Perplexity has post-trained Deepseek-R1 to remove China censorship.
But here's the thing: It's still an obviously biased model. And it starts with what they named it: r1-1776. This indicates that the model has a very strong lean towards “USA truth" which, at this point, should be questioned just as well as “China truth".
Once again, absolutely none of these models, and none of the platforms hosting them, should be trusted to provide unbiased and factual responses.
They did not release a model small enough that I can run, so I can't poke at it, unfortunately.
