XNews

Android Bench 2.0: model+agent benchmark with 28% max pass rate

Android Bench 2.0: model+agent benchmark with 28% max pass rate

Google released Android Bench 2.0, a benchmark for model+agent combinations on real Android development tasks. Top models (GPT-6 Astra, Claude Fable 5.1) achieved only 28% pass rate. Tasks like Flutter-to-Android migration had 0% pass across all models.

Source posts