"when the tokens run out, I'm basically done working."

Oh stop the drama. Open source models can handle 99% of your questions.