The llama.cpp project published b12125 as a pre-release on November 5, 2026. It changes default OpenCL settings for Adreno X2 and X2E graphics hardware, including enabling f32 and q5_K multi-column operations by default on X2.
For X2E, it also selects small-N f32 and q5_K multi-column verification operations by default. The release page lists downloads for several platforms, including Windows and Android builds for Snapdragon devices, but gives no benchmark results.
Why it matters
People running llama.cpp on devices with Adreno X2 or X2E hardware get different default GPU operations, but the release notes do not say whether this improves speed or accuracy.
