llama.cpp Release b10981: OpenVINO Optimizations for Stateful Decode and GPU MoE Inference

Core summary

One sentence to understand this update

llama.cpp released version b10981, featuring significant OpenVINO optimizations for stateful decode and GPU Mixture-of-Experts (MoE) inference, alongside fixes for Gemma-4 and MSVC.

Impact & opportunity

What this could mean

Builders can leverage improved OpenVINO performance for running local LLMs more efficiently, especially for MoE models and Gemma-4, enhancing speed and stability on compatible hardware.