claude-code - 💡(How to fix) Fix [BUG] Single README.md edit burns entire 5-hour Max 5x window in 9m 39s (Opus 4.7 1M, resumed session, v2.1.126, JetBrains, Windows) [4 comments, 5 participants]

Preflight Checklist

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

What's Wrong?

A single small prompt — a minor edit to a README.md file (the only file touched in the session) — on a resumed 1M-context Opus 4.7 session caused Claude Code to work for 9m 39s and consumed the entire 5-hour Max 5x window, locking me out for ~4 hours. No sub-agents, no large repo reads, no parallel sessions.

A previous session of comparable scope (similar README-level edits, same plan, same model) used only a small portion of the 5-hour window — so this is a clear regression, not expected behavior.

This is not the documented peak-hours faster drain — that would still allow multiple prompts. This is one prompt → full lockout, with no pre-turn warning.

Related: #38335, #38345, #45756, #55941. This report covers the 1M context + resumed session + JetBrains plugin on Windows combination specifically.

What Should Happen?

A minor edit to a single README.md file should consume a small fraction of the 5-hour Max 5x window — at most a few percent, even on a resumed 1M-context session with a full prompt-cache miss.

For reference: a previous editing session of comparable scope (similar README-level edits, same plan, same model) used only a small portion of the 5-hour window and let me work through the full session normally. The current behavior is a clear regression from that baseline.

At absolute minimum, if a turn is going to cost more than ~20% of the session window, Claude Code should warn the user before sending the request and offer to start a fresh session or switch off the 1M context window. Silently burning the entire window on one small edit, with no pre-turn cost estimate and no opt-out, is the core bug.

Error Messages/Logs

Worked for 9m 39s
You've hit your limit · resets 12:50am

Steps to Reproduce

1. Have a Max 5x plan with Claude Code v2.1.126 installed in a JetBrains IDE on Windows (native, not WSL).
2. Resume an older session via --resume (or the JetBrains plugin's resume UI) on claude-opus-4-7 with the 1M context window enabled. The session should be old enough that the 1-hour prompt cache has expired (i.e. >1 hour since last activity).
3. Send a single small prompt asking for a minor edit to one README.md file — e.g. "fix typos in README.md" or "rephrase the intro paragraph in README.md".
4. Claude Code works on the turn — in my case the status line showed "Worked for 9m 39s" before the turn ended.
5. Run /status and observe usage at or near 100%.
6. Send any follow-up prompt and observe the 5-hour rate-limit lockout message.

Note: I cannot guarantee 100% reproducibility on a clean account without a similarly-aged resumed 1M session, since the cost appears tied to the cache-miss + full 1M context replay. But the pattern (resumed 1M Opus 4.7 + small edit → session-window exhaustion) is consistent with reports in #45756 and #55941.

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

No response

Claude Code Version

2.1.126

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

PyCharm terminal

Additional Information

No response