We’re in an important moment where AI changes everything, from how we work to the expectations that users have for your apps, and our goal on Android is to transform this AI evolution into opportunities for you and your users. Today in our Fall episode of The Android Show, we unpacked a bunch of new updates towards delivering the highest return on investment in building for the Android platform. From new agentic experiences for Gemini in Android Studio to a brand new on-device AI API to the first Android XR device, there’s so much to cover – let’s dive in!
Build your own custom Gen AI features with the new Prompt API
On Android, we offer AI models on-device, or in the cloud. Today, we’re excited to now give you full flexibility to shape the output of the Gemini Nano model by passing in any prompt you can imagine with the new Prompt API, now in Alpha. For flagship Android devices, Gemini Nano lets you build efficient on-device options where the users’ data never leaves their device. At I/O this May, we launched our on-device GenAI APIs using the Gemini Nano model, making common tasks easier with simple APIs for tasks like summarization, proofreading and image description. Kakao used the Prompt API to transform their parcel delivery service, replacing a slow, manual process where users had to copy and paste details into a form into just a simple message requesting a delivery, and the API automatically extracts all the necessary information. This single feature reduced order completion time by 24% and boosted new user conversion by an incredible 45%.
Tap into Nano Banana and Imagen using the Firebase SDK
When you want to add cutting-edge capabilities across the entire fleet of Android devices, our cloud-based AI solutions with Firebase AI Logic are a great fit. The excitement for models like Gemini 2.5 Flash Image (a.k.a. Nano Banana) and Imagen have been incredible; now your users can now generate and edit images using Nano Banana, and then for finer control, like selecting and transforming specific parts of an image, users can use the new mask-based editing feature that leverages the Imagen model. See our blog post to learn more. And beyond image generation, you can also use Gemini multimodal capabilities to process text, audio and image input. RedBus, for example, revolutionized their user reviews using Gemini Flash via Firebase AI Logic to make giving feedback easier, more inclusive, and reliable. The old problem? Short, low-quality text reviews. The new solution? Users can now leave reviews using voice input in their native languages. From the audio Gemini Flash is then generating a structured text response enabling longer, richer and more reliable user reviews. It’s a win for everyone: travelers, operators, and developers!
Helping you be more productive, with agentic experiences in Android Studio
Helping you be more productive is our goal with Gemini in Android Studio, and why we’re infusing AI across our tooling. Developers like Pocket FM have seen an impressive development time savings of 50%. With the recent launch of Agent Mode, you can describe a complex goal in natural language and (with your permission), the agent plans and executes changes on multiple files across your project. The agent’s answers are now grounded in the most modern development practices, and can even cross-reference our latest documentation in real time. We demoed new agentic experiences such as updates to Agent Mode, the ability to upgrade APIs on your behalf, the new project assistant, and we announced you’ll be able to bring any LLM of your choice to power the AI functionality inside Android Studio, giving you more flexibility and choice on how you incorporate AI into your workflow. And for the newest stable features such as Back Up and Sync, make sure to download the latest stable version of Android Studio.
Elevating AI-assisted Android development, and improving LLMs with an Android benchmark
Our goal is to make it easier for Android developers to build great experiences. With more code being written by AI, developers have been asking for models that know more about Android development. We want to help developers be more productive, and that’s why we’re building a new task set for LLMs against a range of common Android development areas. The goal is to provide LLM makers with a benchmark, a north star of high quality Android development, so Android developers have a range of helpful models to choose for AI assistance.