Voting up here too. For many use cases it would make a real difference knowing the token usage. For some projects I just dropped the default llm model...