Ollama nearly doubles Gemma 4 speed on Apple Silicon, on by default
Ollama's latest build nearly doubles Gemma 4 token generation on Apple Silicon with multi-token prediction. It's automatic, needs no config, and doesn't change the output.
Ollama shipped a build that makes Gemma 4 generate tokens nearly 90% faster on Apple Silicon, and you do not have to do anything to get it. The speedup comes from multi-token prediction: instead of producing one token at a time, the model drafts several and verifies them together. Ollama auto-tunes how many it drafts as it runs, so the win is on by default, needs no flags, and does not change a single character of the model’s output.
That last part is the one that matters. Speculative tricks that trade quality for tokens-per-second are a dime a dozen. This is the same answer, faster, which is the only kind of speed worth having on a machine you actually work on.
For anyone running models on a Mac, this is the good kind of update. It makes the hardware you already own do more without asking you to learn anything or accept a worse answer. Gemma 4 on a 24GB or 32GB Mac was already the comfortable middle of the local range (the sweet spot Will It Run keeps pointing at). Now it is quicker to talk to.
Update Ollama and run your Gemma 4 workflow again. The number that matters is your own tokens per second, and it just went up for free.