All guides

The cache and the one-hour clock

Warm turns cost a tenth. Coming back after lunch rewrites everything at twice list price.

3 min

How the cache works for you

Claude Code caches your conversation prefix with a one-hour lifetime; every turn inside that hour refreshes it. Warm turns read the prefix at the cache rate, which is why a healthy session shows a 95%+ cache hit rate.

After an hour idle the entry expires. The next turn uploads and re-caches the whole context at the cache-write rate, which is twice the input price for the one-hour tier. On a 300k-token session that single resume costs more than the previous fifty turns combined.

What to do

Finish the task in one sitting where you can. When you must break for longer than an hour, do not resume the giant session: write the handoff and start fresh.

Do not leave a session idle overnight and pick it up in the morning out of habit. The dashboard's cold-resume finding counts exactly these turns.

See how much this one is costing you: npx readyclaw scan