#GoogleIO 2025 Developer Keynote Recap

Key Takeaways

You saw how Project Astra can make AI feel natural, and a lot of those capabilities are now available today in the Live API. So, let's select the new 2.5 Flash native audio model. The model is now better at ignoring stray sounds or voices. It also natively supports 24 languages. Hey KC, since there are people, you know, coming from all over the world, could you show us Shoreline Amphitheatre on a map? All right, I've put Shoreline Amphitheatre on the map for everyone. Is there anything else I can help with? Yeah, could you show us some good coffee houses nearby with really great Wi-Fi? Okay, I've searched for coffee houses with Wi-Fi near Shoreline Amphitheatre. Excellent. We're continuing to make it as easy as possible to build adaptively with new features in our Compose adaptive layouts library, like pane expansion. And if you're building adaptively, you're building for Android XR, the extended reality platform built together with Samsung. Also powers headsets like Project Muon from Samsung, which you can start building for right now, knowing it'll be in the hands of consumers later this year. Let's demo a new AI agent uh coming soon in Android Studio to help with version updates. I have a project loaded here that well, I haven't really touched it in a year. The agent is uh trying to build a project and then use Gemini to figure out how to fix the problems. Then it's going to iterate until the build is successful. We'll show you some new capabilities that make it easier to build common but surprisingly complex UI elements like carousels and hover cards. Today we're unlocking new multimodal capabilities from Gemini Nano. Our multimodal built-in AI APIs let you create experiences where users can interact with Gemini using audio and image input. Let's go ahead and give this a try. Now, normally I'd be doing this with my phone, but I'm just developing right now, and so I'm just going to use this webcam. I'm going to snap a photo of this ticket. Let's try this out. And bam! The built-in multimodal AI instantly located my seat selection in the theater. Awesome. You can bring your Figma designs right to life in Firebase Studio with some help from Builder.io. And what I love about this Figma export is that it didn't just generate this large monolith of code, it generated individual, well-isolated components that make up the whole page. And so now, once I update this file, and go back into the web page, I can click around and I have myself a fully working product detail page. I'm thrilled to announce Gemma 3N, a model that can now run on as little as 2 GB of RAM. It's much faster and leaner mobile hardware compared to Gemma 3. We've also added audio understanding, making it truly multimodal. Sign Gemma is a new family of models trained to translate sign language to a spoken language text, but it's best at American sign language in English.