I've been consistently unable to get it to tell me anything at all about Deepseek v4 Pro, it keeps insisting it is beyond its cutoff date, but it knows all about R1.

If it's routing to different models on the backend, it's either pinned for the user, or they're all really old.

Won't it be even worse?

We need to re-send the full context when switching model. It's like sending your data to all provider, so at lease some of them will use it for training ...