Claude Fable 5 prompt caching supports 1-hour TTL
Developer Rhys Sullivan clarified that Anthropic's Claude Fable 5 API natively supports a 1-hour prompt caching Time-To-Live (TTL) duration, debunking assumptions of a strict 5-minute limit. Although CLI environments like Claude Code optimize for 5-minute intervals, the API refreshes the 1-hour cache with each subsequent request.
Anthropic's 1-hour caching TTL is crucial for real-world agentic workflows where coding loops, testing, and debugging take place over extended intervals. The premium for writing to the 1-hour cache is higher than the 5-minute alternative, but the 90% discount on subsequent reads makes it highly cost-effective for multi-turn developer interactions.
- –**Developer Debugging Cycles:** The 1-hour cache stays warm across typical human review intervals, avoiding costly cold-start fees.
- –**Client Defaults vs. API Power:** Confusion arose because CLI tooling presets prioritize 5-minute intervals, causing developers to overlook native API flexibility.
- –**Agent Loop Efficiency:** Long-horizon agents benefit significantly from the 1-hour TTL, enabling more coherent planning and state tracking without constant cache invalidation.
DISCOVERED
50d ago
2026-06-12
PUBLISHED
50d ago
2026-06-12
RELEVANCE
AUTHOR
RhysSullivan