On topic of Chinese models, Ernie just released several models (v4.5) ranging from tiny 0.3B to 424B. Several are multimodel, several have <think> modes.
Models are unfortunately either too small for RP (lmao 0.3B) or too big for local machines that aren't inference intent (21B)... no 7B/8B/13B. Also no quants or GGUF yet, but I'm sure they'll be created soon.
While not Dispy, effectively sets the expectation bar for whatever DS would come up with, next.
https://huggingface.co/baidu
https://ernie.baidu.com/blog/posts/ernie4.5