Models & Labs
llama.cpp adds Ling 3.0 VL support
llama.cpp Releaseshigh confidence
Why it matters
- →Enables local inference of Ling 3.0 VL, a significant open-weight vision-language model.
- →Stabilizes complex multimodal architectures like BailingMoeV3 for reliable local deployment.
- →Expands hardware support to include Snapdragon NPUs and ROCm 10.0 for broader accessibility.
The llama.cpp project has released build b11156, adding support for the Ling 3.0 vision-language model by integrating it into the BailingMoeV3 architecture. The update addresses specific architectural requirements, such as maintaining shared NORM rope lists during mrope section gating, to ensure stable inference. New binaries are provided for a wide range of hardware, including CUDA 12/13, ROCm 10.0, OpenVINO, and Snapdragon NPUs on Linux and Windows. This release allows developers to run Ling 3.0 locally with optimized performance across diverse GPU and NPU backends.
Read originalThe story around this
Earlier coverage that leads up to this article, and what followed. Lines connect each piece to the closest one after it, converging here.
How we got here
- 1Llama.cpp b9980 Release Focuses on Multimodal Model Handling
llama.cpp Releases · July 13, 2026 · Same story
- 2llama.cpp v0.3.0 Release
llama.cpp Releases · August 26, 2026 · Same story
More from llama.cpp Releases
Coding Toolscoding
llama.cpp b12336 fixes HunyuanVL crash
This release quietly patches a critical memory overflow in the mtmd module that caused crashes when loading HunyuanVL models. The issue stemmed from F64 position embeddings overflowing an F32 buffer, a subtle bug that broke multimodal inference for users relying on this specific architecture. By enforcing strict F32 requirements, llama.cpp stabilizes support for Tencent's vision-language model without adding new features or changing the core API. It’s a necessary maintenance update that keeps local multimodal workflows from breaking unexpectedly.
llama.cpp Releases
Models & Labsother
llama.cpp b12340 adds CUDA 13 and ROCm 10.0
This release quietly closes the hardware gap for local inference by adding default support for CUDA 13 and ROCm 10.0 across Linux and Windows. Users with newer NVIDIA GPUs or AMD accelerators no longer need to compile from source to get performance, while Apple Silicon builds now ship KleidiAI disabled by default, shifting the focus back to standard ARM optimizations. The inclusion of Snapdragon support for Linux and Android marks a significant step toward on-device AI beyond the x86/ARM desktop ecosystem. It is less about new features and more about ensuring llama.cpp remains the universal runtime for every available accelerator.
