[BUG] Prompt Cache TTL Reduced from ~5 to ~3 Minutes Without Documentation (Mid-December 2025)

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

Prompt caching timeout appears to have been reduced from 5 minutes to ~3 minutes without documentation or announcement.

Previously, the prompt cache TTL was documented as 5 minutes, and in practice worked reliably for ~4 minutes. Recent observations show the cache now expires after approximately 3 minutes, causing:

1. More frequent cache misses - Users must respond faster to maintain cache hits
2. Increased costs - Cache misses mean full context re-processing
3. Degraded UX - Less time to review Claude's output before responding
4. No changelog entry - This behavioral change wasn't announced

What Should Happen?

1. Documented cache TTL should match actual behavior
2. Changes to caching behavior should be announced in release notes
3. Cache duration should remain at 5 minutes (as documented) or provide clear rationale for reduction

Error Messages/Logs

No error messages - this is a silent behavioral change. Observable via:
  - API billing showing fewer cache hits
  - Timing tests confirming ~3 minute expiration vs previous ~4-5 minutes

Steps to Reproduce

1. Start a Claude Code session with a large context (e.g., reading multiple files)
2. Let Claude complete a response
3. Wait exactly 3 minutes before sending next message
4. Observe cache miss (full context re-sent, visible in API usage/billing)
5. Compare to previous behavior where 4-minute gaps still hit cache

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

Worked correctly until approximately mid-December 2025. Exact version unknown as this is a server-side change.

Claude Code Version

2.0.73 (Claude Code)

Platform

Other

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

Business Impact:

• Users who take time to carefully review Claude's output are penalized with cache misses
• Complex coding tasks requiring thought between messages now incur higher costs
• The 5-minute TTL was a reasonable balance; 3 minutes feels rushed

Request:

1. Confirm the current cache TTL
2. Document any changes to caching behavior
3. Consider restoring the 5-minute TTL or providing configuration options

Related:

• Anthropic documentation still references 5-minute cache TTL
• No announcement found in Claude Code changelog