Android Dev AI Bench: Google Ranks Top Coders

Google is shaking up its Android development AI benchmark with a major update. The Android Bench, which evaluates large language models on 100 real-world Android dev tasks, now includes eight new heavy-hitters like Claude Fable 5, GPT 5.4, and Qwen 3.7 Max. The leaderboard also adds cost and efficiency metrics alongside open-weight models, making it easier for developers to pick the right tool for the job. The results are telling: Google’s own Gemini 3.1 Pro sits in fifth place, while Claude Fable 5 leads with a strong 84.5% accuracy. Google’s also adopted a new, simpler framework so devs can run their own tests and contribute feedback. Why it matters: This benchmark reveals that even as AI coding tools explode in popularity, performance varies wildly across models, and being from the biggest name doesn’t guarantee top results — a crucial reminder for developers betting on these tools for production code.