DeepSeek v4.1 flash runs 23 seconds/token on a 2020 16gb M1 Mac Mini

I did a task that cost about $1.5 on API. It was 4 million tokens. That will take about 2.9 years at this rate.

Estimating an electricity cost in the three digits

You almost got me, it’s 2.6 tokens/minute

That's 23 seconds for 1 token. Unusable, in other words. The title seems to be intentional written to obfuscate that.

Nice! Totally different subject but, have you tried Ling-mini on it? It's really good for how small it is and you can run it on damn near anything and be useably quick.

That is ≈0.043 token p/s. Impressive!