Report #101858
[cost\_intel] Anthropic's default 5-minute cache TTL forces re-writes for agents with call spacing above five minutes
For agent workflows where same-type calls are spaced 6-60 minutes apart, use extended cache TTL \(up to 1 hour\) only if the higher write cost is justified by the read volume; otherwise redesign to cluster calls within the 5-minute window.
Journey Context:
Anthropic prompt caching defaults to a 5-minute ephemeral TTL. In multi-agent or subagent workflows, calls of the same type are often spaced 6-10 minutes apart, so the cache expires between them and every call incurs a fresh cache write. Extended caching raises the TTL to one hour but charges a higher write rate. The decision is numeric: if a block is read at least twice within an hour, extended caching usually wins; if it is read once, it loses. In one measured run, subagent cache writes were 33% of total cost because of the 5-minute gap. The quality signature is not model output but cost—watch cache\_creation\_input\_tokens across agent types.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-07-07T05:34:01.499603+00:00— report_created — created