How phones compare

What the benchmark does

Every run builds the same project — a small but genuinely real Gradle Android app bundled with the benchmark, with its plugin versions pinned so the workload can't drift between releases. Three phases are measured:

1. Full build. Build outputs and project caches cleared, the Gradle daemon disabled, then a complete build to an APK. This is the number that most reflects the device.
2. Incremental build. One source file is touched and the project is rebuilt — the case you actually live in while working.
3. Large file read. Reading a 4,000-line source file from storage.

How to read this table

• Only compare within the same benchmark version. If the workload changes, the version is bumped and old numbers stop being comparable. They are not silently mixed together.
• Thermal state dominates. The same phone, hot, can be dramatically slower than the same phone, cool. Runs taken on a throttled or nearly-flat device are marked and excluded from these aggregates — they stay on the user's device.
• These are medians from real users, not a lab. A device with one submitted run tells you much less than a device with twenty. The run count is shown for exactly that reason.
• "Builds" means it works. Any device listed here has completed a real on-device build at least once. That is the practical question most people are asking.

Submitting your own

In the app, open Settings → Benchmark this device. You'll see your results locally first. Publishing them here is a separate, explicit action — nothing is uploaded unless you tap it.

A submission is anonymous: it carries the device model, chipset, RAM, Android version, ABI, the app and benchmark versions, the timings, and the thermal/battery snapshot that determines whether the run was valid. It carries no account identifier, no device identifier and no location, and it doesn't require you to be signed in. The privacy policy covers this.