It’s not an issue. When we run LTO optimizations, strip all symbols, we get 2MB compressed for decently complex GPU-accelerated rendering (minfx.ai)