[BUG] Hitting much faster usage limits than before
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?
Hi Claude Code Team,
I wanted to report an unexpected change in usage behavior that I noticed between yesterday and today.
My workflow has remained completely the same, but starting today I'm hitting usage limits significantly faster than before. Most notably, this is the very first time I have ever filled the 5-hour session window in Claude Code Max 200$ sub which has never happened to me before despite using it regularly with the same workload.
This strongly suggests something changed on the backend regarding how usage/tokens are being counted or how limits are being applied, rather than anything on my end.
Could you please investigate whether there were any recent changes to usage calculation, rate limiting logic, or session window behavior? Happy to provide more details if helpful.
What Should Happen?
Usage consumption should remain consistent with previous sessions when the workflow and prompts have not changed. The 5-hour session window should not be exhausted under normal, unchanged usage patterns.
Error Messages/Logs
No explicit error messages. Symptom: Usage limits reached significantly faster than usual. First time ever hitting the 5-hour session window limit in Claude Code despite identical workflow.
Steps to Reproduce
Use Claude Code with an unchanged workflow (same types of prompts, same project structure as previous days)
Continue working as normal throughout a session
Observe that usage limits are hit significantly faster than on previous days with identical workloads
Note: No specific code or file is needed to reproduce. The issue manifests across the entire session regardless of task type. This is the first time the 5-hour session window has ever been fully consumed despite the workflow being unchanged.
Claude Model
None
Is this a regression?
Yes, this worked in a previous version
Last Working Version
_No response_
Claude Code Version
2.1.52
Platform
Anthropic API
Operating System
Windows
Terminal/Shell
Cursor
Additional Information
_No response_
43 Comments
Found 3 possible duplicate issues:
This issue will be automatically closed as a duplicate in 3 days.
🤖 Generated with Claude Code
This is a new issue, not a duplicate.
I'm experiencing a similar issue recently, the most dramatic example being today when the 5 hours usage limit filled in under 10 minutes from a handful of turns with limited tools use. This is on a Max plan. Claude Code CLI on Windows 11.
I have no insight into the cause, it's hard to imagine how it's even possible to use up a Max plan 5 hour window that quickly. It's the third or forth time I've hit the limit in the last couple days, in all cases after a negligible amount of work (compared to previous months).
I suspect the cause could be related to the Opus 4.6 model, as its token consumption rate is significantly higher than that of version 4.5. But it is just a speculation from my side.
I'm in exactly the same situation. It started about 12 hours ago. My limits are being consumed much faster than before. I've tried both Opus and Sonnet, and both are significantly oversubscribed. Max subscription, Claude CLI, macOS
They must do here something. Maybe an ULTRA MAX plan for 300$. Then ULTRA MEGA MAX plan for 500$. Sounds like a reasonable solution. (Ha ha)
Opus 4.6 definitely uses more tokens, even a lot more, but this is something else. A full Max 5 hour window in a few minutes with one agent? Make that make sense.
That destroys the point of even subscribing Claude Code. Whole UX gets completely nuked. @claude must do here something, as they will lose their whole loyal userbase.
same issue here , using opus 4.6 at Claude version 2.1.56 without any hit limit issue , after upgraded to version 2.1.59 today , 1 hour already hit the limit.
Sometimes the Claude Code team fixes bugs quickly, this one might be bad enough to qualify, but then so are this one and this one and they've gone unfixed for months.
Deciding which buggy version of the app to revert to is always a challenge. Which one is the least broken? Does anyone happen to know which prior version we should jump back to while we wait for a fix?
this is hilarious. After 1 hour???! This is a good hint. Thanks, will try this version 2.1.56 as well, maybe it does its job.
Hopefully they'll fix it ASAP, as this one single bug destroys my complete workflow.
using max plan , half an hour hit the 75% limit
Seeing this too, it's the input tokens blowing up in particular.
<img width="2560" height="1511" alt="Image" src="https://github.com/user-attachments/assets/99a461b7-db0c-4202-9c95-24c72b2ab6b5" />
the cache read resets to the system prompts/tools portion all the time and the rest of the context gets recached, and as the context grows this keeps repeating. even just a casual chat with nothing happening quickly grows out of proporiton
Seeing this as well. Been using Claude Code everyday this month and tbh I hit the limit couple of times each day. But today something feels super wrong. I've been hitting the limit 10min after the reset, third time just now. Claude Code entered plan mode on 2 different projects and that was it, now I have to wait another 5h... It even used $10 extra usage in 1-2min.
Is the built-in cache broken?
Claude Max
Claude Code v2.1.58
Opus 4.6
In my eyes, they will need to do something here without blowing up everyones pockets.
Saw this:
This is fixed in v2.1.62. Please run claude update. We've also reset rate limits: https://x.com/trq212/status/2027232172810416493
https://github.com/anthropics/claude-code/issues/29178
Thanks for the update! Will test it over the weekend, if it's still the same behavior.
the issue is not related to the TOOL it self, this morning the consuming is very heavy not like every time, im in max plan, and i made only one prompt that takes 30k token and the session hit 23% ?! something wrong
I've been trying with v2.1.62 and I must say it works better but its still way off. I use ccusage to monitor the tokens and as you can see in the image below the tokens used to create cache is 4-6x more than the days earlier. And it seems like Claude isnt reading from the cache as often as before.
Still hitting the limit after ~40min-1h
<img width="2024" height="861" alt="Image" src="https://github.com/user-attachments/assets/c2813a42-90b4-4e85-985f-9118986f88a2" />
Claude Max
Claude Code v2.1.62
Opus 4.6
Windows 11
Same for me. The strange thing on my end: according to Claude Monitor, the limit has not been reached, while Claude Code has been showing the limit as reached for about an hour.
<img width="767" height="477" alt="Image" src="https://github.com/user-attachments/assets/3f82600c-8ecb-42d9-ae9a-20a5cf2866f9" />
<img width="602" height="549" alt="Image" src="https://github.com/user-attachments/assets/855a95c1-838c-4021-aa6b-01182d6a4419" />
Maybe this is the reason for the burning fast for every one, they reduced the token usage per session 🧐
Same issue here since the 4.6 updates. We coded cozempic to get actual visibility into where tokens go — turns out a lot of context is noise (progress ticks, duplicate system-reminders, stale file reads) eating your budget silently.
\
\\\pip install cozempic
cozempic current --diagnose
\
\Shows exact token count + context % + breakdown by message type. The \
treat\command prunes the dead weight so you get more useful work per token before hitting limits. Open source, would appreciate feedback.very useful insights -- what does @claude think about this one?
Anthropic will hit 100B in revenue this year (up 10x) for sure
Regards,
Junaid Ali Qureshi
On Sat, 28 Feb 2026 at 3:58 PM 0zkbrewer @.***> wrote:
It is a very sensitive issue, i've even had additional 25$ spent without even doing anything, something looks wrong and token consumption is now like 10X, and i don't see claude's team reacting to the multiple raised issues. Fix v2.1.62 didn't make any change for me and we need some feedback from claude's team
@claude
Definitely true, thanks to the new tokenconsumption.
I hit my weekly limit (Max 20x plan) in a few hours with 1 Claude Code session, while mostly being away from my computer and not leaving anything running in the background. Something's not right. >10x my daily usage while barely using Claude at all.
I'm on the Pro plan and a _singular_ Opus prompt ate up an entire 5 hour window's worth of usage. I wish I was joking
@claude
this does sound like a bad joke from Anthropic.
Wondering if the new default “High effort” could be the culprit.
<img width="536" height="171" alt="Image" src="https://github.com/user-attachments/assets/c34fc911-79c7-48c1-8064-a9985962b7e0" />
would be too easy.
On the Max plan, yet drained out entirely yesterday. Issue is - my project tasks are the exact same every month and this is the first time I have hit limits on Max. It says I have used 0% of my current session - still I can't start a new chat or use Claude code. To top all of this, Anthropic provides terrible support.
<img width="1597" height="797" alt="Image" src="https://github.com/user-attachments/assets/69b964e7-0c8a-4982-899f-bbdef98c6fbc" />
Same here.
Using the Pro plan I never hit the weekly limit before, hitting in average 20 5-hour limits per week.
Since Monday, I noticed the weekly usage started at more than 30% (I didn't have the view before in the UI) and by my calculations, taking usage measurements periodically, I estimate the actual weekly consumption to be equivalent to 10 5-hour sessions... so for sure I have now less than half the weekly usage I had last week.
I didn't notice a lot of change in the 5-hour slot quotas, because I started using Opus and cannot compare with my previous usage.
Anthropic released the new version 2.1.68.
Brilliant change (<<- please don't believe me here): https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md
I'm experiencing this too.
Being on MAX plan for many months, since maybe a week ago, suddenly I constantly hitting session limit within an hour. My workflow has always been just a single cc instance.
Basically it has become unusable and I could not get any work done pretty much. Something has to be wrong. It felt like after using an old computer for too long, the accumulative debt makes the system slow. I feel like cc is including unnecessary data in the llm context?
current version:
2.1.69as of 2026.03.04.To debug, can someone advise a way to inspect the actual HTTP call cc makes so I can see what's in the context of the requests?
@claude
+1 here. I have a Max subscription plan and all of a sudden hit the 5 hour usage limit in under 30 minute (Opus 4.6 high). When literally yesterday I was never hitting them in the 5 hour window and would have 2-3 hour long sessions of agents just running.
I just updated to the latest claude code (
2.1.81) and then it started happening. I don't recall what the previous version was, I want to say the last time I updated claude code was when the mobile app remote feature was announcedSomething is broken.
Update
When it reset, I literally typed 1 message and it called a python script 2x and it made my hourly usage jump from 0 to 24% that makes absolutely not sense (I am on a Max Plan). Whatever is calculating usage right now is completely busted. I cannot get anything meaningful done.
Terminal output:
<img width="995" height="426" alt="Image" src="https://github.com/user-attachments/assets/8307ba0c-1eab-40eb-98e6-833ff6f46b99" />
<img width="962" height="377" alt="Image" src="https://github.com/user-attachments/assets/6e58dd14-ca43-4eeb-ac43-0ea08c872841" />
You should open a new report for this, they consider this issue resolved so it's unlikely they'll every see your message
Same here. I did 4x, 5x less work on my max x20 plan than 1 week ago. I would eat 10-20% of my 5h rate limit doing some housekeeping (issues/roadmaps/planning) and then start multiple parallel sessions for 2-3h to finally consume my rate limit, even during peak hours.
Today I saw my 5h session (as well as my All models weekly draining much faster) being consumed within 50min without starting any parallel session. Just some chirurgical reconciliations between a roadmap and my repo's issues, and all usage was gone.
It felt like I was using the pro plan I had 1 month ago.
Previous to this, I updated claude-cli : "Successfully updated from 2.1.87 to version 2.1.88"
And I'm using claude in vscode v2.1.88.
I've also hit several time the "you're reached your usage limit", with the API error in the chat, while the usage limit was actually not reached (60% for instance). Retyping "Let's resume" would allow me to continue the work.
This 5h session definitely felt full of bug around usage limit for me.
Yes now its crazy, breaking the limit every 5hours
I'm on pro subscription and claude code has become unusable after forcingly updating to v2.1.119. I didn't want to, but the previous version kept throwing _Interrupted_ at any prompt.
1 simple prompt (edit an html file) consumed the whole 5h usage.
@claude hope this is a involuntary bug that you will solve and not a willingful downgrade move due to capacity issue.
Also, when running double consumption due to peak hour, the client should CLEARLY state that!