The open‑source llama.cpp library released version b11533 on 2026‑10‑09. It fixes OpenCL kernel compilation on A6x GPUs, which were crashing on devices with an Adreno 623 chip. The patch skips a problematic kernel and adds a constant‑fold workaround for GEMV kernels. It also updates the GPU detection list to recognise the Adreno 623. Users of Linux, macOS or iOS can download pre‑built binaries from the release page.
Why it matters
Developers can now run llama.cpp on Android devices with Adreno 623 GPUs without compiler crashes.
