Claude Code Subagents Were 48% of My Bill. Their Output Was 0.9%
I thought my expensive Claude Code habit was long sessions. I was half right. When I finally broke down a month of usage, Claude Code subagents cost 48% of my entire bill , and the tokens they wrote back to me were 0.9% of the total. Almost all of the money went into reading, not writing. Each subagent starts with roughly 51K tokens of context , and it re-reads that context on every single…
Claude Code subagents made up 48% of the total budget, while the output they generated accounted for only 0.9%. The cost primarily consisted of input tokens, not output tokens. Each subagent begins with around 51K tokens of context, and it re-reads this context on every request it makes, which significantly inflates the cost. Fan-out workflows contributed to 48% of the Claude Code expense, while main sessions saw 54% of the costs when request sizes exceeded 400K tokens.
To reduce Claude Code costs, it is suggested to cap the number of agents per run, batch small units into one agent, reduce the starting context size, and hand off to a new session before the context becomes too large.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
