On 2026-10-16 the ggml‑org/llama.cpp project released version b11671. The update adds Muse Spark Open, a multimodal model that can process text, vision and audio inputs. It uses a mixture‑of‑experts architecture with 256 experts and supports 8‑way routing. Vision reuses the Muse Glimmer tower and audio is handled by a four‑stage causal convolutional pyramid. The code changes were contributed by Xuan Son Nguyen, Pedro Cuenca and Sigbjørn Skjæret.
Why it matters
Developers can now run a single open‑source model for text, image and audio tasks on local hardware.
