ggml‑org released llama.cpp version b12247 on November 10, 2026. The update fixes out‑of‑bounds reads and writes when using Vulkan through MoltenVK on macOS and iOS. It also resolves a sentinel‑mismatch error in the test‑backend‑ops suite and disables robustness checks for several math pipelines. Pre‑built binaries for macOS Apple Silicon, Intel, iOS, and various Ubuntu CPU targets are provided.
Why it matters
Mac developers can run LLaMA models on Apple devices via Vulkan without the crashes that the previous version caused.
