Yes, relatively speaking. But as I mentioned, because the weights are open source, once you download and train it further with, say, a Japanese dataset, it tends to adopt whatever additional training you provided.