谷 水 (@emmett) 在 CCometixLine 的token消耗量准确性测试 中发帖
由于发现智谱官网的token计费是实时更新的,故依据官网消耗token数据,对 CCometixLine 的token计算准确性进行测试。
为何测试
在claude code中使用过任何三方模型的用户应该体会过通过 /cost 命令显示的token计数基本上没有用,ccusage 的消耗量也非常不可靠。故我很在意使用statusline显示token使用量是否可靠。
测试过程
三轮简单提问测试。首先在智谱官网记录初始可用的token数,每次提问后计算智谱token的差值、CCometixLine 的 显示用量,进行对比。
测试结果
初始剩余可用token:1,769,132 tokens
index
prompt
ccline token
ccline percent
z.ai glm 4.5 diff
z.ai glm 4.5 left
z.ai 4.5 air dif...