[BUG] Weekly usage limit exhausted in ~2.5 days instead of 7 — matches pattern in #52502

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?

We're hitting the same pattern described in this issue, with concrete data over the last 7 days.

Plan: Max 20x ($200/month)
Client: Claude Code CLI, version 2.1.278

7-day aggregate (Sep 16–22, via /stats → Last 7 days):

• Total tokens: 9.6B across 34 sessions
• Model split: Opus 5 = 93.9% of usage (In: 51.6k, Out: 24.0M, Cache read: 8.8B, Cache write: 179.0M), Sonnet 5 = only 5.7%
• Longest session in the window: 5d 3h 25m (a session left running for days, not actively worked on the whole time)
• Weekly limit: 30% consumed within the first day of the billing cycle (Sep 22), projecting full exhaustion in ~3.3 days instead of 7

<img width="701" height="564" alt="Image" src="https://github.com/user-attachments/assets/8c5379aa-e286-4240-a822-843ffd7c262e" />

<img width="641" height="565" alt="Image" src="https://github.com/user-attachments/assets/f2be9013-396c-4b6f-bbb9-5ee7ac1b2aa1" />

Environment:
Claude Code version: 2.1.278

Happy to share full /usage and /stats screenshots if useful for debugging.

What Should Happen?

Single-session example within this window:
One session (4h45m API duration) cost the equivalent of $419+ in API pricing — $395 of it from Opus 5 alone (307k input, 1.2M output, 614M+ cache read). The built-in usage breakdown attributes 93% of that session's usage to subagent-heavy work, with these subagents involved: general-purpose (29%), laravel-architect (9%), phpunit-test-developer (6%), php-laravel-developer (6%), test-validator (1%).

Error Messages/Logs

Steps to Reproduce

Why we don't think this is just normal usage growth:
Our workload and task types haven't changed over the past several weeks, but our weekly limit — which used to comfortably cover 5 working days — now exhausts in ~2.5 days. That's roughly a 65% reduction in effective runway, far larger than the ~17% reduction attributable to the temporary weekly-limit promotion expiring (see support.claude.com's May–September 2026 Claude Code weekly limits promotion history). The magnitude and the subagent-heavy/long-session profile match what's already reported here rather than a general limit adjustment.

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.278

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

IntelliJ IDEA terminal

Additional Information

_No response_