📄 中文摘要
在使用 LLM 功能时，账单可能会在不知情的情况下激增。大多数成本飙升是可预测和可观察的。造成这种情况的原因包括：上下文膨胀，导致输入令牌数量增加；重试风暴，导致一个操作产生多个调用；错误模型漂移，使得昂贵模型成为默认选择；代理/工具循环，导致工具调用失控；冗长输出，造成支付过多费用。为了解决这些问题，可以采取相应的跟踪和限制措施，例如设置提示预算、限制重试次数、调整模型选择、限制工具调用深度和输出长度等。
为什么你的 LLM 账单激增：7 个原因（及解决方案）
出处: Why your LLM bill spiked: 7 causes (and a way to fix them)
发布: 2026年2月17日
📄 English Summary
Why your LLM bill spiked: 7 causes (and a way to fix them)
LLM features can lead to unexpected spikes in billing, often due to predictable and observable factors. Key causes include context bloat, where prompt tokens gradually increase; retry storms, resulting in multiple calls for a single action; wrong model drift, where an expensive model becomes the default; agent/tool loops, causing runaway tool calls; and verbose outputs, leading to excessive charges. Solutions involve tracking input tokens, capping retries, adjusting model routing, limiting tool call depth, and setting maximum response lengths to manage costs effectively.