English(EN) We analyzed 1.5M+ of our Claude Code requests (mostly Opus). ~95% of all tokens spent by us are cache hits, which cut our token bill by ~84%. Imagine if they ch

分析发现Claude Opus缓存命中率将token账单削减了84%

作者 PulseAugur 编辑部 · [1 个来源] · 2026-07-08 17:30

SemiAnalysis分析了Anthropic的Claude模型（主要是Opus）的超过150万次请求。他们的发现表明，使用的约95%的token是缓存命中，从而估计将token成本降低了84%。分析还指出，Claude Haiku由于在生命周期较短的会话中使用或作为子代理，其缓存命中率可能较低。 AI

影响强调了缓存机制在大型语言模型中为开发者节省成本的巨大潜力。

排序理由对第三方关于模型性能和成本的使用数据进行分析。

在 X — SemiAnalysis 阅读 →

基础设施

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

报道来源 [1]

X — SemiAnalysis TIER_1 English(EN) · SemiAnalysis_ · 2026-07-08 17:30

We analyzed 1.5M+ of our Claude Code requests (mostly Opus). ~95% of all tokens spent by us are cache hits, which cut our token bill by ~84%. Imagine if they ch

We analyzed 1.5M+ of our Claude Code requests (mostly Opus). ~95% of all tokens spent by us are cache hits, which cut our token bill by ~84%. Imagine if they charged us full price for those too. As a side note, haiku is lower probably because it's used as a subagent or in short h…

报道来源 [1]

We analyzed 1.5M+ of our Claude Code requests (mostly Opus). ~95% of all tokens spent by us are cache hits, which cut our token bill by ~84%. Imagine if they ch

相关实体

相关话题