Tell HN: US LLM models are smarter outside US working hours

For sure Claude especially must have a knob that allows them to route traffic to lower quantizations as desired, as well as arbitrary throttling.

It’s clear as day. I’ve also noticed this behavior on Sundays, not just US workday hours.

I'll take this as a notice to wake up early and start token maxing.