You realize the models you’re running locally had to be trained the same way as the proprietary ones, right?

I bet I could come up with some contenders… I won’t though, because I hate that M$ controls the major open-source repository.