Analysis of Claude Code session data reveals Anthropic silently changed the prompt cache TTL default from 1 hour to 5 minutes around early March 2026, causing significant cost and quota inflation. The change resulted in a 17.1% increase in cache creation costs across 119,866 API calls spanning January through April 2026.

Data from two independent machines shows consistent 1-hour TTL behavior from February 1 through March 5, followed by a sudden shift to 5-minute TTL that began around March 6-8. This regression caused a 20-32% increase in cache creation costs and triggered quota limit hits for subscription users who had never previously reached their limits. The most expensive month was March, with users overpaying by $719 for Sonnet and $1,198 for Opus.

February's near-zero waste profile (1.1%) suggests 1-hour TTL was the intended default for Claude Code. The reversion to 5-minute TTL appears to be the primary cause of recent quota limit issues reported by subscription users. The data indicates this was likely an infrastructure regression rather than an intentional change, as the 1-hour default had been stable for over a month across different accounts and machines.