Google identified top AI models for developing Android apps, and the winner was Gemini

Artificial Intelligence in Android App Development: New Google Ranking

Google conducted a series of tests to determine which language models (LLMs) are best suited for generating code for Android. The results placed the top performers on a recently published list – Android Bench.

What Was Tested
Google experts evaluated how models interact with key Android tools and libraries:

- Jetpack Compose – UI generation;
- Coroutines + Flows – asynchronous programming;
- Room – working with local databases;
- Hilt – dependency injection.

They also tested more specific tasks: correct handling of critical changes when updating the SDK, interaction with system components (UI, multimedia, cameras), adaptation to devices with flexible displays, and more.

Project Goal
Google stated that existing benchmarks do not cover all the problems developers face. Therefore, Android Bench was created to show which models deliver the best results in real-world Android app development scenarios. The ranking also aims to encourage further improvement of LLMs and increase programmers’ productivity.

Testing Results
Position | Model | Score
1 | Google Gemini 3.1 Pro Preview | 72.4 %
2 | Anthropic Claude Opus 4.6 | 66 %
3 | OpenAI ChatGPT 5.2 Codex | 62.5 % … – Last position: Gemini 2.5 Flash | 16.1 %

What This Means for Developers
Google emphasizes that publishing the ranking will help teams choose the most reliable model and accelerate the creation of high-quality apps across the Android ecosystem.

Thus, Google demonstrates how modern AI models can significantly ease the lives of Android developers by providing concrete data on which ones truly perform better.