XNews

Google Android Bench 2.0 evaluates model-agent combos on real Android development tasks

Google Android Bench 2.0 evaluates model-agent combos on real Android development tasks

Google released Android Bench 2.0, a benchmark measuring model-agent pairs on 30 real Android engineering tasks from production apps. Top scores: GPT-6 Astra + Codex and Claude Fable 5.1 + Claude Code, with highest pass rate 28%. All models failed on cross-platform migration and XML-to-Compose tasks.

Source posts