XNews

Google Android Bench 2.0: top models score 28% pass rate on real app tasks

Google Android Bench 2.0: top models score 28% pass rate on real app tasks

Google released Android Bench 2.0, a benchmark measuring model-plus-agent combinations on real Android development tasks derived from Google best practices. The top result was 28% pass rate, with GPT-6 Astra + Codex and Claude Fable 5.1 + Claude Code leading. All models failed on cross-platform conversion and XML-to-Compose migration tasks.

Source posts