# llama.cpp release adds MUSA support and updates CUDA versions

> The open‑source Llama.cpp library now handles MUSA GPUs and includes builds for CUDA 12.8 and 13.4.

Oossa · 2026-09-30 · https://oossa.com/en/llama-cpp-release-adds-musa-support-and-updates-cuda-versions

Produced and translated with AI assistance. Check the original sources below.

The ggml‑org team released a new version of llama.cpp on September 30, 2026. It fixes a problem where the MUSA GPU driver did not define CUDA_ARCH, causing some kernels to compile empty. The patch adds proper architecture detection and removes redundant checks. The release also lists many pre‑built binaries, now covering CUDA 12.8 and 13.4 on Linux and Windows, plus support for various ARM and Apple Silicon targets.

## The facts

- Release date: September 30, 2026 – “Wed Sep 30 2026 13:56:30 GMT+0200”
- Adds CUDA 12.8 and CUDA 13.4 library builds for Ubuntu and Windows

## Why it matters

Developers can now run Llama.cpp on MUSA GPUs and newer CUDA versions without manual patches.

## Sources & references

1. [ggml-org/llama.cpp b11282](https://github.com/ggml-org/llama.cpp/releases/tag/b11282) – llama.cpp, 2026-09-30

Last updated: 2026-09-30