I want to run the model in Western servers. And GPT/Opus is paid by the company which doesn't really get subsidized tokens.

In the future none of us do, so it's better to trial how the actually adorable models perform.