ggml‑org released version b11804 of llama.cpp on 2026‑10‑22. The update adds an OpenCL routine that computes argmax from 32‑bit floats to 32‑bit integers, improving GPU inference speed. It also includes fresh pre‑built binaries for Apple Silicon, Intel macOS, iOS, and several Ubuntu flavors (CPU, Vulkan, CUDA). The changes are authored by OpenCode and Li He of Qualcomm.
Why it matters
Developers can now run llama.cpp models faster on OpenCL‑compatible GPUs, expanding hardware options for hobbyist AI projects.
