Cache TTL silently regressed from 1h to 5m around early March 2026, causing quota and cost inflation
Summary
A GitHub issue reports that Anthropic's Claude-Code cache TTL default regressed from 1 hour to 5 minutes in early March 2026, leading to higher quota usage and costs. The analysis covers 119,866 API calls across two machines, with a phase-by-phase timeline and cost calculations showing a significant increase in cache creation costs when using the 5m TTL. The post urges official confirmation, transparency on TTL behavior, and consideration of restoring 1h TTL or exposing a user-configurable option.