In an update on 29 June, Ollama described how Gemma 4 can run faster with support for Apple’s MLX and a method in which a smaller model suggests upcoming text for the main model to check.
You do not need to understand the technology to get the point: the computer’s work can be organised more efficiently. That can make a model more pleasant to use without every improvement requiring new hardware.
Ollama’s speed figures come from particular tests on a particular computer. They do not tell you how fast it will run on your own Mac, and support depends on the model and software version.
For anyone curious about local AI, it is still an encouraging direction. Progress is about both what models can do and how accessible they become in everyday life. A project that felt slow before may be worth another try after a relevant update.