llama.cpp b10358 drops with multi-platform builds, 124k GitHub stars
The local LLM runtime just released b10358 with prebuilt binaries for every major platform.
ggml-org shipped llama.cpp b10358, a new release of the wildly popular C/C++ LLM inference library that has amassed 124,000 stars and 21,600 forks on GitHub. Tagged on August 11 and signed with GitHub's verified signature, this incremental release addresses a review comment for PR #25532 (tracked via #26852), indicating continued polish on the library's core engine. While no major feature is flagged, the release's real value is its aggressively expanded build matrix.
The b10358 release streamlines local AI deployment by offering prebuilt binaries across every major platform. For Apple users, it provides macOS arm64 builds with and without KleidiAI (ARM's CPU acceleration library), plus an XCFramework for iOS. Linux users get CPU builds for x64, arm64, and s390x, along with GPU-accelerated variants for Vulkan, ROCm 7.14, OpenVINO, and SYCL (both FP32 and FP16). Windows users can choose from CUDA 12.4 and CUDA 13.3 DLLs, a preview CUDA 13.4 build for arm64, plus Vulkan, OpenVINO, SYCL, and ROCm 7.14 versions. Android and openEuler flavors round out the package, covering Ascend 310p and 910b accelerators. This breadth means AI developers and enthusiasts can run and serve local LLMs on virtually any hardware, from edge IoT devices to data center GPUs, without compiling from source.
- llama.cpp b10358 released Aug 11 with a verified commit fixing PR #25532
- Prebuilt binaries for macOS, Linux, Windows, Android, iOS, and openEuler, including CUDA 12/13, ROCm 7.14, Vulkan, OpenVINO, and SYCL
- Project boasts 124k GitHub stars and 21.6k forks, underlining its dominance for local LLM inference
Why It Matters
Frequent llama.cpp updates keep local LLM inference fast and accessible across consumer and enterprise hardware.