# llama.cpp b11723 adds FLASH_ATTN_EXT for longer context

> The open‑source llama.cpp library released version b11723 with hexagon FLASH_ATTN_EXT tweaks that speed up large‑context inference on CPU and Vulkan.

Oossa · 2026-10-19 · https://oossa.com/en/llama-cpp-b11723-adds-flash-attn-ext-for-longer-context

The ggml‑org team posted llama.cpp version b11723 on Oct 19 2026. The update adds FLASH_ATTN_EXT improvements for longer context windows and SWA models, plus masked block skipping to cut memory fetches. Binary builds are now available for macOS (Apple Silicon and Intel), iOS, and several Ubuntu flavors, including Vulkan GPU support. The changes aim to boost performance for models like gemma‑4.

## The facts

- Version b11723 released on 2026-10-19
- Includes FLASH_ATTN_EXT enhancements for longer context

## Why it matters

Developers can run larger‑context LLMs more efficiently on desktop CPUs and compatible GPUs.

## Sources & references

1. [ggml-org/llama.cpp b11723](https://github.com/ggml-org/llama.cpp/releases/tag/b11723) – llama.cpp, 2026-10-19

Last updated: 2026-10-19