[BUG] Instant auto-compaction not instant - compaction takes same time as before v2.0.64
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?
Version 2.0.64 changelog states "Made auto-compacting instant", but the compaction step/process still takes the same amount of time as before. Currently on v2.0.65 and the issue persists.
When auto-compaction triggers (or when using /compact manually), the process duration is identical to pre-2.0.64 versions - there's no observable improvement in how quickly compaction completes. The feature advertised as "instant"
has no noticeable performance difference.
What Should Happen?
Per the v2.0.64 changelog: "Made auto-compacting instant"
The compaction process should complete significantly faster (ideally instantly or near-instantly) compared to previous versions where it would take noticeable time to process and summarize the conversation context.
Error Messages/Logs
No error messages - the feature just doesn't behave as advertised.
Compaction completes successfully but takes the same duration as pre-2.0.64 versions.
Steps to Reproduce
- Use Claude Code v2.0.64 or later (tested on v2.0.65)
- Start a conversation and accumulate context until it reaches ~80% or triggers auto-compaction warning
- Either wait for auto-compaction to trigger OR run /compact manually
- Observe that the compaction process takes the same amount of time as it did in previous versions
- Expected: "Instant" compaction as per changelog
- Actual: Same duration as before the "instant" feature was introduced
Claude Model
Opus
Is this a regression?
No, this never worked
Last Working Version
_No response_
Claude Code Version
2.0.65 (Claude Code)
Platform
Anthropic API
Operating System
Windows
Terminal/Shell
PowerShell
Additional Information
- The v2.0.64 changelog explicitly states "Made auto-compacting instant" as a feature
- Issue persists in v2.0.65, suggesting the "instant" optimization is either not active or has a bug
This issue has 8 comments on GitHub. Read the full discussion on GitHub ↗