Optimizing Claude subagents to reduce high token costs

Nearby_Life9503 · reddit · 2026-08-31

A user experimenting with Claude CLI subagents noticed that creating sub-processes incurs high token costs because the system sends context, cache, and other info with every sub-agent creation. The current setup uses Opus 5 high for the main model and Opus 5 low for sub-agents. The user is seeking optimization strategies or advice on whether switching to a single model (like Opus 5 medium) would be more cost-effective.

Original post →

More from coding & agent

coding & agent channel →