Google's I/O 2026 developer conference has finally wrapped, and the keynote was, truthfully, a bit of a snooze for non-developers. The headline Gemini news amounted to a half-step update in Gemini 3.5 Flash, with the word "agentic" getting a full workout.

Google is "building a new agentic era," as one executive put it on the stage. The big theme this year: let Gemini do the heavy lifting. Emails, dinner reservations, vacation planning, shopping — Gemini wants to be your everything solution. The twist is Google now wants it done autonomously, within limits, naturally.

Hardware was equally quiet. The most notable announcement was a pair of smart glasses the company is calling "audio glasses" — essentially Google's answer to Meta Ray-Bans, but leaning harder into style with collaborations from Warby Parker and Gentle Monster. The goal seems to be making camera glasses that don't scream, "I am wearing camera glasses."

All told, it was a characteristically uneventful show — which, let's be honest, is basically the I/O keynote tradition at this point. But there was still plenty announced worth digging into, so here's the full rundown.

Mashable 101 Fan Fave: Vote for your favorite creator today!

Gemini 3.5 Flash is the new default model

Google launched the first model in the new Gemini 3.5 line today, and it's available now in the Gemini app and other Google AI products. Google execs said that a new flagship model, Gemini 3.5 Pro, would be coming in June.

Gemini 3.5 Flash beats most frontier models in benchmarks and token efficiency, according to Google. The company says it also surpasses Gemini 3.1 Pro in most respects. You can start using it right away at gemini.google.com.

Gemini Spark wants to run your life

The splashiest announcement of the keynote was Gemini Spark, a cloud-based AI agent that runs continuously in the background, handling tasks while you're busy doing everything else. As I wrote during its announcement, it's possibly the most ambitious thing Google has put on stage in a while. At least, the most ambitious of anything else announced at I/O today.

Powered by Gemini 3.5 Flash and built with Google's Antigravity coding IDE, Spark connects to Gmail, Docs, and eventually over 30 third-party apps, including Uber, OpenTable, Lyft, and Zillow. It rolls out to AI Ultra subscribers in the U.S. next week.

To stop it from going rogue with your credit card, Google talked about Agent Payments Protocol (AP2), which caps what Spark can spend, where it can shop, and what it can buy. For now, you still approve every transaction. The company described Spark as a teenager getting their first debit card and stated that over time, the agent will get looser guardrails as trust is built.

Google's AI subscription tiers got cheaper

Google reshuffled its AI subscription lineup at I/O, and the headline is that things got more affordable. As I reported earlier, the new entry point for AI Ultra is $99.99/month — down from $250 — aimed at developers and power users, bundling Gemini 3.5 Flash, 5x the usage limits of Pro, priority access to Google Antigravity, 20TB of storage, and a full YouTube Premium plan. The $250 tier still exists but drops to $200. The full lineup now sits at AI Plus ($7.99), AI Pro ($19.99), and AI Ultra (starting at $99.99).

Google is also ditching the per-prompt counting model in favor of measuring compute used, meaning a simple text query barely dents your limit while a complex video task costs more. Limits also now refresh every five hours instead of daily, and if you hit your cap, Google automatically steps you down to a lighter model rather than cutting you off entirely.

Gmail is getting a live voice mode

Gmail Live, a new feature reported by Mashable's Haley Henschel, lets you verbally ask your inbox questions instead of typing searches. The pitch is pretty straightforward — ask what your flight's gate number is or what's happening at your kid's school this week, and Gmail pulls the answer from your emails. Similar conversational features are also coming to Google Docs and Keep. Gmail Live rolls out this summer for AI Pro and Ultra subscribers, with a Workspace preview at the same time.

AI Inbox, which started as an Ultra-only feature earlier this year, is getting three new additions and broader access for Pro and Plus subscribers. The updates include personalized draft replies, instant access to relevant Docs and Sheets, and one-click task management to clear out inbox clutter. Those features start rolling out today. Gmail VP of product Blake Barnes said in a media pre-brief that user data isn't used for training on either feature, and that sourcing is being built in so you can see exactly which emails informed a given response.

Google Search gets its biggest overhaul in 25 years

Search — the thing that made Google Google — is getting another serious injection of AI. The company is throwing a lot at its flagship product this year, and most of it falls under the same "let Gemini handle it" umbrella that dominated the keynote.

The most symbolically significant update might be the new intelligent AI Search Box, which Google is billing as the first redesign of its search box in over 25 years. The expanded box now supports natural language queries and lets you attach images, videos, files, and even Chrome tabs alongside your search. Basically, Google wants you to stop Googling and start... talking to Google.

AI Overviews is also getting a back-and-forth conversation mode. Users can now ask follow-up questions directly within the Overview, turning what was a static summary into more of a chatbot exchange. It's a logical evolution, even if it continues to beg the question of what happens to the websites that used to get that traffic.

Are you an Apple superfan? Enter Mashable’s Big Guessing Game to win prizes.

Search Agents

The bigger swing: Google introduced Search agents, which are essentially AI assistants that run 24/7 in the background, scanning news sites, blogs, and social media on your behalf. Google's example use cases — tracking apartment listings, monitoring sneaker drops — are genuinely useful, though the catch is that these information agents are locked to AI Pro and Ultra subscribers when they launch this summer.

The agentic push extends to shopping and booking too, where agents can surface live prices, availability, and direct booking links. Google is also letting agents make actual phone calls to businesses on your behalf — a feature rolling out to all users this summer.

Google is broadening access to Personal Intelligence in Search, its feature that connects your Gmail, Google Photos, and soon Google Calendar, so Search can factor in your personal context. It's opt-in, and Google is quick to note users maintain control over their data, which is the kind of thing you say when you know people are going to ask.

Rounding things out, Google is folding AI coding tools powered by Google Antigravity directly into Search, with a clear push toward non-developers. Think custom fitness trackers or wedding planning dashboards — Gemini for people who've never touched a line of code.

Google and Samsung's smart glasses are real, just unnamed

The hardware story of the show was the first real look at Android XR smart glasses. Google and Samsung pulled back the curtain on two styles — one in collaboration with Gentle Monster, one with Warby Parker — both arriving this fall. There's still no name and no price, but the design is clearly meant to avoid the "obviously tech glasses" look that has plagued the category.

Google is deliberately calling these "audio glasses" to distinguish them from future display glasses. Functionally, they're in the same territory as Meta's Ray-Bans — voice commands, phone pairing, hands-free assistance — but with deep Gemini integration baked in. Samsung handled the hardware, and Google handles the AI and Android XR platform.

Google Shopping gets an overhaul

Shopping reporter Samantha Mangino broke down three new Google Shopping features announced at I/O, and the headliner is Universal Cart — a single cart that aggregates everything you've added across retailers like Target and Amazon into one Google-side view, working in the background to flag price drops and restock alerts. It hits Google and Gemini in the US this summer, with Gmail and YouTube support to follow.

The bigger structural move is Universal Commerce Protocol (UCP), which lets AI agents complete purchases and hotel bookings directly through partners, including Amazon, Walmart, Shopify, and Meta. It also surfaces relevant fees and card rewards automatically so you know what you're actually paying before you check out.

Docs Live

Google announced Docs Live, a new feature that lets you dump your word vomit and thoughts into Gemini and have it turn them into a structured document in real time. From there, you can refine it conversationally — just tell Gemini what to fix, add, or cut, and it updates on the spot.

SynthID gets a big lift

One of the bigger under-the-radar moments of the keynote: Google's AI digital watermarking tool SynthID is being adopted by OpenAI, Kakao, and ElevenLabs. The Google project is now becoming an industry standard in fighting against AI-generated content, and it's also a rare moment of cross-company alignment in the AI space.

In addition, SynthID will also be easily accessible in Chrome and Google Search. As Google CEO Sundar Pichai said in the keynote, that means users will be able to right-click and quickly see if an image or video contains a SynthID, and thus is likely AI-generated.

Gemini Omni is Google's new world model

Google unveiled Gemini Omni at I/O, its new multimodal world model. While Google is positioning it as a model that can "create anything from any output," the keynote demo leaned heavily into video generation. The first release in the family, Gemini Omni Flash, is available today for paid AI Plus, Pro, and Ultra subscribers in the Gemini app and Google Flow, with a free rollout to YouTube Shorts and YouTube Create later this week.

Unlike text-to-video tools like Veo, Omni is multimodal in both directions — you can feed it text, audio, images, or video, and it generates back accordingly. Google DeepMind CEO Demis Hassabis called it a "meaningful step" toward AGI at the keynote. The practical showcase was mostly about video editing through conversation, like swapping backgrounds, changing angles, and adjusting specific details in a clip. There's also an Avatar feature for creating a digital likeness of yourself, though Google says it's still being tested before a broader rollout. All Omni-generated videos get embedded with Google's SynthID watermark to identify them as AI-generated.

YouTube gets two updates, neither of them huge

YouTube got a relatively light showing at I/O compared to the rest of Google's product slate — but there were two things worth flagging.

Gemini Omni is coming to YouTube Shorts Remix, the platform's AI creation tool that generates videos from existing content. Creators will be able to use more advanced AI prompts to remix their Shorts, with AI-generated content labels and links back to the original source automatically applied. Google also expanded its likeness detection tool — which flags content where a creator's face has been AI-altered — to all creators 18 and older.

The second update isn't on YouTube itself but inside Google Search. Ask YouTube lets users surface relevant YouTube videos directly within search results when asking complex questions. Think tutorial-style queries — how to fix something, how to learn something — where a video would actually be more useful than a text answer. It's still in testing but expected to roll out across the US this summer.