PulseAugur
中
实时 07:53:13
English(EN) I Counted 30 Days of My Claude Code Tokens. 98% Were Cache Reads.

Claude 代码令牌使用量:98% 为缓存读取,0.37% 为输出

一位用户发现,在 30 天内,Claude Code 处理的令牌中 98.1% 用于缓存读取,只有 0.37% 归因于模型的实际输出。这一发现是在用户构建了一个用于跟踪令牌使用量的工具后出现的,该工具显示每次请求都会重新发送整个对话历史记录,而这部分内容可以廉价地从 Anthropic 的提示缓存中获取。用户还详细介绍了 Claude Code 如何将会话数据存储在 JSONL 文件中,并提供了一个 Python 脚本来准确计算令牌,避免因回复记录方式而导致的重复计算。 AI

影响 突出了用户在使用大型语言模型令牌管理和成本结构方面可能存在的效率低下问题。

排序理由 用户对产品运营细节生成的分析。

在 Towards AI 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →

Claude 代码令牌使用量:98% 为缓存读取,0.37% 为输出

本文如何被排名

Signal score
4 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Commentary
用户对产品运营细节生成的分析。
Source corroboration
Single-source cluster
Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics
infra, product
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
Breaking (< 6h)
Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

  1. Towards AI TIER_1 English(EN) · Kushal Banda ·

    我计算了Claude代码令牌的30天使用量。98%是缓存读取。

    <h4><em>What Claude Code writes to your disk, how to add it up without counting twice, and where your plan limits live.</em></h4><p>I pay a flat monthly fee for Claude, so for months I never thought about tokens. Then I built a small menu bar app that counts them, pointed it at m…