Google I/O 2026 Preview: Gemini 4, Android 17, and an AI-First Hardware Blitz

With just four days until the keynote, Google is holding its cards close. But the leaks, regulatory filings, and developer briefings that have trickled out over the past month paint a remarkably clear picture of what the company intends to announce at its annual developer conference — Google I/O 2026, on May 19–20 at Shoreline Amphitheatre in Mountain View, California.

This year’s event arrives at a pivotal moment. OpenAI’s GPT-5.5 has captured headlines and mindshare. Apple is preparing a major AI overhaul for iOS 20, to be unveiled at WWDC next month. And Microsoft quietly restructured its OpenAI partnership to go multi-cloud. Google’s I/O keynote — beginning at 10 AM Pacific on May 19 — is its highest-stakes developer event in years.

Gemini 4.0: The Main Attraction

Every indicator points to Gemini 4.0 as the centerpiece of the keynote. The model has been internally codenamed “Omni” and is expected to unify text, image, audio, and video generation into a single pipeline — a capability Google has been publicly teasing through its Project Astra demos since 2024.

According to Android Authority, Gemini 4.0 is expected to push context windows to 2–4 million tokens, dwarfing even the already impressive 2-million-token limit of Gemini 3.1 Ultra, which launched in March. Critically, Google is promising that these context windows will function natively across modalities — meaning a developer can feed a full-length film alongside a screenplay and ask the model to cross-reference them, without stitching together separate pipelines for each input type.

On benchmarks, analysts tracking early SWE-bench runs suggest Gemini 4.0 will challenge GPT-5.5’s 72.3% on agentic coding tasks. Google has also previewed plans for a new “Agentic Foundations” layer in its Vertex AI platform that lets enterprise developers wire Gemini 4.0 into their workflows, with multi-step tool use, persistent memory, and auditable action logs baked in.

The leakage of a “Gemini Omni” tab spotted inside the Gemini web app earlier this month suggests video generation capabilities are nearly production-ready, likely launching as a limited waitlisted preview at the conference.

Gemini Intelligence: AI That Lives Inside Android

Beyond the model itself, Google is set to announce Gemini Intelligence — a framework that embeds Gemini’s reasoning directly into the Android operating system at the system level. This is Google’s most aggressive answer to Apple Intelligence: rather than building app-specific integrations, Gemini Intelligence monitors on-screen context across apps and can proactively complete cross-app tasks without the user switching between them.

Internal previews have shown capabilities like simultaneously summarizing all emails related to a travel booking, pulling up the relevant calendar entries, and flagging a gate-change notification from the airline — all surfaced in a single coherent interaction. The system reportedly uses a fine-tuned on-device model for privacy-sensitive operations, routing to the full cloud Gemini only for complex multi-step reasoning.

CNBC reported this week that Google is racing to deploy these features before Apple Intelligence arrives in iOS 20. Sources described the internal pressure as “existential” for Google’s mobile AI positioning, given that Apple’s tight hardware-software integration allows for on-device capabilities that cloud-first companies struggle to replicate.

Android 17: Security, Satellite, and Find Hub

Android 17 is expected to receive a formal release timeline at I/O, targeting stable rollouts to flagship devices by September 2026. Google previewed several features at the Android Show on May 12:

Find Hub expansion — the successor to Find My Device now supports ultra-wideband precision location for third-party tags, and adds satellite connectivity for emergency location sharing even when no cellular signal is available. This puts Android in direct competition with Apple’s emergency SOS satellite feature, which has been available since 2022.

Hardened permissions — new per-session app permissions that automatically revoke access after a configurable time window, plus tighter controls over clipboard access and background sensor use. The changes respond to mounting pressure from EU regulators over app data collection practices.

Material 3 Expressive — a redesign update enabling richer motion, adaptive color theming, and dynamic layouts that respond intelligently to screen size and orientation across the entire app ecosystem.

Android Auto is also getting significant upgrades, including Dolby Atmos passthrough, a new widget framework, and native support for video streaming apps — meaning rear-seat passengers in compatible vehicles will be able to stream content through the car’s infotainment system rather than casting from a phone.

Android XR: Smart Glasses That Actually See

Perhaps the most surprising confirmed announcement is a public developer preview of Android XR glasses — lightweight spectacles that run Gemini natively and overlay contextual information directly in the user’s field of view. Google confirmed the preview in a developer relations briefing, framing the glasses as a complement to its Android XR headset platform.

Unlike Meta’s Ray-Ban smart glasses, which rely on cameras and speakers with no display, Google’s prototype reportedly includes a small embedded projector enabling a true augmented reality overlay. Developer kits are expected to ship to a limited set of early-access partners at I/O, with broader availability in early 2027.

The timing is deliberate. Meta is expected to refresh its Ray-Ban lineup later this year, and Microsoft is rumored to be working on a next-generation HoloLens built around Copilot. Google is betting that Gemini’s reasoning capability will differentiate its glasses as genuinely useful rather than a novelty.

The Stakes for Google

Google enters this I/O under genuine competitive pressure. OpenAI’s model cadence has accelerated dramatically — GPT-5.4 and GPT-5.5 launched within six weeks of each other. Anthropic’s Claude continues to win enterprise deployments. And Apple’s AI overhaul, while delayed, is widely expected to leverage device-level integration in ways that cloud-first companies like Google will struggle to match.

The I/O keynote is not just a product demonstration. For Google, it is an attempt to reframe the narrative — from “the search company catching up on AI” to “the AI company that runs the world’s most widely used computing platform.” With 3 billion Android users and Gemini now embedded across Search, Gmail, Docs, Chrome, and the operating system itself, Google has the distribution advantage that OpenAI and Anthropic fundamentally lack.

May 19 will reveal whether it also has the product to match.