Ollama says models with architectures supported by the MLX runtime now run on MLX by default on Apple Silicon devices. The change is described in its v0.40.0-rc2 release notes and applies automatically when an eligible model runs.
The key limit is architecture support. An Apple Silicon device alone does not make every model eligible for MLX. According to Ollama, the new default applies to model architectures that the MLX runtime supports.
Ollama names gemma4, qwen3.6 and qwen3.5 among the additional models in the release notes. It does not present the change as covering all models, and says it will continue testing and enabling more.
Read nextChinese AI Models Echo State Doctrine on Sensitive TopicsFor people running eligible models on Apple Silicon, the update changes which runtime Ollama uses automatically. The release notes describe the default behavior, but provide no performance figures or other measured results for the switch.
The MLX default is described in the v0.40.0-rc2 release notes. Ollama says further model support will follow testing, but gives no date for enabling additional models.
