llama.cpp Releases Build b10809 Bumping Project Version to 0.4.0
llama.cpp has released build b10809, officially bumping the project version to 0.4.0. The release includes prebuilt binaries for macOS, Linux, Windows, and Android across diverse compute backends including CUDA 12/13, Vulkan, ROCm 10.0, OpenVINO, and SYCL. As the core open-source inference engine powering much of the local AI ecosystem (such as Ollama and LM Studio), updates to llama.cpp maintain cross-platform hardware support and performance. Keeping track of version releases helps downstream tools and developers integrate the latest hardware acceleration fixes. This release provides binaries for preview features like Windows ARM64 with CUDA 13.4 DLLs alongside standard platforms. However, specific build targets remain disabled in this release, including macOS Apple Silicon with KleidiAI enabled and openEuler builds.
## BACKGROUND
llama.cpp is a widely used open-source C/C++ library designed for fast local inference of Large Language Models (LLMs) on consumer hardware. KleidiAI is an open-source library developed by Arm that provides assembly-optimized micro-kernels to accelerate AI workloads on Arm processors.