OP, I'm very interested in seeing an actual comparison ran through a common tokenizer of tool calls. I think you'll find different results than what you intended for this tool to be. You've mixed up tokens with characters on your screen.