React Native Navigation Benchmarks

I’m trying to figure out how to build the fastest possible React Native app. A big piece of that is understanding what navigation costs — it runs at startup, it runs on every screen transition, and most teams pick a library without knowing the real numbers.

So I built the same app four times — one for each major navigation library — and measured everything: cold start, RAM, FPS, and what the JS thread actually does. This is what I found.

Why Android? Android is where React Native performance bottlenecks most often show up in production, and it gives the richest profiling data. Perfetto Systrace captures every thread boundary — UI thread, JS thread, RenderThread, SurfaceFlinger — at microsecond resolution. The Hermes CPU sampler gives a source-mapped JS call stack for the full startup burst. Together they make it possible to answer why, not just how slow. iOS has equivalent tools but the data is less granular. Everything here is Android-only.

Setup: Expo SDK 56 · RN 0.85 · Hermes · New Architecture (bridgeless/Fabric) · Samsung Galaxy A16 · Android 14 · release/profileable builds. Cold start = OS Displayed metric; FPS/CPU/RAM from Flashlight driving Maestro; breakdowns from Perfetto Systrace + source-mapped Hermes CPU profiles. All data and tooling at the StateOfReactNativeNavigation repo.

Every app renders identical screens from a shared UI package — a 30-row list, a push to a Details screen, a tab switch. The only variable is the navigation library.

The headline numbers

Library	Cold start (ms)	Avg FPS	Avg CPU %	Peak RAM (MB)
react-native-navigation	316	59.8	31.2	195
React Navigation v7	358	59.9	37.8	214
navigation router	398	59.8	34.8	241
Expo Router	917	59.8	37.1	308

Cold start = median of 3 runs; RAM = Flashlight peak over the navigate flow. All four hold ~60 FPS — the differences are in startup cost and memory.

Three things that surprised me

Expo Router is ~3× the cold start — but Reanimated isn’t the main reason. The Expo Router template ships react-native-reanimated@4 and react-native-worklets, which none of the others do. But a controlled experiment showed Reanimated adds only ~62 ms to cold start. The bulk of the gap is the 2× bigger bundle and the router-on-top-of-React-Navigation layering that evaluates 106 JS modules at boot vs 36–43 for the others.

Reanimated is the RAM story. Adding only Reanimated to the leanest app (rn-navigation) reproduced Expo Router’s entire RAM premium: +125 MB, almost entirely anonymous heap from the second Hermes runtime Worklets spins up. Bundle mode is supposed to fix this but it actually regress your cold start given it must parse your entire bundle.

rn-navigation wins because navigation is native. Its JS bundle evaluates in 55 ms. The tabs and stack are Kotlin views — the JS thread barely runs at startup. React Navigation’s JS bundle takes 168 ms because it builds a real component tree reconciled by Fabric. But it’s not all flowers, rn-navigation’s approach also means it will choke on large screens and drop many frames.

The series

• Deep Diving Where Time Is Spent — why Expo Router is slow, what React Navigation adds on top of rn-navigation, the Hermes hot functions for each library, and the Reanimated controlled experiment.
• What’s the Expo Tax? — the fixed cost of adopting Expo (~36 ms, ~13 MB RAM, ~16 MB APK) and the marginal cost of each extra Expo module.
• What Does Navigating to a Screen Cost? — press→paint on a trivial screen and a heavy screen, JS call stacks per library, and why “first frame” means very different things when the screen actually has work to do.

Caveats

• rn-navigation is a bare RN app; the other three are Expo apps. Some of its lead is “no expo-modules-core”, not just the navigation library. RNN owns the React host, which is incompatible with Expo’s host factory.
• One device, one simple UI. Samsung Galaxy A16, Android 14, Hermes, New Architecture. Heavier screens change the FPS and navigation cost story.
• Numbers are indicative. Hermes sampling is coarse; cold-start medians are 3 runs with constant profiling instrumentation (inflates absolute times identically across all apps).
• Expo Router does more for the cost. Deep linking, lazy screen loading, file-based routing, web support. These numbers measure startup on a trivial app — not a verdict on whether the features are worth it.