# llama.cpp release b11755 adds missing MoE kernel for 9‑expert tokens

> The open‑source llama.cpp project released version b11755 on Oct 20 2026, fixing a metal GPU kernel needed for mixture‑of‑experts models.

Oossa · 2026-10-20 · https://oossa.com/en/llama-cpp-release-b11755-adds-missing-moe-kernel-for-9-expert-tokens

The ggml‑org/llama.cpp library published version b11755 on Oct 20 2026. The update fixes a missing metal kernel variant that supports nine experts per token in mixture‑of‑experts (MoE) models. Binaries for macOS (Apple Silicon and Intel), iOS, and several Ubuntu builds (CPU, Vulkan, CUDA 12) are now available for download.

## The facts

- Release tag b11755 published on Tue Oct 20 2026
- Fix adds metal kernel for 9‑expert MoE tokens

## Why it matters

Developers can now run MoE‑based LLMs on Apple Silicon GPUs, expanding model options on macOS devices.

## Sources & references

1. [ggml-org/llama.cpp b11755](https://github.com/ggml-org/llama.cpp/releases/tag/b11755) – llama.cpp, 2026-10-20

Last updated: 2026-10-20