This blog post explores the idea that measuring the cost of large language models (LLMs) in tokens may not be the best approach for calculating return on investment (ROI). The author suggests that a more intelligent LLM might achieve the same outcome with fewer interactions and tokens, prompting a reevaluation of how we perceive costs associated with these models.