Decrypt noted that AI companies charge per token, but most users waste a significant number of tokens on tasks that could be completed with cheaper models. The article provided various strategies, from initial prompting to using AI agents, aimed at helping users reduce costs and improve the effectiveness of AI model usage. For example, Meta employees consumed 60 trillion AI tokens in 30 days earlier this year, with an estimated cost of up to $900 million. High-end models from Anthropic and OpenAI are 100 times more expensive than their low-end counterparts. The article suggested that users select appropriate models based on the task, optimize prompts, start new conversations for new topics, limit output length, and utilize free models.