Ollama's release candidate makes MLX the default runner on Apple Silicon.

v0.40.0-rc0, published September 25, runs every architecture the MLX runner supports on MLX by default on Apple Silicon, with qwen3.8 as the worked example, and the maintainers say more models get tested and enabled during the RC. (release) Mac users get Apple's native ML stack without touching a flag. Anyone with published local-model benchmarks on M-series hardware should re-run them against this build, because the old numbers no longer describe the default path.