[Android] - Higher memory usage on 0.85.x vs 0.83.x with the new Hermes (~25–30% more RAM, same app)

Description

After upgrading an app from 0.83.9 → 0.85.3 with no application code changes (only the react-native / @react-native/* bump and the Hermes that ships with it), RAM usage is consistently ~25–30% higher across an identical scripted flow. It starts ~110 MB higher and widens to ~160 MB higher by the end of the run.

Both builds run the New Architecture + Hermes and the exact same third-party dependencies; the only meaningful difference is the React Native version (and therefore the bundled Hermes / "Hermes V1").

Environment

• Platform: Android, physical device (Google Pixel 9 Pro, Android 16, arm64-v8a)
• JS engine: Hermes (default, hermesEnabled=true)
• Architecture: New Architecture (Fabric + TurboModules) enabled in both builds
• Build type: release-optimized profileable build (Hermes bytecode, minified, debuggable=false)
• Key deps (identical across both builds):
• react-navigation 7, @shopify/flash-list 2, react-native-reanimated 4 + react-native-worklets, react-native-screens 4, react-native-gesture-handler 3

Steps to reproduce

1. Build & install a release (profileable) APK on 0.83.9.
2. Drive a fixed UI flow under measurement. I used Flashlight + a Maestro flow that scrolls a FlashList product feed, opens a product screen, taps a related item, and repeats 10×.
3. Record RAM, then repeat on 0.85.3 with the same app code and flow.

# on 0.83.9
flashlight test --bundleId <app.id> --iterationCount <N> \
  --testCommand "maestro test flow.yaml" --resultsFilePath results_83.json

# on 0.85.3 (identical app code)
flashlight test --bundleId <app.id> --iterationCount <N> \
  --testCommand "maestro test flow.yaml" --resultsFilePath results_85.json

flashlight report results_83.json results_85.json

React Native Version

Regression between 0.83.9 and 0.85.3.

Affected Platforms

Runtime - Android

Output of npx @react-native-community/cli info

System:
  OS: macOS 26.5
  CPU: (10) arm64 Apple M5
  Memory: 168.42 MB / 32.00 GB
  Shell:
    version: "5.9"
    path: /bin/zsh
Binaries:
  Node:
    version: 22.14.0
    path: /Users/josemaria.elias/.nvm/versions/node/v22.14.0/bin/node
  Yarn: Not Found
  npm:
    version: 10.9.2
    path: /Users/josemaria.elias/.nvm/versions/node/v22.14.0/bin/npm
  Watchman:
    version: 2026.05.18.00
    path: /opt/homebrew/bin/watchman
Managers:
  CocoaPods:
    version: 1.16.2
    path: /Users/josemaria.elias/.gem/ruby/3.3.6/bin/pod
SDKs:
  iOS SDK:
    Platforms:
      - DriverKit 25.5
      - iOS 26.5
      - macOS 26.5
      - tvOS 26.5
      - visionOS 26.5
      - watchOS 26.5
  Android SDK: Not Found
IDEs:
  Android Studio: 2025.3 AI-253.32098.37.2534.15336583
  Xcode:
    version: 26.5/17F42
    path: /usr/bin/xcodebuild
Languages:
  Java:
    version: 17.0.19
    path: /usr/bin/javac
  Ruby:
    version: 3.3.6
    path: /Users/josemaria.elias/.rubies/ruby-3.3.6/bin/ruby
npmPackages:
  "@react-native-community/cli":
    installed: 20.1.0
    wanted: 20.1.0
  react:
    installed: 19.2.3
    wanted: 19.2.3
  react-native:
    installed: 0.85.3
    wanted: 0.85.3
  react-native-macos: Not Found
npmGlobalPackages:
  "*react-native*": Not Found
Android:
  hermesEnabled: true
  newArchEnabled: true
iOS:
  hermesEnabled: true
  newArchEnabled: true

Stacktrace or Logs

Expected behavior
Comparable memory between 0.83.x and 0.85.x for the same workload (ideally lower, given the Hermes work).

Actual behavior
0.85.3 uses materially more RAM for the entire run — ~110 MB higher at start, ~160 MB higher by the end (~25–30%).

RN 0.83.9	RN 0.85.3	Δ
RAM at start	~370 MB	~480 MB	+~110 MB
RAM at end	~560 MB	~720 MB	+~160 MB


A snapshot breakdown via adb shell dumpsys meminfo <app.id> attributes the increase mainly to Native heap + Graphics, consistent with the engine/runtime + image buffers retaining more memory.

MANDATORY Reproducer

https://github.com/SirJalias/AwesomeProject

Screenshots and Videos

必須の AwesomeProject reproducer から開始し、React Native 0.83.9 と 0.85.3 に対して、文書化されている Flashlight と Maestro のフローを実行します。結果ファイルと adb shell dumpsys meminfo の出力を比較し、報告された Native heap と Graphics の差異に注目します。回帰を特定し、同等のメモリ使用量を実証するか、回帰テストまたは測定によって確認済みの原因を文書化できれば完了です。