算账：256GB M5 Ultra 本地跑 V4.1 Flash 一年电费 $394，API 仅约 $663

George Kiknadze @kiknaio原推发布于 2026.10.24

在 X 上打开原推

Local DeepSeek V4.1 Flash on that 256GB M5 Ultra (rough math): ~35 tok/s decode (full model doesn’t fit) Running it 24/7: ~1.10 billion tokens/year API equivalent (official off-peak output @ $0.60/M): ~$663/year Macbook electricity consumption (250W, ~$0.18/kWh): $394/year. From model running perspective I really don't think it's feasible. I think it's an amazing laptop and you really would enjoy using it and you'll be super productive but purchasing that only for local LLM seems like overspending to me.