XNews

Google Android Bench 2.0 evaluates AI agents on real Android tasks

Google Android Bench 2.0 evaluates AI agents on real Android tasks

Google released Android Bench 2.0, a benchmark for model-agent combos on real Android development tasks. GPT-6 Astra + Codex and Claude Fable 5.1 + Claude Code led, but many tasks had zero passes, especially XML-to-Compose migrations and cross-platform conversions.

Source posts