According to an engineer's measured report published on Zenn, about 98% of Claude Code's cost comes from the AI 're-reading the entire conversation from the beginning every time'

About this mod

According to an engineer's measured report published on Zenn, about 98% of Claude Code's cost comes from the AI 're-reading the entire conversation from the beginning every time.' The structure is such that the more exchanges there are, the more the cost per interaction balloons, and in a conversation that continued for 13 days, there was apparently even a case where the cost was more than 30 times that of a new conversation.

Just knowing this structure can change how you use it considerably. There were three money-saving techniques that proved effective.

1. Reset conversations frequently The longer you continue a conversation, the more expensive it becomes. Make it a habit to reset the session at a good stopping point.

2. Have large files read by a separate agent Do not load a large number of files into a single session. Split roles and keep sessions small.

3. Have tools called all at once Reducing the number of exchanges directly leads to cost savings. Communicate instructions all together in one go.

Even if you are not using Claude Code, the same idea can be applied to long sessions in ChatGPT or Gemini. Try being conscious of how you break up conversations starting today.