llama.cpp b10087 adds support for Laguna XS.2 & M.1 models

The open-source llama.cpp project, known for enabling local inference of large language models on consumer hardware, released version b10087 on July 22, 2025. This update introduces support for Laguna XS.2 & M.1 models, expanding the range of LLMs that can run efficiently offline. The release is signed with a verified GitHub signature and includes pre-compiled binaries for a wide array of platforms: macOS (Apple Silicon and Intel), Windows (CPU, CUDA 12/13, Vulkan, ROCm 7.2, OpenVINO, SYCL, HIP), Linux (x64 and arm64 with various backends), Android (arm64 CPU), and iOS (XCFramework). Additionally, openEuler builds for x86 and aarch64 are provided, along with UI assets.

This release underscores llama.cpp's commitment to broad hardware support and efficient execution, enabling developers and power users to run the new Laguna model family locally without cloud dependencies. The Laguna models likely target improved performance per parameter or specialized capabilities, though details remain sparse. By supporting both compute-heavy backends like CUDA and lightweight CPU-only modes, llama.cpp b10087 ensures flexible deployment from servers to edge devices. This update follows the project's rapid release cadence, building on its 121k GitHub stars and active community.