Ollama vs llama.cpp vs MLX on a Mac Studio, measured

Qwen3.5 9B through Ollama (GGUF and MLX), llama-bench, mlx-lm and rapid-mlx on an M3 Ultra. Measured tokens per second, memory, and what MTP changes.

Commenting on Bubbles has been removed. Why? If you want to talk about this post, write to the author, use the comments on their blog if it has them, or reply with a "RE:" post on your own blog. Suggestions? feedback@bubbles.town