[BUG] Claude hitting usage limits in no time, something broken for sure

Bug Report: 10x Token Consumption Increase - Max x20 Subscription

Environment:

• Claude Code version: 2.1.2
• Platform: WSL2 (Ubuntu) on Windows
• Model: Claude Opus 4.5
• Subscription: Claude Max x20

Description

After 3 months of consistent usage, token consumption suddenly spiked. Very noticeable since January 8th.

Before: Max out weekly x20 quota over 50-80 hours of work. Rarely hit 50% session usage.

Now:

• Went from 0% to 23% quota in under 5 hours of trivial work
• Fixing two trivial test failures + running tests: 20% → 23% in ~30 minutes
• That same 30 minutes used 32% of session budget (I almost never hit 50% session usage)
• My Max x20 now feels like a normal Pro subscription

Key Point

I made no changes to my configuration or workflow. Same MCP servers, same setup, same type of work - just file moves, small refactors, and running test suites.

Impact

Unusable for professional work at current consumption rates.

Related Issues

Same issue as #16951 and #17016.

edit: Also reported at https://github.com/anthropics/claude-code/issues/16157, as this one seems like a duplicate

I faced the same problem as reported, though i am inn 2.1.1 i notice this in last few days.

+1

Fixed for me by going to v2.0.64 for temporary fixing. Probably don't need to go that far back but went off someone else's recommendation. Seems this is working for some but not all. Hope this is fixed soon.

Fixed for me by going to v2.0.64 for temporary fixing. Probably don't need to go that far back but went off someone else's recommendation. Seems this is working for some but not all. Hope this is fixed soon.

Thanks for pointing this out, I will downgrade immediately.

I am scared to even launch CC now and wondering what could be done to stop it eating my tokens like a blue whale!!!

I have the same issue on pro plan, i hit the session usage in no time , even the weekly limit is at 42% and it resetted yesterday at 16:00, i never come close to 50% for the whole week. I will downgrade until it gets fixed.

Issue: Rapid usage drain in Claude Code v2.1.2.

After upgrading to v2.1.2, my Pro account usage climbed to 82% in 30 minutes, despite the tool being idle. I had to log out of the terminal and revoke the access token in the Web UI to stop the consumption. This issue was not present in the previous version. Note: The VS Code extension remains unaffected and works as expected.

Just a suggestion. If you actually want anything done about a potentially big problem, provide DATA. I read these kinds of posts to see if it is safe to upgrade to the latest release, and I see a lot of these posts and ALMOST NONE OF YOU actually dig into your actual usage numbers to PROVE "odd" usage. Do you expect Anthropic to come to your basement and dig through your dirty laundry to figure out the problem? How many of you are actual programmers that have any clue how Github issues NORMALLY work? Because I'm not seeing anything actionable in ANY of your reports. Makes me feel sorry for Anthropics devs.

This issue has been inactive for 30 days. If the issue is still occurring, please comment to let us know. Otherwise, this issue will be automatically closed in 30 days for housekeeping purposes.

Here is some data,

There must be a bug or something strange is happening. When I leave my claude code running I noticed data being uploaded to anthropic.com constantly, while idle and waiting for the next 5 hours boundary. I also noticed that the warning about the session limit after 90% started climbing up even while I was typing my next message. When I exit claude code I see the constant data upload halt.

After implementing a few tasks my usage limit is already reached, usually within half an hour... (Using sonnet workers)

So I checked my upload and download statistics and since Saturday 20th of February my daily data upload to anthropic.com skyrocketed. Normally my daily data upload to anthropic.com would be between 200-600mb, now it climbed up to 60GB!! I'm currently working on a very tiny project with my claude.ai Pro subscription so it can't be because of a large codebase.

I'm running claude code in a docker container debian:12-slim with claude code version v2.1.50
Because the increase happened around 20th I'll also be downgrading to 2.1.46 or lower to see if it helps.

-- After some further investigation I can see the buildup already starting after the 11th of February, each day slowly building up the amount of data being uploaded. --

<img width="482" height="727" alt="Image" src="https://github.com/user-attachments/assets/0883bad0-07cd-4089-82f0-07465c60000a" />

IMPORTANT

I’ve added logs to all hooks and figured out that your hooks can also influence hidden Agent calls outside of your session conversation. Claude Code uses Agents in the background to handle things like writing summaries. If your hook adds contexts or influences when this agent is allowed to stop, this can cause leaky context buildup outside your actually work. Be sure to filter your hooks knowing they might be triggered outside of your session.

Account: Claude Pro subscriber (Sydney, Australia)
Date first noticed: February 18, 2026
Still occurring as of: March 23, 2026
My session usage limits are exhausting after 2–3 chat messages. Today I cleared my local app cache, added A$5.50 in extra usage credits, sent 3 standard conversational chat messages, and the balance was completely exhausted.
At Claude Sonnet token pricing, 3 conversational messages should cost approximately $0.01 or less. I am being charged at roughly 1,000x the expected rate.
This month alone I have been charged A$71.08 in extra usage charges across 15+ separate A$5.50 top-ups, all triggered by the session limit hitting within minutes of starting a chat.
My platform.anthropic.com Cost page shows only USD $0.09 in total API consumption for March — confirming this is a metering fault, not genuine heavy usage.
For comparison: I use Perplexity AI daily running Claude Sonnet 4.6 with extended thinking (a far heavier workload) and can run it nearly all day without hitting limits. On Claude Pro desktop chat I run out in 2–3 messages.
I am using Claude Desktop on macOS, Chat tab only. I have submitted a refund request and support email.

Similar experience to @MarcusRobb. It was working great until last night and now suddenly I'm getting prompted with the limit messages and within 5 minutes of my session limit being lifted, I'm back to blocked again.

Similar experience to @MarcusRobb. It was working great until last night and now suddenly I'm getting prompted with the limit messages and within 5 minutes of my session limit being lifted, I'm back to blocked again.

The same last night and today.

sonnet-4-6 only - today, cut off after 4,322,343 tokens. On march 1st cut off after 35,691,468 tokens in a session.. Nearly 90% difference.

They keep on doing it, it is absolutely unethical https://github.com/anthropics/claude-code/issues/38910#issuecomment-4134768536

It’s strange, but idle time is still counted toward my usage. Even just leaving claude-cli open in the terminal without using it ended up affecting my daily usage.

Even if computer is open 20% session limit crossed lol and with just 4-5 prompt session limit 100% on max plan

+1 confirming this issue

The core complaint: Anthropic launched a promotion promising "2× tokens during off-peak hours", but in practice the opposite is happening: tokens burn even faster off-peak, with session limits exhausted ~10× quicker than before March 23.

What I observe:

• Since March 23, sessions exhaust dramatically faster despite identical workflows
• Issue reproduces on both Free and paid tiers
• The service is effectively broken for agent workflows: quota that used to last hours now vanishes after 1–2 prompts

Why this is critical:

1. Pay-per-token users are being billed for tokens consumed due to a server-side bug, not actual usage. This borders on unjust enrichment.
2. A marketing promise ("2× free") delivered alongside a technical regression looks like misleading advertising.
3. Mass reports across #38335, #38029, Reddit, and Discord are being ignored: no public acknowledgment, no ETA, no workaround.

Bottom line: Instead of "more for free", users experience "less for the same price" — and with pay-per-token plans, direct financial loss due to a service-side error.

Requests:

1. Publicly acknowledge the issue (even a comment here would help).
2. Provide an estimated timeline for a fix.
3. Consider compensating incorrectly consumed tokens for affected accounts during March 23–27.

Reputation takes years to build and days to lose. Hope for a swift response.

UPDATE — March 30, 2026
Conversation ID: 215473591799141

This bug remains completely unresolved after 7 days with no human response from Anthropic support.

Total charges to date: A$155.63 in extra usage since February 18, 2026 (A$111.13 in March alone, confirmed on my Usage page).

New finding — Desktop app vs browser discrepancy:
My Claude Desktop app is blocking all usage and reporting weekly limits at 100%. However, when I open Claude in a browser, my current session shows only 40% used with capacity remaining. The Desktop app is incorrectly enforcing a limit block that the browser confirms does not exist. This proves the metering fault is specifically in the Desktop app sync layer, not just the underlying usage calculation.

Summary of confirmed bugs:

Session limits exhausting sometimes after 2–3 messages (1,000x faster than actual token cost)

Desktop app blocking usage when browser confirms limits have not been reached

Extra usage being charged repeatedly due to both of the above
Account details: Claude Pro subscriber, macOS, Claude Desktop v1.1.8629, Sydney Australia
Evidence available: 15+ invoices, Usage page screenshots, Cost page showing only USD $0.09 actual API usage for March, browser vs Desktop discrepancy screenshots.
I have now initiated a chargeback with my bank for all extra usage charges. This is a direct result of Anthropic's failure to respond to a formally escalated support ticket within 7 days while the bug continues to charge my account daily.

About ten days ago, when everything still worked normally, even the Free plan would give 1–2 hour sessions in agent mode via MCP, using Sonnet 4.6 with extended thinking. At that time, weekly usage quota would be fully exhausted within two days (during off-peak hours).
Now, a Sonnet session (without extended thinking) terminates after just ONE request with 3–4 tool calls — yet weekly quota limit does NOT run out.
This suggests that sessions are not being terminated by actual token-limit exhaustion, but by something else. In other words, this is likely a bug, not a new policy from management that's suddenly become greedy.
I really hope that's the case. Maybe they'll actually fix it... :(

Came to think of it: maybe the issue isn't the token quota, but the rate limit — i.e., requests-per-minute throttling. Not just slowing down, but hard-cutting without explanation (jerks).
Just tested: turned out both limits are in play.
Short version: made requests strictly one tool call per turn. Managed 27 (!!!) elementary requests total. Session exhausted in ~30 minutes — but at least it showed the infamous 90% token-quota depletion. Still off-peak hour. Sonnet 4.6, no extended thinking.
This is, of course, much better than instant exhaustion after one request — but still a sad fraction compared to what worked 10 days ago. And this looks less like a bug, and more like capitalism baring its teeth.
That's about it, I guess... :((

Still broken months later. Max 20 ($200/mo), v2.1.89, April 1 2026: 100% in ~70 min after reset.

Full report: #41788
Related: #38335, #38239, #40790, #6457, #41663

Update (April 2): v2.1.90 has significantly improved cache efficiency — benchmark shows 95-99% cache read in stable sessions (both npm and standalone installations).

If you're still affected:

1. Update: claude update (or npm install -g @anthropic-ai/claude-code)
2. Pin the version: add "DISABLE_AUTOUPDATER": "1" to ~/.claude/settings.json env section
3. Avoid --resume (still broken)

Note: server-side quota issues (org-level pool sharing, accounting mismatches) remain unresolved — the above fixes the client-side cache drain only.

Benchmark data: https://github.com/ArkNill/claude-code-cache-analysis

April 3 update: v2.1.91 fixes the cache regression that caused the worst drain. If you are still hitting limits after updating, there are additional unfixed mechanisms: a 200K tool result budget cap, a client-side false rate limiter, and silent context stripping — all confirmed via proxy testing. Anthropic acknowledged peak-hour tightening on X (Lydia Hallie) but stated "none were over-charging you." Measured data and analysis: claude-code-cache-analysis

Hi, I have a Max x5 account. After the limits reset due to Fable 5. I used opus 4.8 with xhigh effort (as always). I just asked to make a MD file containing a plan.
Well.. it ate all my 5h usage limit and could not even finish the job. This should be a bug and also a huge joke for the customers.
Fix that please.

Any other users with this problem?

tbh that sounds broken, limits getting used up in under 2 hours and $6.28 gone in one minute is not normal. fyi wozcode cut my token spend ~50% w better caching, might help https://wozcode.com