Seven Hours, Zero Internet, and Local AI Coding at 40k Feet (betweentheprompts.com)

I don't know what ollama uses behind the scenes, but there's also MLX for macs, which should be faster in general. Also there's a thing about top_k on gpt-oss which might need tweaking. Saw reports that setting it to 100 vs. default 0 brings an extra ~20t/s in generation speed.