TL;DR: Llama.cpp build b9543 introduces video support for Qwen3.5 models and frame merge capabilities for Qwen-VL vision-language models.
Summary: The latest llama.cpp release adds native video support for Qwen3.5 models. It also implements a frame merge feature for Qwen-VL-based models to improve multimodal processing efficiency and includes compatibility fixes for LLaVA-UHD.
Why it matters: AI developers can now run local, high-performance Qwen video-language models on consumer hardware. Builders should update llama.cpp to the latest build to experiment with local Qwen3.5 video processing.
Source: github.com