跳到正文
RadarAI· X:Rohan Paul (@rohanpaul_ai)·· 3 小时前AI 评分58

哈佛论文发现编码 Agent 94.8% 的 token 是缓存重读,能耗或被低估 25 倍

AI 导读

哈佛一篇新论文记录 1 名工程师 10 周真实使用 Claude Code 的情况,共约 10 亿 token、42 个会话,发现 94.8% 被处理的 token 是缓存上下文重读,每个生成 token 平均读取约 260 个缓存 token。

正文 · 原文

Rohan Paul· @rohanpaul_ai · X·2026-11-18 03:25· 56 分钟前AI 评分61

AI 导读

作者转发一篇哈佛论文,该论文记录 1 名工程师 10 周真实 Claude Code 使用,约 10 亿 token、42 个会话,发现 94.8% 处理的 token 是缓存上下文重读,每个生成 token 平均读取约 260 个缓存 token。

正文 · 原文

New Harvard paper finds that 94.8% of a coding agent's tokens were re-reads of cached context, i.e. 94.8% of the text the AI processed was earlier conversation it was re-reading, not new work.

And that re-reading raises its estimated energy use about 25 times, from roughly 10 to 250 kilowatt-hours over 10 weeks.

Energy estimates for LLMs usually price single request-and-response calls. Agents re-read a growing context on every call and hand work to subagents, which those estimates can miss.

The authors logged 10 weeks of real Claude Code use by 1 engineer, about 1 billion tokens across 42 sessions. They built an energy estimator that charges cached tokens for attention reads but not full reprocessing.

About 260 cached tokens were read per generated token, and context grew 5.3× over a session. The cache-aware estimate came to about 250 kWh, against about 10 kWh from the output-based EcoLogits estimator.

If you track agent cost or emissions, count cached reads and subagent calls, and start new tasks in fresh sessions when possible.

来源:Rohan Paul · x.com

来源:RadarAI · aihot.news