llama.cpp b10155 adds MiMo-V2.5 audio input support

llama.cpp, the popular C++ inference engine for large language models, has released version b10155 with major multimodal support. The update adds MiMo-V2.5 audio input capability using RVQ-based models (residual vector quantization), allowing local execution of models that can process both text and spoken audio. This unlocks new use cases like voice assistants and transcription that run entirely on-device.

The release ships with prebuilt binaries across a wide range of platforms: macOS (Apple Silicon and Intel), Linux (x64 and ARM64 with Vulkan, ROCm, OpenVINO, SYCL, HIP support), Windows (CPU, CUDA 12, Vulkan, OpenCL), Android (ARM64 CPU), and iOS. This broad compatibility makes it easy for developers to integrate local multimodal AI into applications without relying on cloud APIs.