Slowdowns and errors on assistant API at peak hours?


I’m doing some runs (run / threads) on the assistant API, using gpt-4-1106-preview model.
I’m current based in Tokyo and during the daytime, I have a fairly stable experience, but after 8~9pm JST. we encounter several slowdowns (run queued, in_progress status forever) and sometimes outages (server_error api returns).

I note that it happens only after 8pm JST as I’m doing some research sessions with a colleague on this timeslot and we cannot work efficiently compared to during the daytime.

Maybe some server overloads when Europe and US east coast are waking up ?

This is quite critical because we cannot rely on the API 24/7…