How Meta Accelerates Its Android apps with Baseline Profiles
Original post Accelerating our Android apps with Baseline Profiles
Article Summary
Meta's Android team just dropped their playbook for achieving up to 40% performance improvements across billions of devices. The secret? Getting surgical with Baseline Profiles.
Jimmy Cleary, Hyun Jun Jin, Andreas Gampe, Minjang Kim, and Shane Nay from Meta share how they've leveraged Android's Baseline Profiles feature to dramatically improve app performance. With apps like Facebook and Instagram loading over 20,000 classes on startup, they needed a solution that could adapt to thousands of daily commits while optimizing both cold starts and post-startup user journeys.
Key Takeaways
- Meta loads 20,000+ classes on startup in Facebook and Instagram alone
- Started conservative at 80-90% thresholds, now include classes appearing in just 20% of traces
- Custom ClassLoader logs real user behavior to create production-representative profiles
- Compiled code is 10x larger than bytecode, requiring careful tuning to avoid regressions
- Improvements range from 3% to 40% across cold start, scroll, and navigation metrics
By collecting real user data and aggressively experimenting with profile thresholds, Meta achieved consistent double-digit performance gains across all major Android apps.
Continue reading
This page summarizes “Accelerating our Android apps with Baseline Profiles” by Meta, published Oct 1, 2025. All credit for the original work goes to the authors.