📄 中文摘要
在人工智能项目中，普遍存在一个假设：如果大型语言模型（LLM）的成本很高，模型本身一定很贵。然而，实际情况往往并非如此。很多团队在实践中发现，LLM成本的飙升并不是由于模型定价，而是由于架构决策的影响。从实验阶段转向生产阶段时，有几个关键因素开始显得尤为重要：首先是调用模型的频率，频率的增加会迅速累积，导致成本上升。其次，生产环境中的额外调用、冗余的验证过程或代理的多次内部调用都会显著增加开支。因此，理解这些因素对于控制LLM成本至关重要。
是什么导致你的大型语言模型成本飙升？
出处: What’s Actually Making Your LLM Costs Skyrocket?
发布: 2026年2月11日
📄 English Summary
What’s Actually Making Your LLM Costs Skyrocket?
A common assumption in AI projects is that high costs for large language models (LLMs) are primarily due to expensive model pricing. However, this is rarely the case. Many teams discover that the real issue lies in architectural decisions rather than model costs. As projects transition from experimentation to production, three key factors become critical: the frequency of model calls, which can compound quickly; unnecessary validation passes; and agents making multiple internal calls. These elements contribute significantly to unexpected cost increases, highlighting the importance of understanding the underlying drivers of LLM expenses.