[BUG] Unexpected token usage spike and continued consumption while Claude Code is inactive
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?
Description
For about one week, my Claude Code token consumption has increased dramatically and unexpectedly.
The most concerning part is that the usage appears to continue increasing even when I am no longer actively working with Claude Code. This makes the situation feel completely out of my control as a paying user.
I have tried to investigate this on my side as much as possible:
- checked my local setup
- reviewed my usage patterns
- stopped active work sessions
- verified that I was not intentionally running additional Claude Code tasks
- contacted Anthropic support multiple times
Despite this, I have not received a clear explanation, a real fix, or any convincing response that explains why my consumption keeps increasing.
Why this is serious
From a customer perspective, this feels like an abnormal billing/usage behavior. If tokens or usage credits are being consumed while the user is not actively working, that is a critical issue.
At minimum, Anthropic should provide:
- a clear explanation of what is consuming tokens
- detailed usage logs visible to the user
- a way to stop all background Claude Code activity
- confirmation whether this is a known issue
- reimbursement or credit correction if the usage is incorrect
Expected behavior
Token usage should only increase when Claude Code is actively processing user-requested work.
If background processes, retries, indexing, MCP calls, IDE extensions, or hidden sessions consume tokens, this must be clearly visible and controllable by the user.
Actual behavior
Usage has increased massively over the past week, and it appears to continue increasing even when I am not actively using Claude Code.
Support experience
I have reported this to Anthropic several times, but so far I have not received a meaningful answer or solution.
This is especially frustrating because the difference with ChatGPT/OpenAI usage transparency and reliability has become very noticeable. At this point, it is becoming hard to justify keeping an Anthropic subscription instead of switching to OpenAI.
Request
Please investigate this urgently and provide:
- A detailed explanation of what caused the usage spike.
- A way for users to audit token consumption by session/process/tool.
- A way to stop any background usage immediately.
- Confirmation whether other users are affected.
- A billing/credit review for affected accounts.
What Should Happen?
With my Claude Max subscription at $200/month, I should be able to work comfortably, as I was able to do before.
Until about a week ago, the usage limits felt reasonable and I could use Claude Code for my normal development workflow without major interruptions.
Now, despite using the same subscription and similar work patterns, I am being blocked by limits 3 to 4 times per week. This prevents me from working properly.
Expected behavior:
- Token usage should match my actual active work.
- Usage should not continue increasing when I am no longer using Claude Code.
- A $200/month Claude Max subscription should provide enough capacity for a normal professional development workflow.
- If limits or token accounting changed recently, Anthropic should clearly communicate what changed.
- Users should have clear visibility into what consumes tokens and why they are being rate-limited.
Error Messages/Logs
no msg
Steps to Reproduce
This issue is related to account-level token usage / rate-limit behavior, so I cannot provide a minimal code example that reproduces it independently.
Steps observed on my account:
- Use Claude Code normally with a Claude Max $200/month subscription.
- Work with a similar development workflow as before: same machine, same projects, same type of tasks.
- Stop active Claude Code work sessions.
- Wait and monitor the usage / remaining limit.
- Observe that token usage appears to continue increasing even when I am no longer actively using Claude Code.
- Resume normal work later in the week.
- Hit usage limits much more frequently than before, around 3 to 4 times per week.
Before this issue started, the same subscription and similar usage pattern allowed me to work comfortably. Since about one week ago, I am frequently blocked by limits and can no longer work normally.
Because the behavior depends on account usage accounting and possibly background Claude Code processes, Anthropic may need to inspect server-side usage logs to reproduce and diagnose it.
Claude Model
None
Is this a regression?
Yes, this worked in a previous version
Last Working Version
_No response_
Claude Code Version
2.1.165 (Claude Code)
Platform
Other
Operating System
Windows
Terminal/Shell
Windows Terminal
Additional Information
_No response_
Showing cached comments. Read the full discussion on GitHub ↗
9 Comments
Found 3 possible duplicate issues:
This issue will be automatically closed as a duplicate in 3 days.
🤖 Generated with Claude Code
I do not think this should be automatically closed as a duplicate yet.
The linked issues appear related, but my report adds an important current
regression: on a Claude Max $200/month subscription, my usage was
previously comfortable and workable, but since about one week ago my token
consumption has increased dramatically and I am now blocked by limits 3 to
4 times per week.
The key issue is not only that limits are reached. The issue is that usage
appears to continue increasing even when I am no longer actively working
with Claude Code.
This may be related to the existing reports, but it needs investigation as
an active/current regression affecting paid Claude Max usage. Please keep
this issue open unless Anthropic can confirm the exact root cause, whether
it is the same bug, and what users should do to stop the unexpected
consumption.
Le ven. 5 juin 2026 à 18:43, github-actions[bot] @.***>
a écrit :
I do not think this should be automatically closed as a duplicate yet.
The linked issues appear related, but my report adds an important current regression: on a Claude Max $200/month subscription, my usage was previously comfortable and workable, but since about one week ago my token consumption has increased dramatically and I am now blocked by limits 3 to 4 times per week.
The key issue is not only that limits are reached. The issue is that usage appears to continue increasing even when I am no longer actively working with Claude Code.
This may be related to the existing reports, but it needs investigation as an active/current regression affecting paid Claude Max usage. Please keep this issue open unless Anthropic can confirm the exact root cause, whether it is the same bug, and what users should do to stop the unexpected consumption.
I was fighting this issue for the last two days. What finally made it stop for me was going to Settings > Claude Code > Authorization Tokens and deleting every token in the list (I had 11 pages of them going back months). Obviously that doesn't solve the root of the issue, and I have no idea what caused it in the first place, but hopefully doing this can at least stop it eating your usage like it did for me.
<img width="957" height="716" alt="Image" src="https://github.com/user-attachments/assets/8b253115-bcc4-467f-9eaf-5d737cb5c321" />
The issue has not been fixed on my account.
I am still experiencing abnormally high token consumption, and my usage limits continue to be depleted much faster than they should be. This is happening even when I make simple requests.
As a paying customer, this is extremely frustrating. I am unable to use the service normally because my limits are exhausted far too quickly.
At this point, the situation is seriously making me consider cancelling my subscription immediately unless a solution can be found.
Please investigate my account and explain why my usage is being consumed at such an abnormal rate.
Just happened to me about two hours ago, 15% roughly spike of my weekly usage out of the blue. They just reset it though completely, atleast it looks like it for now.
For anyone experiencing abnormal Claude token consumption, here is what solved the issue for me.
After making the following changes, the abnormal token usage completely disappeared and my consumption returned to normal levels:
I also verified that my APP
.claudeignorefile correctly excludes large directories that Claude should not scan:.claude/.git/.codegraph/.code-review-graph/_UnityProjects/Scripts/AI-LLM/Data/Imports/RAG-Scans/Before these changes, I was experiencing extremely high and abnormal token consumption. Since implementing them, the problem has disappeared completely on my setup.
If you're seeing unusually high token usage, I recommend checking your hooks, background tools, automatic graph/index updates, and making sure your
.claudeignoreis properly configured.The pattern you're describing -- usage climbing while Claude Code is idle -- points to background processes that keep accumulating tokens without a visible task running. A few things worth checking based on what has resolved this for others:
The "usage appears to continue increasing even when not actively working" description matches the hooks + stale-token pattern more than a billing bug, so those are the first two places to check.
(I build Claudiverse -- a session monitor for Claude Code on Mac -- and this failure mode shows up in user reports often enough that I keep a checklist. Disclosure: I am the founder.)
The "usage increasing while inactive" symptom usually points to one of two things: a background agent that was never properly killed (left running in a detached process group after the UI session ended), or a hooks/MCP server that is polling or making tool calls on a timer independent of the visible conversation.
Steps that helped me diagnose this:
claude --list-sessionsor equivalent if your version supports it, to see if there are sessions the UI doesn't show.~/.claude/logs/directory (or wherever your version writes debug output) can show you which session IDs are generating API calls. If you see calls on a session ID that doesn't match an active terminal, that's your culprit.The lack of per-session usage breakdown in the UI is the root UX problem here -- you have a total counter but no way to drill into which session or tool is responsible. Issue #49588 requested an exposed usage API for exactly this case; it was closed as not planned, which is frustrating given how common this scenario is.
If Anthropic support hasn't given you session-level logs, push for them -- they have the API-side data, they just don't surface it by default.