📄 中文摘要
使用大型语言模型（LLM）进行API调用时，许多用户可能会发现自己支付的费用比实际需要的高出3-5倍。这并不是因为模型本身昂贵，而是因为在大多数任务中使用了不合适的模型。根据计算，如果60%的任务足够简单，可以使用更便宜的模型，那么用户每百万个token的费用将多出7.92美元。对于每天超过100个请求的用户而言，这种浪费每月可能超过100美元。简单任务包括文件读取、格式化修复、测试样板生成和简单重构等，这些都不需要复杂的推理能力。选择合适的模型可以显著降低成本。
使用单一大型语言模型的隐性成本
出处: The Hidden Cost of Using One LLM for Everything
发布: 2026年3月27日
📄 English Summary
The Hidden Cost of Using One LLM for Everything
Many users of large language models (LLMs) may find themselves paying 3-5 times more than necessary for API calls. This is not due to the models being expensive, but rather because they are using the wrong model for most tasks. Calculations show that if 60% of tasks are simple enough for a cheaper model, users are overpaying by $7.92 per million tokens. For those making over 100 requests a day, this can lead to more than $100 in wasted expenses each month. Simple tasks include file reads, formatting fixes, test boilerplate generation, and straightforward refactoring, all of which do not require complex reasoning. Choosing the right model can significantly reduce costs.