Google Unveils Gemini 3.5 Flash and Spark Agent at I/O 2026

Google used its I/O 2026 keynote on Monday to drop the most aggressive Gemini update yet: a new flagship Flash-tier model that outperforms last year's Gemini 3.1 Pro on coding, and a personal AI agent named Spark that runs around the clock on virtual machines in Google Cloud.

The pricing is the headline. Gemini 3.5 Flash launched at $1.50 per million input tokens and $9.00 per million output tokens — a fraction of what frontier models from OpenAI and Anthropic cost. Google says output speed is roughly 4x faster than comparable models.

What's New in Gemini 3.5 Flash

Flash-tier models used to be the budget option. Gemini 3.5 Flash flips that framing. Google says the model beats Gemini 3.1 Pro on coding and agentic benchmarks while costing a fraction of frontier-class pricing.

As of May 19, 2026, Gemini 3.5 Flash is the default model in the consumer Gemini app and in AI Mode inside Google Search worldwide. That means hundreds of millions of users are now getting a more capable model on every query, with no action required on their part.

Why this matters: The Flash-tier upgrade pulls the cost curve down for any developer building AI features at scale. At $1.50 input and $9 output, you can run a chatbot, summarizer, or agent loop for cents on the dollar relative to top-tier Anthropic or OpenAI models.

Spark: A Personal Agent That Runs 24/7

Gemini Spark is Google's answer to the "agentic AI" wave that has dominated 2026 product launches. Spark is a personal agent inside the Gemini app that can reason across information from connected apps — Gmail, Calendar, Drive, Maps, third-party connectors — and take actions on your behalf.

The technical twist: Spark runs on virtual machines in Google Cloud, not on your phone. That means it keeps working 24/7, even when your device is closed. It can monitor inboxes, draft replies, book appointments, and queue up multi-step research tasks in the background.

Spark is powered by Gemini 3.5 and is in beta. Google says it will roll out first to trusted testers and Google AI Ultra subscribers starting next week.

Google AI Ultra: A New $100/Month Tier

Google introduced Google AI Ultra, a new $100-per-month subscription tier aimed at developers, creators, and power users. The tier includes early access to Spark, 20TB of cloud storage, and priority access to new model releases.

The pricing slots Ultra between ChatGPT Plus ($20) and ChatGPT Pro ($200) and competes directly with Anthropic's Claude Max plan. Google is betting that bundling consumer storage with a high-end AI agent will draw users who already live inside Drive, Gmail, and Photos.

Gemini Omni: World-Modeling for Video

Google also unveiled Gemini Omni, a model built specifically for cinematic video generation and editing. Omni takes text prompts, reference images, and existing video clips as input and produces edited footage.

Google framed Omni as a "world model" — meaning it tries to keep physics, object permanence, and scene continuity consistent across frames. That's the bar OpenAI's Sora and Runway's Gen-4 have been pushing against for a year.

Industry Implications

For OpenAI, the Flash-tier price drop is the more uncomfortable announcement. GPT-5.4 and GPT-5.5 remain stronger on the hardest benchmarks, but Google just made it cheaper to serve good-enough AI at consumer scale than to route everything through OpenAI's API.

For Anthropic, Spark is the more direct shot. Anthropic has been winning the agent-coding race with Claude Opus 4.7 and the Claude Agent SDK, but Google's distribution advantage — Spark inside the Gemini app, which lives on billions of Android devices — could blunt that lead with mainstream users.

For developers, the cost curve matters most. Coding tools, customer-support agents, and document-processing pipelines that have been routing to Anthropic or OpenAI for quality will now have a real budget alternative.

Expert Perspectives

Reaction on X focused on two themes. Developers flagged the Flash-tier pricing as the most consequential part of the keynote — calling it a "race to the bottom on token costs that we benefit from." Researchers focused on Spark's persistent agent architecture, arguing that the always-on model is the right design for agents that need to monitor and react over hours and days.

Sundar Pichai used the keynote to argue that Google's full-stack approach — owning the model, the cloud, the consumer apps, and Android — is the moat. The pitch is that no other AI lab can ship an agent that talks to Gmail, Maps, Photos, and Calendar without permission walls.

What's Next

Gemini 3.5 Flash is rolling out worldwide now as the default model in the Gemini app and AI Mode in Search. The API is live on Vertex AI and Google AI Studio at the announced pricing.

Spark beta access opens to Google AI Ultra subscribers next week. Google has not given a public timeline for general availability, though the rollout pattern suggests a broader release later this year.

Gemini Omni's launch window is "coming weeks" with no firm date.

Bottom Line

Google I/O 2026 is the clearest sign yet that the AI race has moved past raw frontier-model benchmarks into distribution and price. Gemini 3.5 Flash makes good-enough AI radically cheaper, and Spark turns the Gemini app into an always-on agent that competes with Claude and ChatGPT on a different axis. If you build with AI APIs, your cost model just changed — and if you live inside Google's apps, your assistant is about to get a lot more proactive.