llama.cpp pre-release b11729 adds Qwen3 architecture support with bidirectional attention
ggml org's llama.cpp published pre release build b11729 on Oct 19, 2026, adding support for the Qwen3 model architecture using bidirectional attention and a de…
Published on MyPrivateClaw
Oct 19, 2026, 11:18 PM UTC
Coverage date
Oct 19, 2026
Last updated
Oct 19, 2026, 11:18 PM UTC
News summary
ggml org's llama.cpp shipped pre release build b11729 on Oct 19, 2026 (commit verified on Oct 19, 2026), adding support for the Qwen3 model architecture using bidirectional attention and a dense 2 output projection via PR 30607. The release is pre release only, with no stable tag yet. The b11729 release ships 38 assets including prebuilt binaries for macOS/iOS, Linux, Windows, and Android across CPU, Vulkan, CUDA 12.8, CUDA 13.4, ROCm 10.0, OpenVINO 2026.4.1, SYCL FP32/FP16, and Snapdragon (CPU, Adreno GPU, Hexagon NPU) backends, with CUDA 13.4 libraries offered as a separate download. The b11729 release was created with GitHub's verified GPG key (GPG key ID B5690EEEBB952194) and carries attestation attestations/57363774. Because the release is pre release only, no formal release notes exist for the Qwen3 feature; the release page returned a loading error, and the body text shown was th…