The open‑source llama.cpp library released version b11560 on Oct 12, 2026. It adds IM2COL_3D support for Qualcomm’s Hexagon NPU, including DMA pathways and fixes to FP16 handling. The change also removes inefficient loops and improves thread handling. The update ships builds for many platforms, from macOS and Windows CPUs to Android devices with Snapdragon CPUs, Adreno GPUs, and Hexagon NPUs.
Why it matters
Developers can now run llama.cpp models faster on Snapdragon phones that include a Hexagon NPU.
