Google Android Bench 2.0: AI agents score 28% max on real Android dev tasks

Google released Android Bench 2.0, a benchmark measuring AI model + agent combinations on real Android development tasks from Google's best practices. The 30-task suite covers app conversion, build-from-scratch, architecture migration, and new feature development. Top scores reached 28% (GPT-6 Astra + Codex and Claude Fable 5.1 + Claude Code), with many tasks like Flutter-to-Android migration and XML-to-Compose migration seeing zero passes across all models.