Reports on Android Bench’s October 26 update list long-horizon Android development pass rates of 40% for Claude Sonnet 5.5, 29.3% for GPT-6.1 Sol and 21.3% for Gemini 4 Argon. Completion rate, measuring progress toward a solution, is a separate metric. Models used different agent tools and reasoning settings, while sample size and confidence intervals add uncertainty. These are not clean model-only comparisons. Teams choosing app-building or migration tools should use the results as context, then evaluate their own repositories against completion criteria, cost and required human rework.