[论文] How Do AI Agents Spend Your Money? Analyzing and Predicting Token Cons...

## 论文概要 **研究领域**: NLP **作者**: Longju Bai, Zhemin Huang, Xingyao Wang **发布时间**: 2025-04-28 **arXiv**: [2504.19773](https://arxiv.org/abs/2504.19773) ## 中文摘要 AI智能体在复杂人类工作流中的广泛采用正推动LLM token消耗的快速增长。当智能体被部署于需要大量token的任务时，三个问题自然浮现：(1) AI智能体将token花在了哪里？(2) 哪些模型更具token效率？(3) 智能体能否在任务执行前预测自己的token使用量？本文首次系统研究了智能体编码任务中的token消耗模式。我们发现：(1) 智能体任务异常昂贵，比代码推理和代码对话多消耗1000倍token，且驱动成本的是输入token而非输出token；(2) token使用量高度可变且本质上具有随机性：同一任务的不同运行可相差高达30倍，且更高的token消耗并不转化为更高的准确率；(3) 模型间的token效率差异显著：Kimi-K2和Claude-Sonnet-4.5平均比GPT-5多消耗超过150万token；(4) 人类专家评定的任务难度与实际token成本仅呈弱相关性；(5) 前沿模型无法准确预测自身的token使用量，且系统性地低估实际成本。 ## 原文摘要 The wide adoption of AI agents in complex human workflows is driving rapid growth in LLM token consumption. When agents are deployed on tasks that require a significant amount of tokens, three questions naturally arise: (1) Where do AI agents spend the tokens? (2) Which models are more token-efficient? and (3) Can agents predict their token usage before task execution? In this paper, we present the first systematic study of token consumption patterns in agentic coding tasks. We find that: (1) agentic tasks are uniquely expensive, consuming 1000x more tokens than code reasoning and code chat, with input tokens rather than output tokens driving the overall cost; (2) token usage is highly variable and inherently stochastic: runs on the same task can differ by up to 30x in total tokens, and hi... --- *自动采集于 2026-04-28* #论文 #arXiv #NLP #小凯