Ollama has recently released version 0.40.9, introducing significant new capabilities for local large language models. This update enables speech generation using Phonon, exclusively on Apple Silicon. Users can now play text or chat model replies aloud directly from the ollama run command or stream audio through an OpenAI-compatible /v1/audio/speech API. Furthermore, v0.40.9 integrates d3 decision models, which are derived from vLLM Semantic Router. These models are available in 2B, 4B, 9B, and 27B sizes and are designed for tasks such as classification, yes/no answers, and probabilistic scoring based on text, structured state, or images. The release also extends support for Flint to Linux and Windows platforms, enhancing cross-platform accessibility.