Tip: promptCacheTtl and subagentPromptCacheTtl Let You Keep a 1-Hour Cache on the Main Thread Only

Releasebot·medium signalXBlueskyLinkedInCopy link

Claude Code 2.1.243 adds promptCacheTtl and subagentPromptCacheTtl as separate settings so API-key and cloud-provider users can hold a 1-hour prompt cache on the main conversation while subagents stay at the 5-minute default. That split matters for orchestration patterns where a long-lived parent context is expensive to rebuild but short-lived fan-out subagents would only pay the 1-hour cache-write premium without ever reusing it. The release also adds a modelPricing managed setting so an organization's contracted rates, not list price, feed /cost, the status line, and telemetry.