Anthropic silently changed Claude Code's default prompt cache TTL from 1 hour to 5 minutes in early April, without any public announcement. This change dramatically increases token consumption for developers doing long-context agentic work. One user's analysis of 1,140 sessions showed cache busts jumping from 39 to 199 per day, raising daily costs from $6.28 to $15.54 — a projected $277.80 monthly increase. Anthropic's Jarred Sumner defended the change by saying shorter caches are cheaper for one-shot queries, but this doesn't hold for sustained agentic sessions. Developers can check their own JSONL logs under ~/.claude/projects/ to confirm the change, and can set CLAUDE_CODE_DISABLE_1M_CONTEXT=1 to cap context at 200k tokens as a partial workaround.

5m read timeFrom xda-developers.com
Post cover image
Table of contents
You can find the switch in your session logsEvery cache bust is a full rebuild, and it adds up fastYou can check your own instance

Sort: