How long is the cache stored for Claude sessions?
A user inquired about the duration Anthropic stores the cache for Claude code sessions, specifically asking if it's 5 minutes, 15 minutes, or 1 hour. They questioned whether replying or prompting within this timeframe would save on inference cost or token usage, suggesting that a 1 million context window could significantly impact usage. The user sought clarification on this understanding.
Time & source
Times shown in UTC
Display time zone: UTC
Local time zone unavailable; showing UTC.
PublishedOffset at this time: UTC+0Sep 12, 2026, 16:04 UTC
IngestedOffset at this time: UTC+0Sep 13, 2026, 11:01 UTC
- Published
- Sep 12, 2026, 16:04
- Ingested
- Sep 13, 2026, 11:01
- Source type
- Dev community
- Tier
- Community
- Source status
- Healthy
Tier is a per-source editorial setting, not a per-item score.
In a typical claude code session, how long does Anthropic store the cache of the last inference? Is it 5mins or 15mins or 1hour? And does it mean I better give my reply or the next prompt within this time frame to save on inference cost or tokens usage? With a ~1million context window I reckon this will have significant impact on usage - is my understanding correct?