Maximizing Claude Code Session Value Through Context and Cache Management

Managing Prompt Cache and Model Switching

Prompt cache efficiency is critical for reducing token costs in Claude Code. Changing the model or effort level mid-session invalidates the cache, forcing a full re-read of the conversation history at full price. The cache expires after one hour on subscriptions or five minutes on API keys. To minimize costs, set your model and effort level at the start of a session or immediately after running /clear. If you need to discard recent turns, use /rewind instead of /compact, as rewinding only removes turns from the end of the context (preserving the cache for prior turns), whereas compacting rewrites the conversation history.