Comment by chasd00

The token price isn't the only reason to run a model locally though. You can do additional training to specialize or remove censorship that may be a no-no per TOS with cloud GPUs.