Subscriber usage impact of 2.1.117 Opus 4.7 auto-compact threshold change not documented
Summary
Claude Code 2.1.117 changed the auto-compact threshold for Opus 4.7 sessions from ~200K tokens to ~1M tokens. This is listed in the changelog as a bug fix:
Fixed Opus 4.7 sessions showing inflated /context percentages and autocompacting too early — Claude Code was computing against a 200K context window instead of Opus 4.7's native 1M.
However, this change has a significant subscriber impact that does not appear to have been communicated:
Because every turn re-sends the entire conversation context, letting sessions grow 5x larger before compacting results in up to 5x the tokens shipped per turn on Opus 4.7. For Pro/Max/Team subscribers whose quotas were implicitly sized around the prior compact-at-200K behavior, this is effectively a 5x reduction in how many turns a subscription can sustain per 5-hour window.
Observed impact
Within 24 hours of 2.1.118 releasing (inheriting the 2.1.117 change), on an unchanged workflow:
- Exhausted 93% of a 5-hour Max session limit during a ~40-minute research-only session (zero code written)
- Cache-write volume was abnormally high: 207k cache writes / 386k cache reads for a session with only a handful of user turns
- Both a Max 20x individual plan and a Premium Team plan were exhausted on the same day
Prior to 2.1.117, similar daily workloads used a small fraction of the same limits.
Compounding change in same release
2.1.117 also changed:
Default effort for Pro/Max subscribers on Opus 4.6 and Sonnet 4.6 is nowhigh(wasmedium)
Which also increases per-turn token consumption.
Requested
- Acknowledge in release notes when platform changes materially affect subscriber quota consumption
- Consider notifying subscribers proactively when auto-compact thresholds or default effort change
- Confirm that downgrading to 2.1.116 +
DISABLE_AUTOUPDATER=1is a supported configuration until subscriber impact is addressed
Workaround
\\\`bash
Downgrade to last release before the change
(native installer — adjust for your platform)
ln -sf ~/.local/share/claude/versions/2.1.116 ~/.local/bin/claude
export DISABLE_AUTOUPDATER=1 # add to shell rc
\\\`
Version info
- Affected version: 2.1.117 / 2.1.118
- Last unaffected version: 2.1.116
- Platform: macOS (Darwin 24.6.0, arm64), native binary
3 Comments
Closing as duplicate of #52153 — please follow that thread. I've added the 2.1.117 changelog evidence and impact data there.
Well-documented issue. The 200K→1M threshold change is brutal for subscribers. Cozempic prunes the accumulated bloat (stale tool results, old file reads, redundant compaction summaries) so each turn sends far fewer tokens even on the 1M window. Guard daemon handles this automatically.
pipx install cozempic && cozempic guard— https://github.com/Ruya-AI/cozempicThis issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.