The ggml‑org team released llama.cpp version b11622 on 2026‑10‑14. The main change fixes how floats are written in Jinja templates: it now uses the shortest decimal that round‑trips, just like Python’s json.dumps. The update also lists a wide range of pre‑built binaries for Apple Silicon, Intel, Linux (CPU, Vulkan, CUDA 12/13, ROCm, OpenVINO, SYCL), Android Snapdragon, Windows (CPU, CUDA, Vulkan, OpenCL) and openEuler.
Why it matters
Developers can now rely on accurate numeric values when generating prompts or tool calls, reducing bugs in LLM‑driven applications.
