# llama.cpp b11685 adds SYCL wide stores for q4_K/q5_K dequant

> The open‑source LLM runner releases b11685 with a SYCL optimisation for quantized weights and new binaries for macOS, iOS and Linux.

Oossa · 2026-10-16 · https://oossa.com/en/llama-cpp-b11685-adds-sycl-wide-stores-for-q4-k-q5-k-dequant

The ggml‑org team published llama.cpp version b11685. It adds a SYCL optimisation called wide stores for the q4_K and q5_K weight dequantisation. The release includes pre‑built binaries for macOS Apple Silicon, macOS Intel, iOS, and several Ubuntu builds (CPU, Vulkan and CUDA 12.8). The change is aimed at speeding up quantised model inference on SYCL‑compatible GPUs.

## The facts

- Release tag: b11685
- Published: 2026-10-16

## Why it matters

Developers can run quantized LLMs faster on SYCL GPUs, expanding cheap hardware options for AI apps.

## Sources & references

1. [ggml-org/llama.cpp b11685](https://github.com/ggml-org/llama.cpp/releases/tag/b11685) – llama.cpp, 2026-10-16

Last updated: 2026-10-16