Anthropic Adjusts Claude Code Cache Settings Amid User Quota Issues
Severity: Low (Score: 36.9)
Sources: Devclass, Theregister
Summary
Anthropic has reduced the time to live (TTL) for the Claude Code prompt cache from one hour to five minutes, claiming this change should not increase costs. However, users have reported that their quotas are depleting faster, with some Pro users experiencing as few as two prompts in five hours. Sean Swanson, a user, highlighted that the five-minute TTL is detrimental for long-session, high-context use cases. The cache settings were altered after a one-hour cache was introduced on February 1 and reverted on March 7. Writing to the five-minute cache incurs a 25% increase in token costs, while the one-hour cache costs 100% more. Anthropic's Jarred Sumner stated that many requests are one-shot calls, making the five-minute cache more economical. The company is also exploring a 400,000-token context window to mitigate costs associated with cache misses. Developers are concerned that cache rebuilding and misses are leading to quota exhaustion. The situation has raised questions about the efficiency of Anthropic's processing time under current quota structures. Key Points: • Anthropic reduced Claude Code cache TTL from one hour to five minutes, impacting user quotas. • Users report faster depletion of quotas, with Pro users getting as few as two prompts in five hours. • Cache settings changes and bugs may contribute to increased costs and performance issues.