Google I/O 2026 Recap: Gemini 3.5, Omni, Spark, and the New Search
Google I/O 2026 Is Over โ Here's What Actually Matters
Google I/O 2026 closed with over 140 announcements across a keynote that ran longer than 3 hours. The short version: Google is turning AI from a helper tool into the background infrastructure for every product it ships.
Four developments stood out.
Gemini 3.5 Flash โ New Model, New Speed
Gemini 3.5 Flash launched during the keynote and became the default model on AI Mode globally. 4x faster than other frontier models, at less than half the cost. Google is processing 3.2 quadrillion tokens per month โ up 7x year over year.
Flash isn't just a speed upgrade. Google deepened its agentic capabilities โ the model can chain multiple steps autonomously instead of answering a single question. Building AI-powered features? This is the model to benchmark against.
Gemini 3.5 Pro arrives in June. Flash is available today in the Gemini app and AI Mode.
Gemini Omni โ Video Generation From Any Input
Gemini Omni takes text, images, audio, and video as input and produces video output. This isn't simple text-to-video โ Omni accepts any combination of inputs and renders cohesive video.
On stage, Google demoed uploading a personal video and saying "change the environment to a rainforest" or "switch to a bird's-eye view," and Omni handled it. Character images, alternate scenes, and sketches can all be fed in as reference material.
Personalized voice avatars are part of Omni too โ generate video with your own voice without a studio setup.
Omni Flash is available today in YouTube Shorts Remix and YouTube Create. Remix Shorts by inserting yourself into a creator's video or changing the aesthetic. All outputs carry SynthID watermarks. Creators 18+ can opt out of remixing.
Gemini Spark โ A Personal Agent That Runs 24/7
Spark is the next step in the AI agent race. Unlike typical chatbots, Spark runs on the cloud and connects directly to Gmail, Docs, Sheets, Calendar, and third-party apps like Canva, OpenTable, and Instacart.
Example: ask Spark to track your monthly credit card bills. It automatically scans email, finds invoices, summarizes them, and sends a digest. Ask "what did my kid learn at school this week?" โ it pulls info from school emails and responds.
Spark runs 24/7 in the background โ even when your machine is off. It waits for confirmation before taking sensitive actions like sending emails or making purchases.
Rolling out: trusted testers this week, Ultra subscribers next week, Workspace business customers this summer.
Spark connects to Sheets, Gmail, and third-party apps through a new protocol. Building integrations? This is a platform worth watching.
Google Search โ The Biggest Redesign in 25 Years
Google's Search box โ the one billions of people open every day โ has been completely rebuilt. Google calls it the biggest upgrade in 25 years.
The new Search box works like a mini AI agent. Beyond autocomplete, it suggests follow-up questions, supports multi-modal input (text, images, video, Chrome tabs), and "thinks" before you finish typing.
AI Agents in Search is new. "Information agents" run in the background monitoring blogs, news, prices, and real estate โ summarizing results and suggesting next steps. Rolling out this summer for Pro and Ultra subscribers.
Agentic booking is also notable: search for "private karaoke room, late-night food delivery, Saturday night," and Search checks availability, shows pricing, and completes the booking. Google can even call local businesses โ technology developed since Google Duplex in 2018.
Google Antigravity integrates into Search โ generating custom UIs, visualizations, and mini apps directly in search results. Example: search for wedding planning, and Search builds a checklist dashboard. Free for everyone.
What Else?
Beyond the four main themes, Google also announced:
- New AI pricing: Ultra tier drops from $250 to $100/month, Premium at $200/month with Project Genie
- Personal Intelligence goes free: supports 98 languages across nearly 200 countries
- Google Pics: new image generation/editing app in Workspace
- Gmail Live, Docs Live, Keep Live: search and create content with voice
- SynthID watermarks: verify AI-generated content across Gemini, Search, Chrome, Pixel
What Should Developers Do?
Building AI features: Benchmark Gemini 3.5 Flash โ 4x faster at half the cost. Watch the Gemini Omni API if your product involves video or content creation. Study the Spark protocol โ it could become the standard for AI agent integrations.
Building search/ecommerce: Agentic booking and Universal Cart will change how users shop online. Google Antigravity enables custom UI in search results โ worth exploring.
Building mobile apps: Gemini Intelligence is deeply integrated into Android 17, Wear OS, and Android Auto. On-device AI with Gemini 3.5 Flash is the new direction.
Conclusion
Google I/O 2026 isn't about Android anymore. Android got pushed to a separate "Android Show" a week earlier. I/O this year was about AI โ Gemini 3.5, Omni, Spark, and a completely reimagined Search.
Whether developers adopt any of this is a separate question. What's clear: Google is betting that the future is "AI running in the background for you," not "AI helping you."
This post is part 23 of the Tin Tแปฉc Cรดng Nghแป series.
This post was assisted by AI (Amy ๐ธ). Content reviewed by the author.