[BUG] 5x Max non-peak: 10% of 5h window consumed by 2 lightweight questions (6% context), 1% by a single 3-tool-call request

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?

On a 5x Max plan ($200/month), non-peak hours (2026-03-30 05:00 GMT-4), quota consumption is wildly disproportionate to actual
work:

Most damning data point: A single trivial request (3 tool calls: 1 failed gh issue view, 1 ls, 1 file write

• a short text response of ~100 output tokens) consumed 1% of the non-peak 5-hour window. This is not

explainable by any reasonable token accounting model.

In the same session, two earlier questions (a settings review with ~15-20 file reads/bash commands + a web search
with 2 Sonnet sub-agents totaling ~74K tokens) consumed ~10% of the 5-hour window, while context window
utilization was only ~6% (~60K of 1M).

The off-peak 2x promotion ended March 28 — even accounting for that, the consumption is 5-10x what it should be
based on visible token volume.

What Should Happen?

• 3 tool calls with ~500 output tokens should not consume 1% of a 5x Max 5-hour non-peak window
• A lightweight session (~900K total tokens, 6% context) should not consume 10% of the same window
• Quota consumption should be proportionate to actual token usage at published API rates

Error Messages/Logs

Steps to Reproduce

1. Start Claude Code v2.1.87 on 5x Max plan during non-peak hours
2. Send a trivial request that triggers ~3 tool calls (e.g., "write a file to my desktop and tell me where to file an issue")
3. Observe quota meter — expect ~2% consumption for this minimal task
4. Alternatively: send 2 questions involving ~20 tool calls + 2 Sonnet sub-agents, observe ~10% consumption at only

6% context utilization

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.87 (Claude Code)

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

iTerm2

Additional Information

Token breakdown

Sub-agents (Sonnet, self-reported):

• Agent 1: 43,574 tokens, 24 tool calls
• Agent 2: 30,453 tokens, 15 tool calls

Main context (Opus, estimated):

• ~15-20 tool calls, each a full API roundtrip with growing history
• System prompt: ~20K tokens
• Thinking tokens: unknown (invisible to user)

### Related issues

• #38335 (108 upvotes) — Max plan session limits exhausted abnormally fast
• #38345 — 5x plan exhausted after 1-2 prompts
• #39649 — Multiple simultaneous usage anomalies on Max 5x
• #40721 — v2.1.87 rate limit reached immediately on startup

### Context

• 1M context enabled for weeks, was not an issue before March 28
• effortLevel: high (unchanged, was not an issue before)
• Off-peak 2x promotion ended March 28 — explains ~2x increase at most, not the 5-10x observed

---