Gemini for Science is a new collection of science tools and experiments to expand the scale and precision of scientific exploration.
Introducing Gemini Omni, which allows you to create anything from any input and edit naturally using conversational language.
We're expanding our tools to help you understand how content was created and edited across the web.
Today at I/O, we announced new ways to vibe design with Stitch, transforming your design experience into a live, collaborative partnership with the Stitch Agent — allowi…
At Google I/O 2026, we introduced an intelligent cart that corrals everything you're shopping, expansions to UCP, tech to power agents that buy for you and more.
Android Halo brings intelligence from your agent right to your status bar.
At Google I/O we released Gemini 3.5, our latest series of models combining frontier intelligence with action.
By Simon Cooke, Developer Relations Engineer (X) and Mayank Jain, Product Manager (X)
Android Performance Analyzer (APA) is Android’s new profiler and performance analysis tool for the Android mobile ecosystem.
APA is intended as a profiling tool for any developer building for Android who needs to make their app or game run better and faster. It is helpful for all performance-minded engineers, especially those using Vulkan in their game engines who want to squeeze every bit of performance out of their code.
APA aims to be the tool that helps you optimize apps and games for all modern Android devices and simplifies your most common workflows, with a simple interface that anyone on your team can quickly learn and be productive.
Available today in open beta is APA’s new System Profiler that you can use to analyze the CPU, GPU, Memory, and power usage of your app or game - and see how it interacts with system behavior.
Developed in collaboration with Samsung Austin Research Center (SARC) and LunarG, APA relies on Perfetto for system tracing and its upcoming frame profiling/debugging features (stay tuned!) are powered by LunarG’s GFXReconstruct technology for graphics capture and replay.
Devices running Android 12+ will provide the best experience for capturing system-wide performance and GPU counters and render stages.
We’re also working across the Android ecosystem with our esteemed industry partners to bring more profiling & optimization related data into APA.
APA ships in two different forms, and you can download whichever one suits your needs best
The standalone desktop app is intended to be used without an Android Studio project or Gradle build - and provides deep customization of recording configuration, built-in Vulkan layers for graphics analysis, deep inspection of GPU counters and much more.
APA is also cross-platform: works natively on Windows, MacOS, and Linux.
Capturing your profile data
You don’t always want to take a capture immediately at application or game launch. APA allows you to choose, and capture traces from your device at launch or triggered manually. The user interface allows you to select which GPU counters and other data is captured in a trace - and if you have more complex needs, you can provide your own custom Perfetto configuration.
Deep-Dive System Analysis
With APA, you can analyze the entire system’s behavior in one view. For example, you can easily examine CPU cores - both their frequencies and the work scheduled on them or inspect processes & their thread activity.
For graphics-heavy apps, APA provides GPU performance counter data across hardware from Qualcomm, Arm, Imagination, and Samsung. You can even track battery and power consumption to see the impact of your code on power consumption.
To understand exactly where frames are spending time, SurfaceFlinger events provide deep visibility into the rendering and display composition pipeline, from initial code acquisition to final display. And with the new screenshots feature, you can visually scrub through to easily find the exact areas where you want to focus your attention.
You can open existing Perfetto traces, zoom through the timeline for precise detail, and use rulers to measure the duration of work and events. APA also lets you bookmark and annotate interesting findings, and you can pin critical tracks to the top of your screen to keep your focus exactly where it needs to be as you optimize.
Tabbed interface and split windows: You can open multiple traces in side-by-side tabs or split a single trace into two windows to compare different regions of the same trace simultaneously.
Tabbed interface showing two traces side-by-side.
Project-based workflow: APA uses a project model that allows you to keep track of multiple traces from the project sidebar. This is especially useful for gathering the results of A/B testing and longitudinal tests, and keeping all of your results together for comparison & quick access purposes.
The new project window helps you manage multiple traces.
Navigate visually using screenshots: APA lets you capture screenshots during a trace (without any noticeable performance overhead) to home in on areas where you saw something affect performance by scrubbing through the timeline. Or even just to get your bearings.
Scrubbing the timeline using screenshots for navigation (trace taken from NetMarble’s Seven Deadly Sins: Origin).
Persistent view customizations: When you pin or vertically resize tracks, we save those customizations so that they persist the next time you open the trace.
Vulkan debug trace markers for render passes: We support Vulkan debug annotations for render passes - which allow you to view Render Pass names you set from your codebase directly in the tracks and slices shown in APA.
This immensely helps you to make logical connections between the workloads you see in the profiler to where they are originating from in your codebase.
Vulkan Debug Markers allow you to keep track of what kind of work is being performed in your trace.
Use AI to build SQL queries for custom analysis work: APA supports trace analysis via SQL queries and ships with a new Perfetto SQL skill for use with your favorite AI agents. This makes it easier to build queries without needing to remember Perfetto SQL schemas or the SQL syntax.
Ask Gemini to analyze traces for you: We’ve also added another Perfetto Analysis skill to answer high-level questions for you - like “Why is my app startup slow?” - helping you to find starting points when analyzing complex traces, using your favorite AI agent to pinpoint the answers.
Agentic trace analysis in Android Performance Analyzer
FPS and Frame Duration times : You can review the FPS and Frame duration time at a glance in the tracks to correlate it with other activity happening in your trace.
Speed and robustness improvements: Rendering a trace is now typically 6x to 26x faster than Android GPU Inspector, and APA is significantly more stable when working with large traces.
We’ve worked with our early access partners to create detailed case studies showcasing how APA could be used to improve performance for Vulkan apps & games.
The Forge used Android Performance Analyzer to identify the need to batch calls to vkCmdBindDescriptorSets, which reduced CPU setup costs by ~50%. This, in turn, slowed heat production on their device by 2-3x, leading to longer session times. They also used APA to identify opportunities to move font and UI rendering work over to the GPU, improving scalability.
You can read the full case study from The Forge here.
Note: This case study demonstrates how to use custom SQL queries in the profiler to generate a total rendering cost metric.
The Forge case study showing frames presented consistently at a stable 30 FPS using APA
Netmarble used Android Performance Analyzer to fine-tune their game Seven Deadly Sins: Origin, focusing particularly on improving performance by making changes to the precision of their shaders, and exploring the impact of upscaling on the performance of their renderer.
This allowed them to reduce the GPU cost of rendering some scenes by up to 90%.
Read the full NetMarble case study here.
Netmarble validating pre- and post-optimization performance changes using APA for their game: Seven Deadly Sins: Origin
Google has been improving the Filament glTF Viewer, our physically-based rendering engine.
We spent some time digging into the viewer with a variety of scenes, and showed how to use Android Performance Analyzer to identify scenes that are too complex for the GPU, and how to trim them down to hit a target 60FPS, by improving texture compression and optimizing geometry. Memory consumption was also reduced in this process.
You can read our exploration of Filament here.
Screenshot showing GPU wait time was reduced from 25ms to 20ms by introducing dynamic resolution and measuring it through APA
The Android Performance Analyzer is available for you to try out and use today:
This is beta software, which means that you might run into an occasional bug – please report it to us if you find any (Help Menu > Submit a bug report).
We’re excited to see how you use the new Android Performance Analyzer, and how it will help your project’s performance and reliability.
Explore this announcement and all Google I/O 2026 updates on io.google.
Today, we are excited to announce that official support for Unreal Engine and Godot has arrived for Android XR. Alongside these engine expansions, we are also launching new tools designed to boost your productivity and enable new XR capabilities: the Android XR Engine Hub and the Android XR Interaction Framework.
The Android XR Engine Hub is currently available for Windows and is your mission control for development. It unifies your workflow across Unity, Unreal Engine, and Godot by serving as a high-speed bridge that streams device-created perception data straight from your device into the engine of your choice.
The Hub bridges the gap between desktop power and mobile sensor data. Instead of requiring a full build to see how your app reacts to the world, the Hub streams OpenXR extensions from the physical Android XR device directly to your Windows machine.
This means you can iterate on complex interactions in "Play Mode" while receiving live, high-fidelity data from the headset’s sensors. Without this streaming capability, testing even a minor change to eye-tracking or spatial mapping would require a full APK export and installation.
The Hub enables low-latency testing for the following streamed extensions:
Core & Interaction Support
Android XR Vendor Extensions
.png)
XR_ANDROID): Stream expressive avatar data to your editor to refine social presence without building.By virtualizing the device's hardware capabilities and streaming them over a low-latency desktop bridge, the Android XR Engine Hub allows for game engine developers to quickly iterate.
Download the Hub:
Get the Android XR Engine Hub for Windows
Learn more about Direct Preview
Through our commitments to OpenXR standards, we are ensuring that whether you are a veteran studio or an indie developer, you have best-in-class tools to help bring your creative vision to life.
Unreal Engine support is now available in developer preview, targeting version 5.6.1. This integration is built directly on using OpenXR with the support for AndroidXR vendor specific API using the Android XR vendor plugin for Unreal, you can access platform-specific extensions for advanced hand tracking, face tracking, and scene understanding (like plane detection and depth) whilst making use of Unreal blueprints or C++ support.
Get Started with Unreal:
In partnership with the Godot Foundation and W4 Games, we are bringing official Godot support to Android XR for Godot 4.6.2 and higher.
We are already seeing incredible momentum from W4 as they have ported experiences like MoAT and Expedition to Blobotopia that are already live on Google Play, proving that Godot is ready for production-grade spatial experiences today.
To unlock the full potential of the platform, use the Godot OpenXR Vendors plugin 5.1, which provides the necessary Android XR vendor extensions for features like scene meshing, dynamic resolution, light estimation and much more. We're collaborating with Godot to optimize the OpenXR implementation for the Android XR power profile and input standards.
Get Started with Godot:
The Unity OpenXR: Android XR 1.13 package is now available for Unity 6.5 Beta. Unity has expanded Application SpaceWarp support to include both uGUI and TextMeshPro. Keep an eye out for the general release of Unity 6.5 and more platform enhancements arriving this summer.
Android XR Extensions v1.3.1 for Unity
Everything else you need for comprehensive platform integration is available in our latest Android XR Extensions release:
android.software.xr.api.SPATIAL manifest tag directly through XRSessionFeature settings, making it easier than ever to define your app's Spatial API requirements and target levels.TryGetFineEyePoses extension method.Note: Android XR (Extensions): Hand Mesh has been removed; you should now use the unified Hand Mesh Data within the extensions package.
The Android XR Interaction Framework (AXRIF) is now available in developer preview. AXRIF is an unstyled, opinionated input toolkit that abstracts the complex logic required to build interfaces that are consistent with Android XR system interactions.
Instead of focusing on UI visuals, AXRIF prioritizes the underlying mechanics of the Android XR user experience. At its core is the same Transition Manager that powers the system's rich multimodal inputs, enabling state switching between 6DoF controllers, 3D mouse, hand tracking, and eye gaze. By leveraging this framework, developers can significantly reduce the implementation burden required to bring Android XR's full complement of robust interactions to their apps.
At launch, the framework provides three core capabilities:
By adopting AXRIF, your app inherits the platform's native interaction model, ensuring your app feels consistent with the rest of the OS.
Explore the Toolkit:
Interaction Framework Documentation
Download the Unity Package
There has never been a better time to dive into Android XR development. With support across Unity, Unreal, and Godot, the platform is ready for your creative vision, no matter which engine you call home. Explore our official engine partners to get started:
We're thrilled to see developers continuing to bring their apps and experiences to Android for Cars! Over the past year, we've continued to see strong growth and momentum in the app ecosystem on Android Auto and cars with Google built-in. This year at Google I/O, we're introducing updates that benefit both drivers and developers by enabling richer, more differentiated in-car experiences. With new features and templates that allow you to build once to reach users across different infotainment screens and car platforms, it's easier than ever to build for the road.
We're bringing more flexibility, new components, and new template capabilities to the Car App Library. Car App Library 1.8.0-beta01 and Car App Library 1.9.0-alpha01 are now available.
Developers can now build customized, distraction-optimized, media browsing and playback experiences for Android Automotive OS, making it easier to build once and deliver these templated media apps to more users and more cars.
To help you test the experiences on Android Automotive OS, we'll be launching updated system images for the Android Automotive OS emulator. Beginning with revision 3, the API level 35-ext15 system images will support apps built using the Car App Library media templates.
With the 1.9.0-alpha01 release of the Car App Library, we're bringing features to help you build more differentiated, expressive experiences across Android Auto and cars with Google built-in.
We're increasing the modularity of existing templates to give developers more flexibility and options for laying out content. These improvements include expanded headers for better visual emphasis, such as on detail pages, spotlight sections that can be placed in scrollable areas to highlight specific content, and grid item variations to support different content types and states.
App developers, including those shown below, have already begun building upgraded media experiences using these new Car App Library features. You can join these developers and prepare to distribute your own media apps built with the Car App Library templates by applying to participate in our early-access beta program.
You’ve already built the apps, now we’re helping you reach more users. For the first time in Android Auto, users will be able to sit back, relax, and watch videos while parked. Apps, like YouTube, will be able to deliver smooth, 60fps HD video playback. This brings the Android Auto parked experience in line with the high-fidelity, immersive experiences users already enjoy in cars with Google built-in.
This capability will start rolling out to compatible vehicles later this year, for users with phones running Android 17 and higher. If your video app is already adaptive, making it available for parked use cases in cars requires minimal effort. To express interest in making your video app available on Android Auto, fill out this form.
The next generation of Android Auto brings a more expansive user interface and the Material 3 Expressive design system you know from the phone into the car, built to seamlessly fill larger screens no matter what shape they are. With this new design, the investments you've already made in mobile widgets will be available to users of Android Auto this year, and cars with Google built-in later on, opening up new ways to reach and engage with your users while they're on the road. We're excited to unlock these new glanceable user journeys!
You can look forward to even more updates coming to cars later this year.
To deliver a more continuous user experience, we're making it possible for you to provide a templated experience while driving that can seamlessly transition to a native, adaptive app experience when the vehicle is parked.
New components and conversational templates will be coming to the Car App Library, so you can integrate agentic and voice-based flows more seamlessly in apps.
Improved app brand expression across all car surfaces allow experiences built with the Car App Library to feel easily recognizable by your users.
Google Maps SDK support is coming to cars with Google built-in. With this change, you'll be able to use the Google Maps SDK to render map-based content with the MapWithContentTemplate in point-of-interest (POI) and Weather apps on both Android Auto and cars with Google built-in.
Stay updated on these features and start building with the latest at goo.gle/cars-whats-new.
Explore this announcement and all Google I/O 2026 updates on io.google.
