Skip to content

v1.0.2

Choose a tag to compare

@abhay-byte abhay-byte released this 31 May 17:55
· 15 commits to main since this release

FinalBenchmark v1.0.2

Release Date: May 31, 2026
Version Code: 10
APK Size: ~15 MB (arm64-v8a + armeabi-v7a, R8 minified)


Changelog

AI Benchmark

  • Replaced the entire Litert/TFLite backend with a custom Native C++ implementation to resolve infinite freezing during LLM inference
  • AI backend now correctly attempts to load Vulkan/OpenCL/OpenGL ES accelerators before falling back to CPU
  • Fixed the AI threading model to prevent silent fallback to CPU mode caused by EGL context thread affinity issues
  • Recalibrated AI scoring baselines using actual NEON CPU measurements from the Snapdragon 8 Gen 3 reference device

CPU Benchmark

  • Restored optimized CPU iteration targets for mid and flagship tiers to ensure the entire benchmark suite completes within 3 minutes

UI

  • AI benchmark badges now accurately display the utilized accelerator backend (Vulkan, OpenCL, OpenGL ES, or CPU)