You are not wrong, but it depends on the available hardware. I can run llama.cpp with Qwen 35B-A3B with usable speed (30ish t/s tg) on my laptop, but strata doesn't run at all, because my VRAM doesn't suffice.