[BUG] Max 5x: entire 5-hour usage limit exhausted ~23 minutes after reset during normal Claude Code usage

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?

I experienced what appears to be an abnormal usage spike on my Claude Max 5x subscription while using Claude Code.

My 5-hour usage limit had just reset. When I checked the usage page again and saw that the limit was already at 100%, there were still 4 hours and 37 minutes remaining until the next reset.

This means the entire 5-hour usage allowance was exhausted in approximately 23 minutes.

During roughly the same period, my weekly usage allowance also decreased by approximately 10%.

This happened while Anthropic's temporary promotion increasing weekly usage limits by 50% through August 31 was active.

This usage rate is completely abnormal compared with my previous Claude Code usage. I have been working on the same project with essentially the same agent-based workflow for a while, and I have never experienced anything remotely close to exhausting the entire 5-hour Max 5x limit in about 20 minutes.

What was running

At the time, I had two Claude Code sessions running concurrently on two different computers:

• One Claude Code session was running on Windows and was primarily reviewing / inspecting image files related to my project.
• The second Claude Code session was running on macOS and was working with the existing codebase and creating Git commits.

I do not know which of the two sessions caused the abnormal usage spike, or whether the usage was caused by a combination of both.

Running agents and doing this type of work is not unusual for my normal workflow. I understand that two simultaneous Claude Code sessions can consume usage faster than one session, and that image analysis, tool calls, large contexts and agent activity all consume usage.

However, the difference here was extreme: the entire 5-hour Max 5x allowance disappeared in approximately 23 minutes after the limit reset, while about 10% of the weekly allowance was also consumed.

Expected behavior

Usage should increase according to the actual amount of Claude Code work being performed.

Two concurrent Claude Code sessions may obviously consume more usage than one, but a normal workflow that I have used previously should not suddenly consume the entire Max 5x 5-hour allowance in approximately 23 minutes unless an exceptionally large amount of model activity actually occurred.

Actual behavior

• Plan: Claude Max 5x
• 5-hour usage limit had just reset
• I later checked usage and it was already at 100%
• Time remaining until the next reset: 4 hours 37 minutes
• Therefore the entire 5-hour allowance was consumed in approximately 23 minutes
• Weekly usage allowance also decreased by approximately 10%
• Two Claude Code sessions were active at the same time
• One session was on Windows
• One session was on macOS
• One session was inspecting project images
• One session was working on source code and creating Git commits
• It is not possible for me to determine which session generated the abnormal usage
• This occurred during the temporary +50% weekly usage promotion

Why I believe this may be a bug

The same project and very similar agent-based workflows previously consumed usage at a much lower and expected rate.

There was no intentional change in my workflow that would explain such a dramatic increase.

The fact that the 5-hour allowance went from a fresh reset to completely exhausted in approximately 23 minutes makes me concerned that one of the following may have occurred:

• runaway or unexpectedly repeated model requests
• an agent or subagent entering a retry loop
• repeated processing of the same images or project context
• unexpectedly repeated context ingestion
• duplicate requests
• failed requests being retried many times
• background agent activity continuing unexpectedly
• abnormal cache behavior
• duplicate usage accounting between concurrent sessions
• another Claude Code usage accounting issue

Request

Could you please investigate the server-side usage records for this period?

In particular, it would be useful to determine:

1. How much usage was generated by the Windows Claude Code session.
2. How much usage was generated by the macOS Claude Code session.
3. How many model requests each session made.
4. Whether subagents or background tasks generated unusually large usage.
5. Whether the image-review session repeatedly processed the same images or context.
6. Whether either session entered a retry/request loop.
7. Whether failed or repeated requests were counted multiple times.
8. Whether usage from the two concurrent sessions was accounted for correctly.
9. Whether there was any Claude Code usage/accounting issue that could explain exhausting a fresh Max 5x 5-hour allowance in approximately 23 minutes.
10. Why approximately 10% of the weekly allowance was also consumed during such a short period.

If the recorded amount of usage was legitimate, I would also appreciate clarification on what generated such an unusually large amount of usage in these two sessions, because this is dramatically different from my previous experience using the same project and workflow.

This appears to be a regression from my normal Claude Code usage behavior.

What Should Happen?

Usage should be consumed proportionally to the actual Claude Code workload.

A freshly reset Claude Max 5x 5-hour usage allowance should not be exhausted in approximately 23 minutes during a normal workflow that previously consumed usage at a much lower rate.

If two concurrent Claude Code sessions legitimately generated this amount of usage, the usage accounting should accurately reflect the actual model requests made by each session, without duplicated requests, retry loops, repeated processing, or incorrect double-counting.

The weekly usage counter should also decrease consistently with the actual amount of work performed.

Error Messages/Logs

Steps to Reproduce

1. Start with a freshly reset Claude Max 5x 5-hour usage window.

1. Run two Claude Code sessions concurrently on two different computers:

• one session on Windows, primarily reviewing / inspecting image files related to an existing project;
• one session on macOS, working on the same existing project, modifying code and creating Git commits.

1. Continue the normal agent-based workflow. This is comparable to the workflow I have used previously without abnormal usage consumption.

1. Check the Claude usage page shortly after the 5-hour window reset.

1. Observe that the entire 5-hour usage allowance is already at 100% used while 4 hours and 37 minutes remain until the next reset.

This means the freshly reset 5-hour allowance was exhausted in approximately 23 minutes.

1. During approximately the same period, observe that the weekly usage allowance has also decreased by roughly 10%.

I cannot determine whether the Windows session, the macOS session, or the combination of both caused the abnormal usage spike.

The issue is not reliably reproducible on demand because the same or very similar workflow previously consumed usage normally. This appears to be an unexpected usage spike rather than a deterministic action that always triggers the problem.

Claude Model

Opus

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

1.40609.0

Platform

Other

Operating System

Windows

Terminal/Shell

PowerShell

Additional Information

_No response_