跳到正文
RCreddit.com·
暂不在当前实时榜单

Claude Code ships a per-model table of what each effort level costs

AI 摘要

A hidden table within the Claude Code binary, named effort_cost_index, reveals per-model token-usage estimates for various effort levels. Normalized with 'high' set to 1, the data shows sonnet-5 has a 'max' cost 132% higher than 'high', opus-4-8 is +14%, opus-5 is +6%, and fable-5 is +10%. The source notes that the catalog numbers' derivation and workload are undocumented, suggesting they be treated as vendor estimates.

为什么是这条

This is the first time a per-model token-usage estimate for different effort levels has been extracted directly from the Claude Code binary, unlike public documentation.

时间与来源

时间显示为 UTC

显示时区:UTC

本地时区尚不可用,暂时显示 UTC。

收录当时偏移:UTC+02026年9月18日 17:00 UTC

收录
2026年9月18日 17:00
来源类型
开发者社区

本站未收录正文。

前往源站阅读 →
来源·reddit.com