XNews

Google Android Bench 2.0 tests AI agents on real Android dev; top pass rate 28%

Google Android Bench 2.0 tests AI agents on real Android dev; top pass rate 28%

Google introduced Android Bench 2.0, a benchmark evaluating AI models with coding agents on real-world Android development tasks from best practices. Tasks include app conversion, architecture migration, and new feature development. GPT-6 Astra + Codex and Claude Fable 5.1 + Claude Code achieved the highest pass rates (up to 28%), while many cross-platform migration tasks saw zero passes.

Source posts