### Preflight Checklist - [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) 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? Preflight Checklist <li class="tas

[BUG] Max 5x plan: 70M cache read tokens counted against quota — $62/day on a $100/month plan

Open 💬 1 comment Opened Jul 7, 2026 by marionfromfrance

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?

Preflight Checklist
<ul class="contains-task-list">
<li class="task-list-item"><input disabled="" type="checkbox" checked=""> I have searched existing issues and this hasn't been reported yet</li>
<li class="task-list-item"><input disabled="" type="checkbox" checked=""> This is a single bug report</li>
<li class="task-list-item"><input disabled="" type="checkbox" checked=""> I am using the latest version of Claude Code</li>
</ul>
What's Wrong?
Since July 6, my Max 5x session quota drains at 5-10x the normal rate with the same workload I've been running for months. Usage increases even with near-zero interaction — 5% consumed in 2 minutes while not typing anything.
Local usage data (claude-usage dashboard) — July 7, 2026:
<div class="overflow-x-auto w-full px-2 mb-6">
Metric | Value
-- | --
Sessions | 6
Turns | 399
Input tokens (actual) | 72.2K
Output tokens (actual) | 593.8K
Cache reads | 70.23M
Cache creation | 1.94M
Estimated API cost | $62.47 in a single day

</div>
The cache read volume (70.23M) is 100x the actual input tokens. This suggests cached tokens are being counted at full weight against the subscription quota, or prompt caching is not functioning correctly.
Anthropic dashboard at the same time:
<ul class="[li_&]:mb-0 [li_&]:mt-1 [li_&]:gap-1 [&:not(:last-child)_ul]:pb-1 [&:not(:last-child)_ol]:pb-1 list-disc flex flex-col gap-1 pl-8 mb-3">
<li class="font-claude-response-body whitespace-normal break-words pl-2">Session: 30% used (of 5-hour window)</li>
<li class="font-claude-response-body whitespace-normal break-words pl-2">Weekly: 37% used (resets Sunday)</li>
<li class="font-claude-response-body whitespace-normal break-words pl-2">Fable: 0% (never used)</li>
</ul>
Timeline (July 7, Paris time / UTC+2):
<ul class="[li_&]:mb-0 [li_&]:mt-1 [li_&]:gap-1 [&:not(:last-child)_ul]:pb-1 [&:not(:last-child)_ol]:pb-1 list-disc flex flex-col gap-1 pl-8 mb-3">
<li class="font-claude-response-body whitespace-normal break-words pl-2">~7:40 AM: session window opens</li>
<li class="font-claude-response-body whitespace-normal break-words pl-2">9:13 AM: 39% consumed with light usage</li>
<li class="font-claude-response-body whitespace-normal break-words pl-2">9:24 AM: 52% — 11 minutes later, minimal activity</li>
<li class="font-claude-response-body whitespace-normal break-words pl-2">9:26 AM: 57% — 5% more in 2 minutes, near-zero interaction</li>
</ul>
Environment:
<ul class="[li_&]:mb-0 [li_&]:mt-1 [li_&]:gap-1 [&:not(:last-child)_ul]:pb-1 [&:not(:last-child)_ol]:pb-1 list-disc flex flex-col gap-1 pl-8 mb-3">
<li class="font-claude-response-body whitespace-normal break-words pl-2">Plan: Max 5x</li>
<li class="font-claude-response-body whitespace-normal break-words pl-2">Claude Code version: 2.1.201</li>
<li class="font-claude-response-body whitespace-normal break-words pl-2">Model: Opus 4.8 (switched from 1M to 200K context mid-morning — no improvement)</li>
<li class="font-claude-response-body whitespace-normal break-words pl-2">Effort: lowered from High to Medium — no improvement</li>
<li class="font-claude-response-body whitespace-normal break-words pl-2">Fast mode: OFF</li>
<li class="font-claude-response-body whitespace-normal break-words pl-2">OS: macOS</li>
<li class="font-claude-response-body whitespace-normal break-words pl-2">No programmatic usage (no <code class="bg-text-200/5 border border-0.5 border-border-300 text-danger-000 whitespace-pre-wrap rounded-[0.4rem] px-1 py-px text-[0.9rem]">claude -p</code>, no Agent SDK, no GitHub Actions)</li>
<li class="font-claude-response-body whitespace-normal break-words pl-2">No <code class="bg-text-200/5 border border-0.5 border-border-300 text-danger-000 whitespace-pre-wrap rounded-[0.4rem] px-1 py-px text-[0.9rem]">ANTHROPIC_API_KEY</code> in environment</li>
<li class="font-claude-response-body whitespace-normal break-words pl-2">Multiple colleagues on separate accounts experiencing the same issue simultaneously</li>
</ul>
What I've ruled out:
<ul class="[li_&]:mb-0 [li_&]:mt-1 [li_&]:gap-1 [&:not(:last-child)_ul]:pb-1 [&:not(:last-child)_ol]:pb-1 list-disc flex flex-col gap-1 pl-8 mb-3">
<li class="font-claude-response-body whitespace-normal break-words pl-2">Model change (same as usual)</li>
<li class="font-claude-response-body whitespace-normal break-words pl-2">1M context (disabled via <code class="bg-text-200/5 border border-0.5 border-border-300 text-danger-000 whitespace-pre-wrap rounded-[0.4rem] px-1 py-px text-[0.9rem]">CLAUDE_CODE_DISABLE_1M_CONTEXT=1</code>, no improvement)</li>
<li class="font-claude-response-body whitespace-normal break-words pl-2">Effort level (lowered to Medium, no improvement)</li>
<li class="font-claude-response-body whitespace-normal break-words pl-2">Concurrent claude.ai chat (was idle during affected periods)</li>
<li class="font-claude-response-body whitespace-normal break-words pl-2">Background features (no extended thinking, web search, or research mode)</li>
</ul>
Related issues:
<ul class="[li_&]:mb-0 [li_&]:mt-1 [li_&]:gap-1 [&:not(:last-child)_ul]:pb-1 [&:not(:last-child)_ol]:pb-1 list-disc flex flex-col gap-1 pl-8 mb-3">
<li class="font-claude-response-body whitespace-normal break-words pl-2">#41930 — Widespread abnormal usage drain, March 2026 (caching bug, Anthropic reset accounts)</li>
<li class="font-claude-response-body whitespace-normal break-words pl-2">#38335 — Max plan session limits exhausted abnormally fast</li>
<li class="font-claude-response-body whitespace-normal break-words pl-2">#65687 — Unexpected token usage spike and continued consumption while inactive</li>
<li class="font-claude-response-body whitespace-normal break-words pl-2">#69608 — Weekly usage limit draining 10x faster than normal</li>
</ul>
This started on July 6, the same day as the "Elevated errors on several models" incident on status.claude.com. The pattern matches the March 2026 prompt-caching bug that Anthropic acknowledged and fixed with account resets.
Screenshots attached: Anthropic usage dashboard + claude-usage local dashboard showing the 70M cache read discrepancy.

What Should Happen?

Cache read tokens (70.23M) should not be counted at full weight against the subscription quota. Actual billable usage for the day was ~666K tokens (72.2K input + 593.8K output), which should consume a fraction of a Max 5x session window — not 37% of the weekly limit.

Expected behavior:

Prompt caching should reduce quota consumption, not inflate it
Session quota should reflect actual tokens processed, not replayed cache volume
Usage should not increase when Claude Code is idle

Error Messages/Logs

No error messages. The issue is silent — no errors are displayed, 
but the usage percentage climbs abnormally fast in Settings > Usage.

Steps to Reproduce

Open Claude Code v2.1.201 on Max 5x plan
Select Opus 4.8 as model (default)
Start a normal coding session (file reads, edits, tool calls)
After ~20 minutes, check Settings > Usage on claude.ai
Observe 30-50% of 5-hour session already consumed despite light usage
Stop all interaction for 2-5 minutes
Recheck usage — observe it has increased by 3-5% with zero activity
Compare with claude-usage local dashboard: actual tokens are ~100x lower than cache read volume

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.201

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

View original on GitHub ↗

This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗