llama.cpp build 10297 ships a single server fix and a notably thinned-out macOS build matrix.
The headline change: server: fix empty response for /cors-proxy (#26656) — a bug where the CORS proxy endpoint returned empty responses. Everything else in this release is build distribution.
Platform coverage at a glance:
- macOS: Apple Silicon (arm64) only — Intel x64 build is DISABLED, and the KleediAI-enabled arm64 variant is also DISABLED
- Linux: Ubuntu x64/arm64/s390x (CPU), Vulkan (x64/arm64), ROCm 7.2, OpenVINO, SYCL FP32/FP16
- Windows: x64/arm64 CPU, OpenCL Adreno, CUDA 12 (12.4 DLLs), CUDA 13 (13.3 DLLs), Vulkan, OpenVINO, SYCL, HIP
- Mobile: iOS XCFramework, Android arm64 (CPU)
- openEuler: Partially DISABLED — 310p and 910b ACL Graph builds listed but main openEuler flag is DISABLED
The CUDA 13.3 DLLs are the newer detail. Most local-LLM toolchains still target CUDA 12.x, so a prebuilt CUDA 13 binary gives early adopters a path forward without compiling from source.
NOTE
macOS Intel (x64) has been disabled across recent llama.cpp builds. If you’re on an Intel Mac, you’ll need to compile from source or pin to an older release tag.
Available now on the b10297 release page. The project’s hosted UI lives at llama.app.