# llama.cpp 0.7.2 release fixes Metal GPU memory leaks

> The open‑source Llama model library adds a Metal fix and ships new binaries for macOS, iOS and Linux.

Oossa · 2026-10-21 · https://oossa.com/en/llama-cpp-0-7-2-release-fixes-metal-gpu-memory-leaks

The ggml‑org team released llama.cpp version b11791 on 2026‑10‑21. The update fixes memory‑leak bugs in the Metal GPU path used on Apple Silicon Macs and iOS devices. It also cleans up the capture logic for GPU work. Pre‑built binaries for macOS (Apple Silicon and Intel), iOS, and several Linux configurations—including Vulkan and CUDA 12.8—are available for download.

## The facts

- Release date: 2026‑10‑21 ("Wed Oct 21 2026")
- Fix: memory‑leak and logic cleanup in Metal GPU capture path

## Why it matters

Developers running Llama models on Mac hardware will see lower RAM usage and more stable GPU performance.

## Sources & references

1. [ggml-org/llama.cpp b11791](https://github.com/ggml-org/llama.cpp/releases/tag/b11791) – llama.cpp, 2026-10-21

Last updated: 2026-10-21