ggerganov/llama.cpp Released b12098 With Qwen3-VL Video Prompt Support
Release b12098 of llama.cpp introduces accurate support for the Qwen3-VL video prompt format alongside standard multi-platform compiled binaries. This update expands the multimodal capabilities of the library to handle video inputs from the Qwen3-VL architecture more precisely. This update is crucial for practitioners running advanced vision-language models locally, enabling seamless hardware acceleration for complex multimodal workflows. It ensures that open-source infrastructure keeps pace with rapidly evolving multimodal architectures developed by teams like Alibaba Cloud. The release includes pre-built binaries across multiple operating systems and accelerator backends, including CUDA, Vulkan, OpenVINO, and SYCL. Notably, macOS Apple Silicon builds with KleidiAI enabled remain temporarily disabled pending further integration work.
## BACKGROUND
llama.cpp is an open-source library that performs local inference on large language and vision models using the GGML tensor library, serving as a foundational backend for tools like Ollama and LM Studio. Qwen3-VL is a multimodal large language model series developed by Alibaba Cloud that natively processes text, images, and video inputs.