Ollama v0.40.0 Brings MLX Runtime to Apple Silicon and Decision Models

8 October 2026 1 min read

#apple #apple-silicon #decision-models #inference-optimization #mlx #multimodal-ai #multimodal-embeddings #ollama #rag #release

Ollama v0.40.0 marks a significant milestone for local LLM deployment on Apple Silicon. The release implements automatic model routing to the MLX runtime for supported architectures, eliminating manual configuration and delivering optimised performance on Apple hardware. This is particularly valuable for developers running models like Qwen3.8, Gemma4, and Qwen3.5 locally on MacBooks and other Apple Silicon devices.

Beyond performance improvements, v0.40.0 introduces decision models as a first-class citizen in Ollama, alongside native support for multimodal embeddings via EmbeddingGemma2. These additions expand the types of workloads practitioners can deploy locally, from traditional generative tasks to reasoning and embedding-based RAG systems. The release represents Ollama's evolution from a simple inference server to a comprehensive local AI platform.

Read the full article on Ollama release.

Source: Ollama release · Relevance: 9/10