llama.cpp b10297 fixes /cors-proxy bug, adds 20+ platform builds
123K-star llama.cpp patch corrects empty /cors-proxy responses and ships prebuilt binaries for every major platform.
The ggml-org team shipped llama.cpp b10297, a new patch release for the wildly popular C/C++ LLM inference project. The headline fix addresses a server-side bug where the /cors-proxy endpoint returned empty responses, an issue tracked as #26656. This is a small but meaningful correction for developers who rely on llama.cpp's built-in CORS proxy to make local model APIs accessible from browser-based UIs, as empty responses could silently break web interfaces that depend on cross-origin requests.
Beyond the patch, the release reinforces llama.cpp's position as the most portable LLM runtime available. With 123,000 stars and 21,400 forks, this release delivers prebuilt binaries across an enormous matrix of platforms: macOS Apple Silicon (including KleidiAI-optimized builds), Intel Macs, iOS XCFramework, Linux on x64/arm64/s390x, Vulkan and ROCm 7.2 for AMD GPUs, OpenVINO for Intel accelerators, SYCL for Intel GPUs, Windows with CUDA 12.4/13.3 DLLs, OpenCL for Adreno, Android arm64, and openEuler for Ascend NPUs. This breadth means enterprise teams can deploy the same inference engine on everything from a MacBook to a cloud GPU cluster, making llama.cpp a de facto standard for local and edge LLM serving.
- Fixes empty response bug in /cors-proxy endpoint, addressing issue #26656
- Ships prebuilt binaries for 20+ platform/accelerator combos including CUDA 12/13, ROCm 7.2, Vulkan, OpenVINO, and SYCL
- llama.cpp maintains massive adoption with 123K GitHub stars and 21.4K forks
Why It Matters
The CORS fix stabilizes browser-based LLM tools, while multi-platform binaries keep llama.cpp the go-to for on-prem inference.