It's just me, or do models feel worse right before a new release?
Maybe it’s just perception, but I’ve noticed this pattern a few times: right before a new model drops, the current ones suddenly feel a bit less sharp.
More mistakes, more refusals, more generic answers, worse context handling.
I’m not saying they’re intentionally doing it, but it does make me wonder whether routing, safety layers, load balancing, or backend changes start affecting the experience before the new release becomes public.
Has anyone else noticed this, or am I just overreading it?