I spend my days at work bossing claude opus 5.5 around, and I wanted to see what kind of model I could run on my own laptop that doesn't depend on renting compute from anthropic and the like.
Well it's not great. I can use ollama to run qwen 3.5 (the 4b variant) on my macbook, but it's slow, the context window is tiny, and it drifts into nightmarish hallucinations after a couple of messages back and forth.
I guess there's a reason apple sells those monster mac studios with 512gb of RAM, but I wonder if even that can run a model comparable to the flagships of big labs. Probably not!
Depressing.