecosystem.
Spark works across Gmail, Maps, Docs, YouTube, Calendar, and the broader web. That’s not a plugin architecture — that’s native integration with services that collectively have billions of users. If agents are the next computing paradigm, Google’s distribution advantage is staggering.
Omni: The world model
Omni is perhaps the most forward-looking announcement: a world model that simulates physical environments. Applications span video editing, robotics, and gaming — but the implication is bigger. A world model that understands physics could be the missing piece for truly capable embodied AI. If Spark is the hands, Omni is the eyes and the physics engine.
AI Mode: 1 billion users and the web’s existential question
AI Mode in Search has surpassed 1 billion monthly active users in its first year. Google is processing 3.2 quadrillion tokens per month — 7× growth from last year’s 480 trillion. AI Overviews alone has 2.5 billion monthly active users.
This is the scale that makes critics nervous. As one viral blog post titled “On Google Declaring War on the Web” argues, AI Mode decontextualizes information and removes links to sources — making the open web into “unpaid raw material for their synthetic text extruders.” When 2.5 billion people get answers without ever leaving Google, the websites that created those answers stop getting traffic. This is the web’s quiet crisis.
AI Ultra: $100/month for the serious stuff
Google’s new AI Ultra tier costs $100/month and presumably bundles premium access to Gemini 3.5 Pro, Spark, and advanced features. This puts it in the same ballpark as OpenAI’s premium tier and signals that the era of “AI is free because it’s subsidised” is ending — at least for power users.
The infrastructure behind it all
Google expects $180-190 billion in capex this year — six times the $31 billion it spent in 2022. Custom TPUs, massive data centres, and a full-stack approach from silicon to services. Nobody else has this depth of infrastructure, and Google is betting it translates into unbeatable cost and performance advantages.
❓ Frequently Asked Questions
Q: What does this mean for NZ? NZ businesses and developers get cheaper, more capable AI tools immediately. But the NZ angle that matters is governance: Google’s AI Mode is reshaping how information flows, and NZ has no AI regulation to ensure local content creators are treated fairly. The hidden costs of AI adoption that RNZ reported this week — licensing, security, vendor lock-in — apply directly here.
Q: Is Gemini 3.5 Flash actually as good as they claim? Independent benchmarks are still coming, but Google’s claim of frontier performance at half price is bold. The proof will be in developer adoption and real-world performance. Given Google’s track record of competitive benchmarking, expect the reality to be “very good, maybe not quite frontier, but stupidly cheap.”
Q: What should I do? If you’re a developer, start testing Gemini 3.5 Flash against your current stack — the price difference alone justifies the evaluation. If you’re a content creator, pay attention to AI Mode’s impact on your traffic. If you’re a business, factor AI Ultra pricing into your 2026 budget.
🔍 THE BOTTOM LINE
Google didn’t just announce products at I/O 2026 — it declared the shape of the agentic era. Cheap models, acting agents, world simulation, and premium subscriptions. The question isn’t whether this future arrives. It’s whether the open web survives it.
